Here’s what happens when businesses skip the definition step – and how to avoid it.

When “AI” means everything, it means nothing

The term “AI” currently covers an enormous range of technologies – rule-based automation, recommendation engines, machine learning models, large language models like GPT-5.5, and fully autonomous AI agents. These are fundamentally different things with different costs, capabilities, and failure modes. Yet they all get called “AI” in boardrooms and budget meetings.

Think of it like calling everything with an engine a “vehicle.” Technically correct, but not helpful when you’re trying to decide whether you need a forklift or a company car.

In practice, this vagueness creates three concrete problems for anyone trying to make decisions:

• For buyers/clients, it makes vendor evaluation nearly impossible – you can’t compare solutions if you don’t know what problem you’re solving
• For internal teams, it creates confusion about what’s expected and how success will be measured
• For compliance, it makes it nearly impossible to assess risk, accountability, or regulatory fit

In Switzerland specifically, where compliance expectations are high and SMEs operate under tight regulatory scrutiny, “we’re implementing AI” is not an adequate answer to auditors, data protection officers, or procurement reviewers. Definitional vagueness isn’t academic – it creates real legal and financial exposure.

When a new client says “we need AI,” our team at what. always asks the same set of clarifying questions before anything else:

• What specific task or decision are you trying to improve or automate?
• Where exactly does the current process break down or slow down?
• Does this require judgment and adaptability, or is it a predictable, rule-based task?
• How will you measure whether it’s working?
• Who owns the outcome if it goes wrong?

These aren’t gatekeeping questions. They’re the foundation of any project that actually ships.

Why vague AI goals sink projects before they start

Companies often adopt AI because it feels urgent – competitors are “doing AI,” leadership is asking about it, and nobody wants to be left behind. That’s understandable. But adopting AI because it’s fashionable, rather than because you’ve identified a specific problem it solves, is how budgets disappear with nothing to show for them.

Without a clear definition of what the AI is supposed to do, three things become genuinely impossible:

This connects directly to something we’ve written about before: the “garbage in, garbage out” principle. As we explored in Before You Touch AI, Fix Your Workflows First, AI built on broken or undefined processes doesn’t fix those processes – it just makes the confusion run faster.

What a useful AI definition looks like

Instead of arguing about what AI “is” in the abstract, the more productive question is: what does it need to do for your business, and does it actually require AI to do it?

Before starting any AI project, work through this checklist:

• What specific task should this system perform? (One task, defined precisely)
• Does it need to learn, adapt, or handle ambiguous inputs – or would a simpler rule-based automation do the job just as well?
• How much human oversight is needed, and at what points?
• How will success be measured, and within what timeframe?
• Is this meaningfully different from standard automation?

That last question is the most important one. If the honest answer is “not really” – that’s not a failure. It’s a signal to use the simpler, cheaper, more reliable tool.

This matters especially in the Swiss context. Labour costs here are among the highest in the world. Every poorly-scoped AI project is expensive twice over: once in direct spend, and once in the staff hours lost to a tool that doesn’t work properly. Being precise upfront is not over-engineering – it’s basic financial discipline.

It’s also worth noting that many problems framed as AI problems are actually integration problems. When a client tells us their team is spending hours manually reconciling data between their CRM, ERP, and accounting system, they often assume they need AI to fix it. Usually, they don’t. They need their systems to talk to each other. Our tools integration work resolves the majority of these situations without any AI model involved – and does it faster and more reliably.

What our AI automation expert says

What our AI automation expert says

Most clients don’t have an AI problem – they have a clarity problem. Once we define the actual task, the right solution usually becomes clear, and it’s often simpler than they expected.

Sergei Gordeichuk

Head of Technology at what.

Define first, build second

The companies getting real results from AI aren’t necessarily the ones moving fastest. They’re the ones who were clearest about the problem they were solving before they committed to a solution.

A precise brief – one that defines the task, the success criteria, the data requirements, and the governance expectations – is worth more than any tool. AI automation becomes genuinely powerful once that foundation is in place. Not hype, not a demo that gets shelved after a month. Real leverage, applied to a well-understood problem.

If you’re unsure whether what you’re looking at is an AI problem, an automation problem, or an integration problem, that’s exactly the right question to explore before spending anything. Get in touch with our AI automation team – we’ll help you find the honest answer.

The post Pushing AI Without Definition: Why It Backfires appeared first on what. AG.

Ask AI about your company’s current pricing policy and it might cite a version from two years ago. Ask it for a case study with real ROI numbers and it might generate one that sounds plausible but never actually happened. This is called a hallucination – or drift when the pattern keeps repeating.

Why does it happen? The model was trained on data that existed before your question. It doesn’t have access to what you’ve written internally, updated recently, or stored in your systems – and while some models can query the web to fill gaps, that’s neither reliable nor efficient for company-specific knowledge. So it fills the gaps with the most statistically likely answer, which isn’t always the correct one.

RAG – short for Retrieval-Augmented Generation – is the most practical solution to this problem. Instead of relying on the model’s memory, RAG connects it to your actual knowledge base in real time. Company policies, sales documents, internal manuals, FAQs – when someone asks a question, the system first finds the relevant passages from your content, adds them as context, and only then generates an answer based on that material.

It’s genuinely powerful. Your internal documentation can become something like an operational oracle, answerable in plain language. But it’s not magic, and it’s not the right tool for every situation.

By the end of this article, you’ll know what RAG actually does, how to prepare your content for it, and when it’s the right choice versus when you need something more structured.

How RAG works (without the jargon)

RAG sits invisibly between your question and the AI’s answer. The model doesn’t answer from memory alone; it first retrieves relevant material from your knowledge base and uses that as the foundation for its response.

Here’s the process, stripped down:

1. Content preparation: You load your documents (policies, FAQs, procedures) into the system in a structured, readable format.
2. Indexing: The system processes that content using two complementary approaches: classic keyword search and semantic search (more on this below).
3. Retrieval: When a question comes in, the index pulls the most relevant passages.
4. Generation: The AI produces an answer using those retrieved passages plus its general knowledge.

The result is an answer grounded in your actual content, not generated from scratch.

Why hybrid search matters

Keyword search works well when someone uses the exact term that appears in a document. Semantic search goes further: it understands meaning, not just words.

For example: “how to increase company margin” and “reduce operating expenses” point to the same concept but share almost no keywords. Semantic search finds relevant content even when the phrasing is different. In practice, well-built RAG systems use both approaches together. Keywords catch precise references, product codes, and proper names. Semantic search handles everything where intent matters more than exact wording.

RAG vs. fine-tuning

These two are often confused. RAG doesn’t change the model itself – it uses a ready-made LLM (like Claude or GPT) and passes it the right slices of your knowledge base on each question. Update the knowledge base and re-index it, and answers stay current.

Fine-tuning is different. It permanently adjusts how the model behaves or speaks. That’s useful when you need a consistent style or domain-specific behavior – not when you mainly need accurate, up-to-date answers from documents.

One important thing to keep in mind: RAG is only as good as its retrieval. Wrong context in means a confident but wrong answer out. More retrieved content also means more tokens processed, which adds cost. That balance is something you design deliberately; it’s not a case of “more documents equals better answers.”

At what., we don’t always build RAG from scratch. We choose the approach that fits the actual need – custom RAG, managed search solutions, or a combination – so you’re investing where it genuinely pays off.

Read also: Why do you need AI to automate your processes in the first place?

Preparing your AI knowledge base: why format matters more than you think

RAG retrieves what you wrote. If your documents are messy, fragmented, or poorly structured, the system won’t magically understand your business – it’ll find weak chunks and the model will fill the gaps with confidence. That’s when hallucinations sneak back in.

Retrieval quality depends almost as much on your content as on the algorithm behind it.

Markdown vs. PDF

PDFs are great for reading and sharing. For RAG, they’re often a headache. Complex layouts, broken tables, scanned pages – all of that needs OCR or a parsing step before it can be indexed. That adds cost, processing time, and a real risk of garbled text ending up in your knowledge base. Tools like LlamaIndex are widely used to handle this when PDF is unavoidable, but it’s always more effort than clean structured text.

Markdown works better because structure is explicit: headings, sections, and lists tell the indexing system exactly where one topic ends and another begins. For an AI model trying to retrieve the right chunk, that clarity makes a significant difference.

Markdown is also format-agnostic. It converts cleanly to HTML, Word, PDF, and most CMS exports, so your RAG pipeline isn’t locked to a specific vendor or tool. And it’s been the standard in software documentation for years precisely because it’s plain text, version-control friendly, and easy to maintain.

The practical rule: keep PDFs for archiving and distribution. Use Markdown (or equivalent structured text) as the working format for everything that goes into RAG. If you only have scanned PDFs, budget for extraction – it’s doable, but it costs more and introduces more room for error.

How to structure your documents for retrieval

A few simple habits make a meaningful difference in how well RAG performs:

• One topic per section. Use clear headings. Avoid single massive files that cover everything – prefer themed documents or well-separated sections so retrieval returns coherent blocks rather than half a chapter mixed with unrelated content.

• Descriptive, specific titles. “Introduction” or “Appendix” don’t help search. “Remote work policy – Switzerland” or “Handling price objections – enterprise clients” does. The title is often what gets matched first.

• Put codes and references early. If you use internal procedure codes, module names, or SKUs, include them in the heading or the first line. That makes keyword search hit the right place immediately.

• Use numbered lists for processes. Step-by-step procedures retrieve and cite better than dense prose paragraphs. If there’s a sequence, format it as a sequence.

• Cut the noise. Repeated headers and footers, legal disclaimers on every page, duplicate versions of the same document – all of this pollutes your index. Clean content retrieves cleanly.

A note on chunking

Long documents get split into smaller chunks for indexing. Chunks that are too large bring in too much noise; chunks that are too small lose the thread of meaning. Splitting at Markdown headings naturally keeps related content together and reduces the risk of cutting a concept in half.

Good indexing pipelines also use overlap – a few lines from adjacent sections are included with each chunk so the model doesn’t lose context at the boundary. If a document is short and always relevant in a given flow, sometimes including it in full works better than relying on retrieved fragments alone.

The honest question to ask before buying a more expensive model or platform: is your knowledge base actually findable? A well-designed RAG on clean content will consistently outperform a mediocre setup on chaotic PDFs, at the same API cost.

Not everything belongs in RAG the same way

It’s worth being deliberate about what goes where. Three categories are useful:

A common mistake is putting critical step-by-step procedures into RAG and hoping the model will follow them in order. It won’t reliably. Retrieval finds relevant fragments – it doesn’t replace a workflow engine with enforced sequencing.

Read also: Before implementing AI, it’s worth making sure your underlying workflows are solid first.

When RAG is enough – and when you need more

This is the question that saves teams from overbuilding or underbuilding.

RAG paired with an LLM is the right setup when someone asks a question and needs a grounded answer. It’s not the right setup when the interaction requires a process with mandatory steps, tracked state across sessions, or sequenced decisions that can’t be skipped.

Two mental models:

Quick rule of thumb: one question, one answer, no mandatory sequence across sessions – start with RAG and an LLM. Same user, multiple turns, steps that can’t be skipped – add orchestration. RAG then serves as the supporting library, not the backbone of the process.

Three cases where RAG + LLM is the right call

• Internal FAQ or HR policy. “What’s our remote work policy for employees based in Switzerland?” – A well-indexed corpus, an answer grounded in the actual policy document, no multi-step journey required. Find it, explain it, done.

• Sales enablement. “Do we have a logistics case study with ROI?” – A library of commercial documents that users explore based on intent, not a fixed script. RAG handles this naturally.

• Product support (L1). “How do I reset the connection on device X?” – One question, one answer, tied directly to the manual. If retrieval misses, fix the document, not the whole architecture.

Three cases where you need a stronger architecture

• Digital coaching or consulting with a playbook. Multi-week engagements where you’re tracking goals, working through options, and closing with a plan. The current step and session rules need to live outside the model – in a database or state machine. RAG brings in frameworks and examples when that step calls for them. Without orchestration, the AI skips phases or forgets what was agreed two sessions ago.

• Employee or partner onboarding. Week one: documents. Week two: training. Week three: competency check. That order might be contractual or compliance-driven. Finding the right PDF isn’t sufficient – you can’t open module three until module two is complete. RAG supplies the content; a state machine drives the path.

• Guided sales discovery. Qualification, then needs analysis, then proposal – with mandatory questions at each stage. RAG retrieves pricing, battle cards, and objection handlers. An orchestrator enforces the sequence: “no pricing discussion before the needs are declared.” Without that, AI quotes too early or invents a framework that isn’t yours.

Fix your content before you blame the model

The temptation when RAG underperforms is to upgrade the model or switch to a more expensive platform. Usually, that’s the wrong move.

Most retrieval problems trace back to content quality, not model capability. Documents that are too long, poorly titled, or duplicated across versions will confuse even the best retrieval system. The fix is editorial, not architectural.

Before investing in infrastructure, check three things:

1. Is the content ready? Structured, owned, up to date – not a mix of scattered PDFs and six versions of the same policy document.
2. Is this a search-and-answer problem or a follow-a-path problem? FAQs and policies usually need RAG + LLM. Playbooks and multi-step onboarding need orchestration too.
3. Is success clearly defined? “Useful answers tied to sources” is a success criterion. “It sounds smart” isn’t.

A fast way to find out where the real bottleneck is: pick one domain, curate 20 to 30 documents, write down 10 real questions your team actually asks. Run it. Within a few days you’ll know whether the problem is retrieval, content quality, or architecture – and you’ll have spent almost nothing to find out.

Want AI workflows that are reliable from end to end, not just at the retrieval step? Our tools integration services help connect the systems your RAG pipeline depends on – so data flows cleanly into your knowledge base and stays current without manual effort.

Ready to build a knowledge base that actually works?

The right question isn’t “which AI platform should I buy?” It’s “do I have a knowledge base worth retrieving – and a process that knows when to rely on RAG and when not to?”

That’s exactly the kind of question we help teams work through. As an AI automation agency, what. works with businesses to design RAG setups that fit the actual use case – not more complex than needed, not underpowered for the job. Whether that means a lightweight RAG-only setup or a fully orchestrated AI workflow, we help you figure out the right depth before you build anything.

Get in touch for a focused conversation. No sales pitch – just an honest look at whether RAG is the right fit and what it would take to make it work well.

The post RAG Explained: How to Give AI Your Knowledge Base appeared first on what. AG.

Understanding this shift matters for anyone building in or navigating the Web3 space today.

From monolithic chains to modular infrastructure

Early blockchains were built to do everything in one place. A single network handled transaction execution, reached consensus, ensured data availability, and provided settlement finality. Bitcoin and early Ethereum are the clearest examples: unified systems where every node participates in every function.

The problem with this approach is scalability. When one chain must do everything, it hits limits quickly. Higher demand means higher fees, slower transactions, and pressure to make trade-offs between decentralization and performance.

The modular approach breaks these responsibilities apart. Execution, settlement, consensus, and data availability each become distinct layers that can be optimized independently. A rollup, for instance, handles execution off the main chain and posts transaction data to a separate data availability layer like Celestia or EigenDA, while settling on Ethereum. Each component does one job well instead of one system doing everything adequately.

This architectural shift mirrors what happened in software engineering more broadly – the move from monolithic applications to microservices. The logic is the same: specialization enables scale.

For builders and innovators in the crypto space, this modular foundation has opened up new possibilities. Developers can now launch application-specific chains without building validator infrastructure from scratch, plug into shared security layers, and choose data availability solutions that fit their needs and budget.

The stack is expanding beyond blockchain scaling

Here’s where things get genuinely interesting. The modular blockchain stack started as a solution to a scaling problem. But it’s increasingly becoming something else: infrastructure for autonomous AI agents to communicate, transact, and coordinate.

This might sound like a stretch, but the logic follows naturally. AI agents – software systems that can take actions, make decisions, and interact with external services on a user’s behalf – are becoming more capable and more autonomous. As they start participating in real economic activity, they need reliable infrastructure underneath them. They need to prove who authorized them, make payments, verify the trustworthiness of other agents, and leave auditable records of what they’ve done.

Blockchain’s core properties – transparent records, programmable rules, portable identity, and tamper-resistant logs – map almost perfectly onto what agentic AI requires. The stack isn’t replacing AI; it’s providing the trust and accountability rails that make autonomous agent activity safe and verifiable.

The convergence of agentic AI and decentralized infrastructure isn’t a future scenario – it’s already shaping how we think about Web3 architecture today. The projects that will define the next phase aren’t just building on blockchain; they’re building the coordination layer that lets autonomous systems act with trust, accountability, and economic purpose.

Liwaa Chehayeb

Head of Web3 at what.

Read also: the ongoing tokenization era points to a parallel shift — blockchain is moving from purely speculative infrastructure toward systems that support real economic value. Agentic AI is accelerating that same trajectory.

The protocols shaping the agentic Web3 stack

Several emerging protocols and standards are extending the modular stack specifically to support AI agent activity. Together, they form something that could reasonably be called an agentic commerce infrastructure.

Agent Communication: A2A

A2A (Agent-to-Agent protocol) addresses how AI agents communicate with each other. The future of agentic AI isn’t one all-knowing agent doing everything. It’s a network of specialized agents, each handling a particular domain. 

One agent handles research, another handles payments, another handles compliance checks. A2A provides the common language these agents use to coordinate and delegate tasks between each other.

Tool Access: MCP

MCP (Model Context Protocol) handles tool access. It allows agents to connect with external systems – databases, APIs, blockchain explorers, business workflows, payment services. 

Without MCP, an agent is essentially isolated. With it, an agent can actually interact with the digital world: checking a transaction on-chain, retrieving a document, calling a pricing API, or triggering a business process.

Payment Authorization: AP2

AP2 (Agent Payments Protocol) focuses specifically on authorization. When an AI agent makes a payment on a user’s behalf, a critical question arises: was this actually authorized? AP2 is designed to answer that. 

It’s less about the mechanics of moving money and more about consent, permission scopes, and accountability. Think of it as the authorization layer that sits above the actual payment.

Payment Execution: x402

x402 handles the payment execution side. It revives the old HTTP 402 “Payment Required” status code and turns it into a workable standard for internet-native micropayments. 

An agent can pay for a premium data request, an API call, or even compensate another agent for completing a subtask – all automatically, without human involvement at each step. AP2 proves the agent was authorized; x402 handles the actual transaction flow.

Agent Identity and Reputation: ERC-8004

ERC-8004 tackles agent identity and reputation. As agents increasingly interact with other agents outside their own platform or organization, they need a way to evaluate trustworthiness.

Who built this agent? Has it successfully completed tasks before? Can its claims be verified? ERC-8004 aims to create an open reputation layer for agent-to-agent interactions – essentially helping agents decide who’s worth trusting and paying.

Smart wallets, spending limits, and constrained autonomy

One of the most important design questions in agentic AI is simple: how much financial autonomy should an agent have?

The answer should be: constrained. Not unlimited access to a wallet, but programmable, policy-bound access. Smart wallets and account abstraction make this possible. A user or business can configure an agent to spend only up to a certain amount per day, transact only with approved counterparties, request human approval above a certain threshold, or avoid specific transaction types entirely. Every action gets logged for review.

This matters because the goal isn’t maximum autonomy – it’s useful autonomy within clear guardrails. An AI travel agent that books the cheapest flight within your budget is helpful. An AI agent with unconstrained access to your funds is a liability.

Cross-chain coordination and the UX opportunity

AI agents won’t care which blockchain a service runs on. They’ll care whether the task can be completed quickly, cheaply, and safely. This makes interoperability a first-order concern for the agentic Web3 stack.

An agent might need to pay on one network, verify data on another, and use a service that settles on a third. Cross-chain infrastructure that handles this routing invisibly – without requiring the agent (or the user) to manually manage bridges and gas tokens – is essential for this vision to work in practice.

This connects to one of Web3’s persistent challenges: user experience. Managing wallets, private keys, gas fees, and cross-chain transfers has kept most people at arm’s length from blockchain-based services. AI agents could absorb all of that complexity. Instead of a user interacting directly with decentralized infrastructure, the agent handles it – and the user simply receives results.

That’s a genuinely compelling proposition for broader Web3 adoption. The next major interface for Web3 might not be a wallet or a dApp. It might be an AI agent. Projects like the DMCC crypto and AI ecosystem are already building regulatory and commercial frameworks that anticipate exactly this kind of convergence.

Trust, accountability, and the risks that remain

Autonomous agents create real trust challenges. They act faster than humans can supervise, may interact with unknown services, and can spend real money. Mistakes are possible. So is manipulation – fake or low-quality agents gaming reputation systems, compromised wallets, or insecure API connections.

Blockchain helps address parts of this problem. Transparent execution records, smart contract-enforced rules, verifiable identity, and auditable transaction histories all reduce the risk surface. But blockchain isn’t the brain of an AI agent. It’s the accountability layer around it. The agent decides and acts; the blockchain ensures those actions are recorded, authorized, and traceable.

The honest framing here is that agentic AI and Web3 integration is still early. Standards are fragmented, liability questions are unresolved, and regulatory frameworks haven’t caught up. As with tokenization’s positioning problem, the technology is often ahead of the governance and communication structures needed to make it work in practice. The goal should be constrained autonomy – agents that can act effectively within well-defined rules, budgets, and accountability frameworks, not agents operating without meaningful oversight.

A real-world agentic workflow: what this looks like in practice

Abstract protocol names are one thing. But how does the full agentic Web3 stack actually behave in a real scenario? Here’s a concrete example.

Say a logistics company wants to automate vendor procurement. They configure an AI agent with a clear mandate: source a certified freight partner for a shipment from Dubai to Rotterdam, negotiate within a predefined budget, and complete the booking – all without manual intervention.

Here’s what happens under the hood:

• The user defines the task and sets parameters inside a smart wallet: maximum spend of $8,000, approved vendor categories only, human sign-off required above $10,000. The agent gets to work.

• Using MCP, it connects to freight databases, logistics APIs, and compliance registries to pull live rates, carrier certifications, and regulatory requirements for the route.

• It needs help assessing carbon offset compliance for the EU leg. So via A2A, it delegates that subtask to a specialized compliance agent. That agent checks relevant registries, confirms compliance status, and reports back.

• Before committing to a vendor, the agent checks the carrier’s ERC-8004 reputation profile. The carrier has completed 400+ verified cross-border bookings, holds a strong on-chain track record, and has no dispute history. Trustworthy enough to proceed.

• The carrier’s API responds with an x402 payment request for the booking deposit. The agent checks back with AP2 to verify the payment is within authorized parameters. It is.

• The smart wallet validates the spend against the user’s policy rules and executes the transaction. The booking is confirmed, the transaction is logged on-chain, and a full audit trail is created automatically.

• The user gets a notification. Total time: minutes. Manual steps required: zero.

For businesses exploring how AI automation can power these kinds of workflows, the underlying principles align closely with what our AI automation services team builds for clients today. The difference is that Web3 infrastructure adds the trust, payment, and identity layers that make those workflows verifiable and economically sovereign.

Also worth reading: 2026 Will Be the Year of Autonomous Workflows explores how AI-driven automation is reshaping business operations more broadly.

What to watch as the agentic stack matures

The protocols and standards discussed here are still developing, but several trends are worth tracking closely:

• Adoption of x402 for API-level micropayments and agent-to-agent compensation
• AP2 becoming a standard for verifiable agent payment authorization in enterprise contexts
• ERC-8004 development as a trust and reputation layer for cross-platform agent interactions
• Smart wallet infrastructure maturing to support complex, multi-agent spending policies
• Stablecoin micropayments enabling machine-to-machine transactions at scale
• Agent marketplaces where specialized AI agents offer services to other agents or users

Each of these represents a piece of the broader picture: a Web3 infrastructure that was built to scale blockchains, now evolving to coordinate autonomous digital activity across networks, agents, and economic systems.

The modular blockchain stack isn’t just a better way to build blockchains. It’s becoming the coordination and trust layer for the next generation of the internet – one where AI agents are active economic participants, not just tools that answer questions.

If you’re building in this space or thinking about how agentic AI and decentralized infrastructure intersect with your business, our team at what. works with crypto and Web3 projects at exactly this frontier. Explore our Web3 & Crypto services to see how we can help you navigate and build in this evolving landscape.

The post How Modular Blockchain Powers Agentic AI and Web3 appeared first on what. AG.

How AI writing boosts your visibility

Traditional SEO focuses on keywords and rankings, but today’s algorithms need more. AI writing means crafting content that specifically helps artificial intelligence systems read, classify, and recommend your pages.

By structuring information clearly, adding metadata, and maintaining a logical flow, you make it easier for AI to understand your intent. This approach improves how your content appears across AI-powered search and voice platforms.

How to use this AIO checklist

No matter if you relaunch your website or need to boost your online store, follow these AI optimization and AI writing guidelines to maximise consistency and impact.

• Use Q&A Format
• Add Quotes
• Use Bulleted Lists
• Write Clear Headings
• Use Clear Metadata
• Include TL;DR / Summary boxes or key points
• Use Short Paragraphs
• Avoid Keyword Stuffing
• Link Internally
• Show Author Info
• Implement Schema.org
• Check Crawl Access
• Avoid Blocking LLM Bots
• Submit Sitemaps

AI writing & artificial intelligence optimization – best practices

To maximise visibility, AI writing should follow both linguistic and technical best practices.

Technical recommendations

The following practices ensure that AI systems can efficiently crawl, interpret, and represent your content across search, social, and emerging AI platforms.

Visibility across channels

Omnichannel optimization ensures your content performs across all touchpoints: search, social, video, and voice. The 4P framework – Presence, Performance, Personalization, and Persistence – defines how AI optimization extends your brand reach.

• Presence: Optimise keywords and metadata for discovery
• Performance: Track visibility metrics across platforms
• Personalisation: Adjust content tone and visuals per audience
• Persistence: Refresh and re-optimise content regularly

Google AI Overview

AI overviews present curated, high-authority answers at the top of search results. Being featured here places your brand in front of users during their research and discovery phases. This early exposure builds awareness, trust, and increases the likelihood of conversion later.

Related: Want to understand how AI overviews are changing search behaviour and what zero-click search means for your brand? Read our guide on AI Overview Optimization and Zero-Click Search.

AI content optimization guidelines

To make the most of AI optimization, your content should be designed not just for search engines, but for AI systems that interpret and distribute it across multiple channels. Start by focusing on user intent rather than obsessing over keyword density.

To maximise the effectiveness of AI writing and AI optimization, your content must deliver real value and avoid fluff. Thin, vague, or overly promotional content is often overlooked by AI systems. Focus instead on specificity, clarity, and user outcomes, and prioritise genuine expertise by including:

• Unique insights based on experience or research
• Practical advice users can apply immediately
• Proprietary frameworks or firsthand examples

Structure content around user-intent-driven questions using an FAQ or Q&A format. Headings should mimic natural search phrasing, for example: “How does AIO affect SEO strategy?” Provide clear, concise answers to each question.

Break down complex ideas using structured lists for steps, benefits, features, or comparisons. This helps both readers and AI models scan content efficiently. Use natural language headings rather than abstract or technical terms:

✔ What is the difference between AEO and GEO? 

✘ Comparative AI Optimisation Frameworks

Maintain consistent H2 and H3 formatting for readability. Routinely refresh and reuse content by updating outdated information, adding new answers to reflect evolving search behaviour, and consolidating thin posts into comprehensive evergreen pages. Don’t forget to incorporate short, attributed quotes from trusted sources or experts near key claims to reinforce credibility.

Related: New to AIO and GEO? Get up to speed on what these terms actually mean and why they matter in our article From SEO to AI Search.

Structured data guidelines

Structured data allows AI systems to interpret and display your content accurately in rich results. Proper markup enhances search visibility and click-through rates, which improves how search engines perceive your authority. Use Google’s Rich Results Testing Tool and Schema.org to validate markup. Keep schemas accurate and consistent with page content.

External citations guidelines

External links signal credibility to both users and AI systems. References strengthen trust and contextual relevance.

• Link to authoritative, high-domain sources (e.g., .edu, .gov, leading publications).
• Use contextual anchor text that clarifies relevance.
• Update or remove broken links periodically.
• Combine internal and external linking to balance authority flow.

Wikipedia serves as a key reference point for many LLMs and is used by generative AI search tools such as ChatGPT, Gemini, and Perplexity. Research shows that nearly 10% of all citations used by these AI systems originate from Wikipedia, which makes it among the most commonly referenced sources, alongside platforms such as Reddit and YouTube.

GEO and AI writing – how they work together

AI writing and GEO (Generative Engine Optimization) aren’t separate disciplines – they’re two sides of the same coin. Good AI writing gives AI systems the raw material they need. GEO ensures that material actually gets surfaced in the right places.

When your content is well-structured, semantically clear, and technically accessible, you’re not just helping a search engine crawl your page – you’re giving AI engines like ChatGPT, Perplexity, and Google AI Overviews the signals they need to include your brand in generated answers.

That’s why the practices outlined in this guide directly support a broader GEO strategy. Clean formatting, schema markup, and natural Q&A structure are all things AI engines rely on when deciding whose content to cite.

Future-proof your visibility with what.

AI writing and AI optimization form the future of digital growth. With structured, omnichannel content strategies today, your visibility remains resilient against algorithmic shifts. As a leading GEO and growth hacking digital agency, what. future-proofs your brand by leveraging AI-optimized and GEO strategies that drive visibility, relevance, and measurable results.

Want to know where your brand actually shows up in AI-generated answers – and where it doesn’t? That’s exactly what our GEO service is built for. We track your AI visibility, identify the gaps, and optimize your presence across ChatGPT, Perplexity, Google AI Overviews, and beyond.

Reach out to discover your AI writing and optimization approach: Book your free consultation with what.

The post AI Writing & AI Optimization – Guidelines & Best Practices appeared first on what. AG.

Together with ETH Zurich, we developed the ETH Companion: an AI leadership coach built exclusively on scientific sources that supports over 3000 ETH executives. Here’s an honest look at what we learned along the way.

ETH Zurich – world-class research

ETH Zurich is one of the most renowned research universities in the world. Since its founding in 1854, it’s stood for scientific excellence across the natural sciences, technology, engineering, and mathematics.

With over 13,000 employees from around 120 countries, ETH isn’t just a Swiss institution – it’s a global player in education and research. People who work here operate in a highly complex, international environment, with correspondingly high demands on leadership and decision-making.

That’s exactly where the ETH Companion comes in.

What really makes this AI coach special

The core question at the outset wasn’t a technical one – it was strategic: how do you make high-quality leadership coaching accessible to thousands of executives without pouring huge sums into individual coaching programs? Traditional one-on-one coaching doesn’t scale. Digital coaching does.

The key is the quality of the knowledge base. The ETH Companion draws exclusively from trusted articles by ETH psychologists. No random internet content – only scientifically validated material, developed by ETH for ETH.

And that difference matters more than it might seem. A personal leadership coaching program can easily run several hundred thousand francs a year – if you’re serious about leadership development at scale. The ETH Companion scales exactly that: scientifically grounded leadership coaching, accessible to every executive, at any time.

The goal isn’t to build a chatbot. It’s about creating a tool that supports leaders through real challenges – based on approaches that work because they’re scientifically proven.

That sounds straightforward, but it’s conceptually demanding. You need to clarify very early on: what content goes in? Who curates it? How do you make sure the coach stays within scope?

These questions need answers before you write a single line of code.

Defining the MVP is harder than building the MVP

The most challenging part of the project wasn’t technical – it was defining the MVP (Minimum Viable Product).

What should be included? What comes later? Where’s the line between “good enough to test” and “too raw to show real value”? These discussions take longer than expected, but they’re crucial.

We ran scope definition workshops with the ETH team, created user stories, and built the framework step by step. It was absolutely worth it – but it takes time you need to factor in from the start.

A clear tip: start small, then scale. Focus the MVP on the absolute essentials, see if it works – and only then layer in additional features. Anything you didn’t clarify in the MVP will cost you twice as much to fix later.

User testing reveals what documents never will

The best moment of the entire project came during the user testing interviews.

Watching executives interact with the Companion – and seeing their genuine enthusiasm – was something else. It validated not just the product, but every decision that led up to it.

But user testing also gave us concrete insights we simply didn’t have before:

• How executives phrase their questions – and what that means for the coach’s conversational style
• Where the dialogue felt too generic and needed more depth
• Which entry points felt intuitive – and which didn’t

Without this phase, the product would have been worse. No document, no internal test replaces real interactions.

Security is not an optional add-on

A security audit was part of the development process – and that was absolutely the right call.

For a tool that handles sensitive leadership situations and is used by thousands of executives, security isn’t a nice-to-have. It’s a must-have.

Plan for it from the very beginning – not as a last-minute step just before launch.

What really counts: long-term integration, not short-term usage

Right after launch, the focus naturally shifts to: are users happy with the answers?

That’s a fair question – but it’s too narrow. The actual vision for the ETH Companion is bigger: the AI leadership coach is meant to become a genuine part of everyday working life over the long term, and to sustainably improve the quality of decisions executives make.

And the vision goes even further: what was built today for ETH leaders should eventually be available to large organizations outside ETH – as a scalable solution for companies that take leadership development seriously, without investing hundreds of thousands of francs in individual coaching programs.

That sets a different standard. And it calls for different metrics than “did the AI give a good answer today?”

What we’ve delivered – and what comes next

The result is a functional, stable AI leadership coach with a well-thought-out design that’s already in productive use. But what happens after launch matters just as much as the launch itself.

We’re continuing to develop it in ongoing sprints – because an AI tool like this isn’t a finished project. It’s a living product.

What you can take away from this

If you’re planning a similar project, these are the insights that really matter:

• Differentiation comes from content, not technology – the quality of the scientific knowledge base and the coach’s clear focus make all the difference.
• Start small, then scale – define the MVP scope in writing, validate it, then expand.
• User testing is not optional – treat it as a real project phase, not just a checkbox at the end.
• Security and scalability need to be considered from day one, not as an afterthought.

Curious about what AI automation could look like specifically in your company? That’s exactly what we work on every day.

The post ETH Zurich AI Leadership Coach: Our Lessons Learned appeared first on what. AG.

Here’s how we at what. helped DMCC build one of the most active blockchain and Web3 hubs in the UAE – and what we learned along the way.

DMCC overview: why Dubai’s hub is key for crypto & blockchain

DMCC – the Dubai Multi Commodities Centre – isn’t just another free zone. It’s a government-backed ecosystem specifically designed to attract and support blockchain, crypto, Web3, and AI companies looking to establish themselves in the UAE.

Positioned in Dubai, the city sits within eight hours’ flying time of markets representing 65% of global GDP. That geographic advantage is real – but only if the right founders, developers, and investors know about it.

DMCC’s ambition was clear: become the go-to destination for crypto and AI businesses in the region, generate qualified leads, and build a community that actually delivers value.

That’s where we came in.

Inside the DMCC Crypto Centre: a physical hub for Web3 companies

The DMCC Crypto Centre is more than a co-working space. It’s an entire floor at Uptown Tower dedicated to bringing crypto, blockchain, and Web3 companies under one roof.

Physical proximity matters in this industry. Being able to walk down the hall and meet your next investor or technical co-founder creates a network effect that no Telegram group can replicate.

The Crypto Centre offers companies not just licensing, but access to a thriving ecosystem – advisors, regulators, service providers, and other builders all working in the same space.

That infrastructure gave us something real to market.

DMCC AI Centre: building infrastructure for artificial intelligence companies

Alongside the Crypto Centre, DMCC also operates a dedicated AI Centre – part of the same technology ecosystem floor at Uptown Tower.

The strategy was the same: bring AI startups, researchers, and enterprise teams into one physical location where collaboration happens organically. The AI and crypto verticals share resources, events, and a regulatory environment that supports innovation without unnecessary friction.

For us at what., managing the digital presence of both verticals meant understanding the overlap and reflecting that in our content strategy.

DMCC’s goals: lead generation, positioning, and community growth

DMCC’s objectives were threefold: position themselves as the leading crypto and AI hub in the UAE, generate licensing leads from qualified companies, and build an engaged community around the ecosystem.

The licensing landscape in the UAE is competitive. Free zones across Dubai and other emirates all compete for the same Web3 and AI companies.

To stand out, DMCC needed to demonstrate real community value – the kind that comes from events, shared infrastructure, and access to decision-makers.

Our job was to amplify that value digitally.

How what. managed DMCC’s full digital presence

We at what. took on the full digital presence of DMCC’s technology ecosystem – specifically the crypto and AI verticals.

In practice, our work covered:

• Managing and growing all social media accounts across X, Instagram, and LinkedIn
• Live coverage of physical blockchain and AI events across the UAE
• Running and moderating the DMCC Telegram community server
• Publishing blogs and placing PR to support lead generation and positioning

We weren’t just scheduling posts. We were building infrastructure for a community that needed to feel valuable to its members and credible to the outside world.

Using blockchain and Web3 events as a growth channel

Over the course of our engagement, DMCC hosted 25+ physical events. We attended and covered 24 of them live, creating real-time content across Instagram and other channels.

But live coverage was only part of the strategy. The more important work was building a system around those events – post-event analysis that tracked what actually came out of each one.

Which conversations led to licensing inquiries? What topics generated the most engagement?

Physical events in the crypto and blockchain space do something social media can’t easily replicate: they put thought leaders, investors, and builders in the same room.

We focused on capturing that value and making it visible online.

Our greatest challenges: community building and working within strict constraints

A community only grows when everyone in it gets something out of it. That sounds obvious, but it’s where most ecosystem-building efforts fall apart.

You can’t manufacture belonging. If the people in your ecosystem aren’t getting tangible value – connections, knowledge, opportunities – no amount of content will keep them engaged.

DMCC understood this early. They built physical infrastructure, hosted regular events, and created environments where meaningful connections could form.

Our job was to amplify that signal digitally.

Marketing for a semi-government organization comes with strict brand guidelines, structured approval processes, and a small margin for error.

Early in the engagement, approval workflows created friction. Material was taking too long to get published, which matters in a fast-moving space like crypto and AI.

We worked to improve the process from the inside – mapping bottlenecks, improving communication between teams, and reducing back-and-forth without compromising compliance.

The result was a faster, more efficient pipeline for getting content approved and published – making the rest of the work possible.

Results of what.’s engagement with DMCC

We can point out the following headline results from this project:

• 800+ Web3 and AI companies licensed in Dubai through DMCC
• 24 physical blockchain and crypto events attended and covered with live content
• 10,000+ organic community members grown across social platforms

Growing from 500 to 700 licensed companies within a single year – and continuing beyond 800 – reflects what happens when lead generation, community building, and positioning all reinforce each other.

This wasn’t about one viral post or one successful event. It was about building a system where every piece contributed to the same goal.

Our biggest learning on community building

The biggest lesson from this project: community isn’t a content strategy. It’s a value proposition.

If the people in your ecosystem aren’t getting something tangible out of being there, no amount of posting will manufacture genuine belonging.

DMCC’s physical infrastructure gave us at what. something real to market. The Crypto Centre, the AI Centre, the events, the network of 800+ companies – those are genuine reasons for a founder to choose Dubai and DMCC.

Our job was to make that value visible and compelling to the right audience.

Next steps: building your crypto, blockchain, or AI ecosystem

If you’re building or scaling a crypto, blockchain, or AI project and need a team that understands both the technical landscape and how to grow an audience in it, we at what. would love to talk.

Check out our Web3 and Crypto Marketing services to see how we approach community building, lead generation, and positioning for blockchain and AI companies.

Or book a free consultation to explore what’s possible for your project.

The post How We Grew DMCC’s Crypto & AI Ecosystem to 800+ Companies appeared first on what. AG.

AI overview and zero-click search – definition

A zero-click search occurs when a user receives the information they need directly on the search results page or in an AI overview, without clicking through to any website. This trend has grown rapidly with the rise of featured snippets, knowledge panels, voice assistants, and generative AI summaries.

• Nearly 60% of Google searches in the USA and EU end without any click through to websites.
• In March 2025, over 13% of queries triggered AI overviews – a steep rise from just 6.5% in January.
• Globally, approximately 18% of Google searches now display AI-generated overviews, and these reduce organic click-through rates by roughly 35%.

For marketers, the key implication is that traditional click-through metrics are becoming less reliable. Instead of measuring only visits, businesses must now evaluate how their brand appears in AI responses and search snippets – which is exactly where GEO (Generative Engine Optimization) comes in.

Zero-click search: opportunities for businesses

At first sight, zero-click search may sound like a loss of traffic, but for businesses it also opens up new ways to build brand awareness and authority. Instead of focusing only on clicks, companies can use AI overviews and snippets to place their expertise directly in front of potential customers.

Brand visibility without clicks

Even if users don’t click, your brand can still gain visibility when featured in AI-driven snippets or overviews. This is particularly valuable for establishing thought leadership and brand recall. Appearing directly in an answer box or AI summary signals authority to both users and search engines.

Although traffic may not always follow, the trust and awareness generated can influence purchasing decisions later. 

For example, when a brand consistently shows up in AI-generated health or finance results, users may remember and choose it when they’re ready to engage more deeply. This is one of the core reasons GEO has become such a priority for forward-thinking brands.

First-mover advantage in AIO and GEO

Companies that begin investing in AIO and GEO today have the opportunity to get ahead of slower-moving competitors. With AI overviews and zero-click search still relatively new, there is less competition for inclusion compared to mature SEO practices.

This first-mover advantage means that brands experimenting early with structured data, expert content, and omnichannel presence are more likely to secure consistent spots in AI-generated answers – and hold them.

Your position as a trusted authority

Zero-click search rewards content that demonstrates trust and authority. When AI systems decide which content to highlight, they rely on signals such as credibility, accuracy, and consistency across platforms. Brands that produce well-researched, expert-driven content are more likely to be showcased as definitive sources.

Over time, being repeatedly cited in AI overviews reinforces your reputation as a go-to source. For industries like healthcare, finance, and education, authority and trust are particularly important. Establishing this role early in AI search – through solid GEO practices – ensures long-lasting competitive advantage.

How to optimise for zero-click search and GEO

Adapting to zero-click search requires a shift in strategy. Instead of relying solely on clicks, businesses need to structure their content so that AI systems and search engines can recognise, extract, and highlight it. The following approaches show how to stay visible in an AI-first search landscape.

Expert-driven content

To be chosen for AI overviews or zero-click snippets, content must reflect expertise and precision. Search systems and AI models prioritise authoritative sources that provide clear, accurate, and well-structured information. Businesses should rely on subject-matter experts, cite credible sources, and write in a style that answers questions directly.

Structured formats such as FAQs, lists, and step-by-step guides increase the chances of being selected. Combining E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness) with clarity and simplicity is key. The goal is not just to rank – it’s to provide information in a format AI can easily extract and highlight, which is the foundation of effective GEO.

Omnichannel strategies

Zero-click search isn’t confined to Google. AI systems collect information from across the digital ecosystem, including Wikipedia, blogs, YouTube, Reddit, Quora, LinkedIn, and news sites. To optimise effectively, brands must maintain a strong presence across multiple platforms.

By doing so, businesses increase the likelihood of being cited in AI-generated responses, regardless of the channel. Omnichannel strategies also future-proof visibility and ensure that whether a user searches on Google, asks a smart speaker, or queries an AI assistant, the brand’s content is discoverable and relevant.

Identifying zero-click queries

Spotting queries with high zero-click potential is key to deciding whether a keyword is worth pursuing and how to shape your GEO strategy. The main signal to look at is keyword intent. Informational queries are most likely to trigger AI overviews, featured snippets, or other SERP features, while product comparisons with commercial intent also frequently appear.

Once you’ve identified a relevant query with specialised tools, the next step is to check the current SERP landscape. Does the keyword trigger features such as snippets, knowledge panels, or AI overviews? If yes, you need to decide whether to invest in competing for that position or redirect efforts towards other queries.

Zero-click search: challenges and limitations

From declining website traffic to limited control over content presentation, zero-click search raises real concerns about visibility, attribution, and performance measurement.

Possible decline of organic traffic

One major drawback of zero-click search is the potential loss of organic traffic. When users receive answers directly in AI summaries or snippets, they may never visit the original website. This trend challenges traditional performance indicators and revenue models dependent on clicks.

To counterbalance this, brands must increasingly focus on awareness, authority, and long-term trust. While direct traffic may decrease, the visibility gained through zero-click presence – and through consistent GEO efforts – can still support conversions indirectly.

Limited control over how content is presented

Another challenge lies in the lack of control businesses have over how their content appears in AI overviews or snippets. Search engines and AI models reformat, shorten, or even paraphrase information. This means that the way your brand is represented may not always align with your messaging or tone, and in some cases, citations may not even be attributed properly.

To address this, companies must prioritise clarity, accuracy, and consistency. Structured data, schema markup, and proactive GEO strategies increase the chances of being represented accurately in AI search results.

Lack of KPIs

Measuring success in zero-click environments is becoming increasingly difficult. Traditional KPIs such as clicks, impressions, or bounce rates provide limited insight when users no longer visit websites. This makes it challenging for marketers to prove ROI from GEO and AI search optimisation.

New metrics are emerging – brand mentions in AI outputs, inclusion frequency in AI overviews, and sentiment analysis of AI-generated references. Businesses need to adapt their analytics frameworks to capture these signals. Although imperfect, these alternative KPIs give a more accurate picture of visibility and authority in a search landscape where traffic is no longer the only measure.

Will zero-click search replace SEO?

Zero-click search will not replace SEO, but it is fundamentally changing how businesses approach digital visibility. Traditional SEO remains crucial for driving traffic, building backlinks, and ranking for targeted keywords.

At the same time, AI and Generative Engine Optimization ensures that content is discoverable and usable by AI systems. Businesses need to focus not only on ranking but also on how their content is presented and interpreted by AI – which means SEO and GEO need to work together, not in isolation.

Related: Also read our blog post on AIO and GEO: How to Win at AI Search

Elevate Your Digital Presence with GEO – what.

To stay competitive, businesses must adopt a hybrid strategy. SEO continues to attract clicks and engagement, while GEO (Generative Engine Optimization) positions brands as trusted sources across ChatGPT, Perplexity, Google AI Overviews, and beyond.

From website relaunch to full GEO strategy – at what., we future-proof your digital presence with SEO and Generative Engine Optimization methods that enhance visibility, authority, and engagement across every platform that AI touches. We track where you appear in AI-generated answers, where you don’t, and optimise your presence across every AI engine that matters.
Ready to get found in AI search – not just Google? Book your free consultation with what.

The post AI Overview Optimization and Zero-Click Search – Opportunities & Challenges appeared first on what. AG.

What is AIO (Artificial Intelligence Optimization)?

AI Optimization is the practice of making content easily discoverable and usable by AI search systems. Unlike Search Engine Optimization (SEO), which focuses on ranking in search engine results, AIO targets answer and generative engines to pull your content into responses and overviews. AIO can further be divided into AEO (Answer Engine Optimization) and GEO (Generative Engine Optimization) – two distinct but complementary approaches to AI visibility.

What is AEO (Answer Engine Optimization)?

AEO optimizes your content so that AI-powered systems like Google’s AI Overview, voice assistants (e.g. Siri, Alexa), and featured snippets can extract and display direct answers from your website or online store.

AEO aims to position your content as the definitive answer to a query. The answer is pulled directly from your content, word-for-word in many cases, and shown in search summaries, voice results, and knowledge panels.

The key challenge is zero-click results, as the user may get their answer without ever landing on your site. That’s why AEO is best used in tandem with strong branding, internal linking, and strategic positioning to maintain visibility and authority even when traffic doesn’t follow.

To succeed with AEO, your content needs to be:

• Clear and structured, often in Q&A or bullet format
• Fact-based, with precise, unambiguous information
• Marked up with structured data (like Schema.org) to help AI understand context
• Written in a natural, conversational tone to match how people ask questions

What is GEO (Generative Engine Optimization)?

GEO (Generative Engine Optimization) is the practice of making your content usable by AI systems like ChatGPT, Google Gemini, and Perplexity for generating accurate responses.

Unlike AEO, where the content is pulled directly from your site, GEO deals with how AI models synthesize and rephrase your information. Your content becomes part of a broader knowledge base that AI taps into to answer user queries – mostly without quoting you directly or linking back.

To optimize for GEO, your content should be:

• Clear, accurate, and well-structured, so AI can easily understand and process it
• Rich in topical depth to position your site as a trusted source
• Aligned with E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness) to signal credibility to AI models
• Conversational and natural in tone to mirror how users phrase questions in chat interfaces

Opportunities and challenges of AI search

AI-driven search opens up exciting new opportunities for businesses, but it also comes with significant challenges. On the positive side, brands can greatly expand their reach through AI assistants, build greater authority by being cited in answers, and explore new ways to capture user attention as content is placed directly within AI-generated results.

At the same time, the inner workings of these systems often lack transparency. Selection criteria remain unclear, which makes a targeted optimization more difficult. Traditional KPIs also lose relevance, as click-through rates and impressions become less meaningful. In addition, companies face a growing dependence on third-party AI models, whose algorithms and data sources they cannot control.

To remain competitive, businesses will need to rethink their strategies and metrics and adapt to the changing search landscape.

Will AI search replace traditional SEO?

AI search is changing how people retrieve information, but it will not replace traditional SEO anytime soon. Around two-thirds of Gen Z and “Zillennials” already use AI tools like ChatGPT for both personal and professional tasks, while 34% of Gen Z use AI chatbots as their primary search method, a much higher percentage than older demographics.

Looking ahead, forecasts suggest that AI-driven search could account for 40-50% of all online queries by 2030, with some predicting that half of informational searches could already shift to AI by 2028.

However, SEO continues to play a critical role: traditional search engines still drive over 90% of global website traffic, and Google’s AI Overviews, while influential, appear in only about 13% of queries as of early 2025.

This means that AI search will complement rather than replace SEO in the near future. Businesses must therefore prepare for a hybrid landscape where visibility in both AI-generated answers and traditional search rankings determines success.

How AI will change traditional SEO

AI search won’t eliminate SEO – it will reshape it. Businesses need to adjust strategies and metrics to stay visible in AI-driven results.

AI disrupts SEO metrics

Traditional SEO metrics like click-through rates, keyword rankings, and impressions are becoming less dependable as AI-generated overviews begin to dominate the search experience. Users are increasingly getting the answers they need directly from AI without clicking through to websites, which leads to a sharp rise in zero-click searches.

This shift demands a new approach to measurement. Businesses can no longer rely solely on traffic-based metrics. Instead, attention must shift to alternative indicators such as how often a brand is mentioned in AI-generated outputs, whether it’s visible within AI overviews, and how frequently its content is referenced or cited by AI systems. These new KPIs will offer a clearer picture of how content is performing in an AI-first search landscape.

Increased omnichannel focus

AI search isn’t limited to Google. AI systems pull data from across the digital ecosystem, from social media and forums to YouTube, blogs, and AI assistants. This evolution means that visibility can’t just revolve around ranking on a single platform. Therefore, businesses need to adopt an omnichannel strategy that ensures their brand is discoverable wherever AI looks for information.

Maintaining consistent messaging and authoritative content across all relevant platforms will be key. Whether a user is asking a question on ChatGPT or using voice search on a smart speaker, the more widespread and consistent your presence, the more likely AI systems are to include your content in generated responses.

AI-optimized content writing

As AI becomes a primary interpreter and distributor of content, writing must evolve to meet new standards. It’s no longer just about targeting keywords or matching search intent, it’s about creating content that AI systems can easily understand.

That means writing in a way that’s both machine-readable and human-friendly. Semantic variety, such as using different but related terms and structures, helps AI better understand the scope and depth of your content. The goal is to create content that teaches AI systems how to “think” about your subject, while still engaging the human reader.

AI-friendly formatting

Generative AI systems rely heavily on how content is organized to determine what information is relevant and how to use it. Well-formatted content improves your chances of being referenced or cited in AI-generated outputs, whether in search overviews or chatbot responses.

Clear headings, short paragraphs, and logical flow help AI navigate your content more effectively. Incorporating FAQ-style sections, summary boxes, and schema markup further enhances readability. These elements make it easier for AI to identify your content as authoritative, extract key points, and integrate them into responses, even if your site isn’t the top result on a traditional SERP.

Google Overview gains momentum

Google’s AI Overview is already transforming the user journey. AI-generated summaries appear at the top of many search results and often replace traditional links. For businesses, this represents both a challenge and an opportunity. If your content is not optimized for inclusion in AI overviews, it risks becoming invisible, even if it ranks well organically.

To be included, your content must anticipate user questions and deliver relevant, contextual answers in a concise format. Brands that focus on clarity, authority, and structure will be better positioned to appear in these high-visibility spots and to influence the answers AI systems deliver to millions of users.

5 key strategies for AI search optimization

The following strategies help you future-proof your visibility through effective GEO and AIO:

1. Ensure technical readiness: Speed and clean structures are critical since AI systems often have strict timeouts and limited rendering capabilities. Therefore, content should be written in plain, structured formats, with clear metadata, schema.org markup, and accessible sitemaps. Avoid blocking LLM crawlers through robots.txt or CDN restrictions, as this can prevent your content from being indexed by generative engines.
2. Maximize visibility across channels: AI models value not only where you are mentioned, but also the credibility and permanence of those mentions. Being cited in reliable sources such as Wikipedia, news outlets, Reddit forums, or LinkedIn boosts authority. Studies show that Wikipedia alone accounts for nearly 10% of all AI engine citations, which makes it one of the most influential domains for effective GEO.
3. Deliver value: Make use of specific insights, data, and actionable advice while avoiding fluff. Structure content using natural language headings, FAQ formats, and concise lists that align with how users phrase questions. Refresh older content regularly, reuse proven assets, and include quotes from credible experts to strengthen authority signals.
4. Optimize for Google AI Overview: While still evolving, Gemini tends to favor conversational content, long-tail keywords, and structured answers. High-quality, well-formatted pages that directly match user intent and appear on authoritative domains are more likely to be selected. Supporting assets like visuals, pillar pages, and backlinks further enhance the chance of inclusion.
5. Reinforce credibility: Use structured data and external citations. Schema.org markup for articles, FAQs, organizations, and products ensure that AI engines can parse content accurately. External trust signals such as backlinks from high-authority sites and neutral, verifiable mentions are critical for notability and brand recognition in LLM datasets.

Conclusion: Future-proof your visibility with AIO today

To secure long-term visibility, AI Search Optimization demands a dual approach – strong roots in traditional SEO with a forward-looking overlay of AI-centric practices like GEO. As leading GEO agency, what. future-proofs your presence with AI-optimized approaches that drive visibility, relevance, and real results across every platform that AI touches.

Reach out to discover your custom GEO solution: Book your free consultation with what.

The post From SEO to AI Search – What Is AIO and GEO and Why It Matters appeared first on what. AG.

These technologies can unlock significant efficiency gains for smaller teams. However, they also introduce new data privacy and security risks that many organisations underestimate.

Traditional software systems operate within clearly defined rules. AI systems behave differently. They interpret natural language, generate responses dynamically, and often access data across multiple systems. AI agents go further by executing actions autonomously, interacting with internal tools, documents, and APIs.

For Swiss SMEs, this creates a key challenge: how to adopt AI safely while protecting sensitive data and complying with data protection requirements.

AI security is not just about protecting models. It requires careful control over how AI systems access, process, and expose data within the organisation.

Why AI systems create new data privacy risks

AI introduces security and privacy challenges that differ from traditional business software. These risks stem from how AI systems interact with data and integrate across tools.

AI systems interact with data dynamically

Most business applications follow fixed logic. Developers define how data flows through the system and what outputs are produced.

AI systems behave differently. They interpret prompts or instructions and generate responses based on patterns learned during training.

For example, an employee might ask an internal AI assistant:

“Summarise the latest sales strategy document.”

If the AI has access to internal file systems or document repositories, it may retrieve sensitive information. Without the right safeguards, it could surface content that should only be accessible to certain employees.

Because AI generates responses dynamically, data exposure can be harder to predict than in traditional software.

Training data can contain sensitive information

Many organisations improve AI performance by training models on internal data, such as:

• internal documentation
• customer communications
• operational data
• financial information

For SMEs, this can be especially attractive because it allows AI systems to understand company-specific processes.

However, if sensitive data is included without proper controls, models may inadvertently retain or reproduce elements of that data.

Even when using external AI providers, businesses must be careful about what information they submit to AI systems.

AI agents operate across multiple systems

AI agents extend AI capabilities by allowing systems to perform tasks across tools.

For example, an AI agent might:

• retrieve information from a CRM
• search internal documentation
• update records
• send communications

For a Swiss SME, this type of automation can significantly increase efficiency. But it also expands the potential attack surface.

If an AI agent has access to several systems, a security vulnerability could potentially expose or modify sensitive information across the organisation.

Related reading: Discover practical AI agents use cases for Swiss SMEs to see how automation works in practice.

Key data privacy risks when using AI

Organisations deploying AI systems face several common categories of privacy and security risk.

Data leakage through prompts

One of the most common risks arises when employees enter sensitive information into AI tools.

Examples include:

• customer contact information
• confidential client projects
• financial forecasts
• internal strategy documents

If these prompts are processed by external AI providers, organisations may lose control over how the data is handled.

For SMEs that work with client data or regulated industries, this can create serious compliance risks. Clear internal AI usage policies are essential.

Uncontrolled access to internal systems

Many AI implementations integrate with internal systems such as:

• CRMs
• document management systems
• databases
• internal knowledge bases

If permissions are too broad, AI tools may gain access to information beyond what is necessary.

For example, a marketing AI assistant might only need product documentation but could accidentally access confidential financial reports if permissions are poorly configured.

Applying the principle of least privilege is critical when securing AI systems. This becomes especially important when connecting AI to your broader tools integration infrastructure, where data flows across multiple business systems.

Model inference attacks

Advanced attackers may attempt to extract information from AI models themselves.

Through repeated queries, attackers can sometimes determine whether specific information was included in training data.

Although these attacks are relatively complex, they illustrate an important principle: AI models can unintentionally encode sensitive information.

This risk is particularly relevant for organisations training models on proprietary datasets.

Third-party model risks

Most SMEs rely on external AI platforms rather than hosting their own models.

This introduces important questions:

• Does the provider store prompts?
• Is the data used to train future models?
• Where is the data processed and stored?

Swiss companies must also consider data protection obligations under the Swiss Federal Act on Data Protection (FADP) and, in some cases, GDPR requirements.

Understanding how AI providers handle data is therefore essential.

Security risks specific to AI agents

AI agents introduce additional challenges compared to traditional AI tools.

Autonomous decision making

Agents are designed to perform tasks independently. These may include:

• updating records
• sending emails
• retrieving and processing data
• triggering workflows

While this automation can be extremely useful for SMEs with limited staff, it also means that systems may perform actions without direct human review.

If configured incorrectly, an agent could expose or modify sensitive information.

Human oversight remains important.

Prompt injection attacks

Prompt injection is an emerging threat in AI security.

In these attacks, malicious instructions are embedded inside content that the AI processes, such as:

• webpages
• emails
• documents

If an AI agent reads this content, it may follow the malicious instructions instead of its original task.

For example, a document could contain hidden instructions telling the agent to reveal internal data.

Protecting against this requires strict controls over how agents interpret and execute instructions.

Tool abuse and API exploits

AI agents often rely on tools and APIs to perform tasks.

These might include:

• CRM integrations
• financial systems
• messaging platforms
• scheduling tools

If attackers manipulate an agent’s behaviour, they may gain indirect access to these systems.

Careful management of API permissions and system integrations is therefore essential.

Also relevant: Learn how AI agents turn data overload into actionable insights while maintaining security.

Data privacy regulations and AI

Swiss SMEs deploying AI must also consider regulatory requirements.

The Swiss Federal Act on Data Protection (FADP) places obligations on how personal data is processed and protected. Companies operating internationally may also need to comply with GDPR.

Key principles include:

Data minimisation

Only the necessary data should be used in AI systems.

Training models on excessive datasets increases risk.

Purpose limitation

Data collected for one purpose should not automatically be reused for AI training without clear justification.

Transparency

Customers and users should understand when their data may be processed by AI systems.

Cross-border data processing

Many AI platforms process data outside Switzerland. Businesses must ensure that appropriate safeguards are in place.

Ignoring these issues can create both legal and reputational risk.

Best practices for securing AI systems

Swiss SMEs can reduce AI security risks by implementing a few practical safeguards.

Implement AI data governance

Companies should define clear policies covering:

• what data can be used with AI tools
• which datasets are restricted
• how AI tools are approved internally

Even small organisations benefit from clear AI usage guidelines.

Limit data access for AI systems

AI systems should only access the data necessary to perform their tasks.

This includes:

• restricting database permissions
• separating sensitive systems
• limiting document access

Applying least privilege access significantly reduces potential exposure.

Monitor AI inputs and outputs

Logging and monitoring AI activity helps detect unusual behaviour.

Companies should track:

• prompts entered into AI systems
• responses generated by models
• actions triggered by agents

This visibility is essential for identifying potential misuse.

Use secure model deployment

Some SMEs may choose enterprise AI platforms that offer stronger security controls, such as:

• private deployments
• enhanced access controls
• secure data handling policies

This can provide more confidence when working with sensitive data.

Implement human oversight

Fully autonomous systems remain risky.

Introducing checkpoints for sensitive actions – such as financial updates or customer communication – can prevent costly mistakes.

Related reading: Before implementing AI systems, consider fixing your workflows first to ensure your foundation is solid.

A practical framework for AI privacy & security

A structured approach can help SMEs implement AI safely.

One useful model is the AI Security Stack.

1. Data governance

Define policies around:

• training data
• prompt usage
• sensitive information handling

2. Model security

Protect the model itself through:

• secure hosting environments
• access controls
• version management

3. Prompt and interaction controls

Implement safeguards to prevent prompt injection or malicious instructions.

This includes:

• validating inputs
• limiting model behaviour
• restricting access to sensitive tools

4. System permissions

Carefully manage what AI systems can access.

Agents should only interact with approved systems using limited permissions.

5. Monitoring and auditing

Continuously monitor AI activity.

Audit logs should track:

• queries
• responses
• system actions

This creates accountability and supports incident investigation.

The future of AI security

AI adoption among Swiss SMEs will continue to grow as tools become more accessible and powerful.

At the same time, regulators and security experts are placing greater emphasis on AI governance and responsible deployment.

Emerging frameworks such as the NIST AI Risk Management Framework and evolving European regulations highlight the importance of structured oversight.

For SMEs, addressing AI privacy and security early provides a clear advantage. Businesses that implement responsible AI practices will be better positioned to build customer trust and long-term resilience.

Also relevant: Understand why your business needs AI automation in the first place, and how to do it responsibly.

Conclusion

AI systems and autonomous agents offer significant opportunities for Swiss SMEs, enabling smaller teams to automate work and operate more efficiently.

However, these systems also introduce new challenges around data privacy, system access, and security risks.

Because AI interacts dynamically with data and increasingly performs actions across systems, traditional security approaches are no longer sufficient.

Swiss organisations adopting AI should focus on:

• clear data governance
• controlled system access
• monitoring AI interactions
• implementing human oversight

By building security and privacy into AI deployments from the start, SMEs can safely unlock the benefits of AI while protecting their data, customers, and reputation.

If your organisation is exploring AI tools or AI agents, it’s important to ensure your systems are designed with strong data privacy and security practices from the outset.

Ready to implement AI securely? As specialists in AI automation services, we help Swiss SMEs design and deploy intelligent automation that prioritizes security, data privacy, and compliance from day one. Our approach combines technical expertise with an understanding of Swiss regulatory requirements, ensuring your AI systems deliver efficiency gains without compromising on protection. 

Get in touch with our team to assess your current setup and build a secure AI strategy tailored to your business needs.

The post Data Privacy & Security for AI Systems and Agents appeared first on what. AG.

Throwing AI at a chaotic back office is like installing a turbocharger on a car with flat tires. Before you think about AI transformation, there’s a more valuable question to ask.

The Swiss SME Reality: Great at the Core, Painful Around It

Swiss SMEs are often excellent at what they do. Whether it’s manufacturing, logistics, professional services, trade, or healthcare-adjacent work – the craft is usually solid. 

What gets painful is everything surrounding the core product:

• Customer data gets manually copied from the ERP into accounting, then again into the CRM. 
• Quotes get built with twelve browser tabs open. 
• Approvals bounce around on WhatsApp. 
• Recurring tasks happen manually not because anyone chose that approach, but because “the software can’t do it” – or more often, because nobody ever configured it properly. 
• Reporting depends on one person who somehow “knows the magic,” and the moment they’re on holiday, everything stalls.

Switzerland adds its own layer of complexity here. Multiple languages, tight compliance expectations, high labour costs, and clients who expect calm, competent service – that combination means every hour burned on admin drag is a genuinely expensive hour. It’s not just inefficient. It’s costly in a way that compounds quietly over time.

So when an SME asks “should we be using AI?”, the better question is: where exactly is time leaking today, and why?

What to Do Before You Touch AI: A Practical Audit

This isn’t a six-month consulting ritual. It’s a clear-eyed look at how work actually flows through your business – and where it gets stuck.

Step 1: Map the Work as It Actually Happens

Pick one workflow that feels expensive. Customer onboarding, invoice processing, building quotes, handling service requests, scheduling – pick the one that your team would describe as “always a mess.” 

Then map it in plain language: who kicks it off, what information they need, where that information lives, how many handoffs are involved, and where things reliably get stuck or re-entered.

You’re not hunting for AI opportunities here. You’re hunting for friction.

Step 2: Put a Number on the Admin Tax

Most SMEs underestimate how much time disappears into administrative busywork, because it’s spread across many people in small increments. Two minutes here, five minutes there – it doesn’t feel significant until you add it up across a month and realize it’s quietly become a second job.

Spend one to two weeks tracking time spent on copy-pasting between systems, searching for the latest version of a file, chasing approvals, reconciling mismatched data, and fixing errors that only exist because a manual step went wrong.

The output of this exercise is a number. And numbers beat gut feelings every time.

Step 3: Fix the System Before You Automate It

Here’s the trap that catches a lot of businesses: automating a broken process doesn’t fix it. It just makes the confusion run faster.

Before anything gets automated:

• Define one source of truth for customer data.
• Standardize your templates – quotes, invoices, intake forms. 
• Reduce approval ping-pong by clarifying who actually owns each decision. 
• Clean up naming conventions and folder structures. 
• And critically: remove any process steps that only exist because “we don’t trust the data” – because that’s a signal the data problem needs to be solved first, not worked around.

Once the system makes sense, then you decide what deserves automation.

Why “Boring” Automation Usually Wins Over AI

Here’s the part that most AI vendors won’t tell you: some of the biggest efficiency gains in Swiss SMEs come from changes that are almost embarrassingly unsexy.

• Automatic invoice matching rules,
• Structured intake forms instead of free-text emails,
• A clean CRM pipeline with required fields that actually get filled in,
• A quote builder that pulls from one consistent data source,
• Scheduled exports into accounting instead of someone uploading a file manually every Friday,
• Document generation from templates,
• And – arguably most impactful – proper integration between your shop, ERP, and accounting system so data stops having to be manually teleported between platforms.

None of that requires AI. It requires systems that actually talk to each other, and workflows that reflect how work should flow rather than how it grew organically over years.

This is where the real savings show up quickly. In admin-heavy environments, it’s not unusual to find that 40–50% of effort is avoidable once you remove duplicate entry, rework, and time spent searching for information that should be findable in seconds. Not because anyone is being lazy. Because the system is noisy.

The key insight is simple: automation doesn’t need to be clever to be valuable. It needs to be reliable.

So When Does AI Actually Make Sense?

AI automation becomes genuinely compelling under a specific set of conditions – and it’s worth being honest about what those are:

• You’re dealing with large volumes of unstructured content – emails, PDFs, support tickets, notes – and you know clearly what you want extracted or acted on. 
• A process is stable enough that you can define what “good output” looks like.
• Your data quality is already decent, or you’ve done the work to fix it.
• The failure modes are acceptable and you have the monitoring in place to catch when things go sideways.
• The ROI is proven and measurable.

In other words: AI is great once your house has walls.

If the foundation is missing, AI tends to become a demo that never makes it into daily operations. It gets piloted, gets praised in a meeting, and then quietly stops being used because nobody really trusts it – or it keeps producing outputs that require more human correction than the original manual process did.

The most common reason AI transformations fail isn’t the technology – it’s the missing groundwork underneath it. Read more about it in our blog post about 5 mistakes you’re probably making in your AI transformation.

A Simple Decision Framework for SMEs

If you want a sane way to sequence this, here it is:

• Process first. Map what’s happening, find the friction, simplify before you optimize.

• Automation second. Connect your systems, standardize your templates, eliminate the manual steps that don’t need to be manual.

• AI third. Apply it where it uniquely helps – text-heavy workflows, tasks that require contextual judgment at scale, triage, summarization, or handling unstructured inputs.

If you start with the last step, you risk buying a sophisticated solution to a fundamentally boring problem. And boring problems are exactly where money quietly leaks.

Clean Workflows First, AI as a Multiplier

The real competitive advantage for Swiss SMEs in 2026 isn’t having AI. It’s having clean data, clear processes, and reliable automation that works every single day without someone manually pushing it along.

Once you have that foundation in place, AI stops being a gamble and starts being a genuine multiplier. Not magic. Not hype. Just leverage – applied to a system that’s actually ready for it.

If you’re not sure where to start – whether that’s cleaning up your process landscape, connecting your tools, or figuring out where AI automation actually fits – it’s best to seek help from a specialized agency.

As an AI automation agency with deep expertise in tools integration, we work with Swiss SMEs to identify the highest-value opportunities and build systems that deliver measurable results.

The post Before You Touch AI, Fix Your Workflows First appeared first on what. AG.