What is AI visibility, and why now?
AI visibility is the degree to which generative AI systems mention your organisation and cite your pages as a source. The discipline is called Generative Engine Optimization, or GEO. The term covers three distinct surfaces: the AI Overview above regular search results in Google, the answers from assistants such as ChatGPT, Claude, Gemini and Perplexity, and the new generation of AI agents that choose on a user's behalf.
Google showed your page and the user clicked. An AI assistant reads your page, extracts an answer and shows that answer instead. The click has become optional. You now optimise to be lifted, not just listed.
That shift is already visible in the search results themselves. Of the eight Dutch business search queries I measured for this article, three showed an AI Overview above the organic results. For a third of those queries, the first answer no longer comes from a website at all.
And it moves fast. For the query "AI visibility" (in Dutch: "vindbaarheid in ai"), the best-ranking Dutch article right now dates from May 2025. On this topic, fifteen months old is simply outdated: it says nothing about how AI Overviews currently select sources.
What I notice at the boardroom table: AI visibility is still filed under marketing almost everywhere. That's a category error. When an AI assistant recommends an insurer or adviser on a customer's behalf, that is distribution. In insurance we have seen this before with comparison sites. Whoever was late back then paid for it for a decade.
What 136 measured search results show
Most of what gets written about AI visibility is advice without measurement: use structured data, write clear answers. Probably true, but rarely backed by numbers. So I measured it myself.
What I did. I pulled the complete search results via Ahrefs for eight Dutch business search queries: ai act, ai training, ai workshop, ai literacy training, ai customer service, ai marketing agency, build a website with ai and ai visibility. From those I took every organic position with a known domain authority: 136 data points. For each result I have the position, the Domain Rating (DR, Ahrefs' authority score on a scale of 100), the number of referring domains and the estimated monthly traffic.
I then calculated the rank correlation (Spearman) between authority and position. A strong negative value would mean: the higher the authority, the better the position. A value near zero means authority predicts nothing.
| Metric | Result |
|---|---|
| Data points | 136 organic positions across 8 search queries |
| Correlation, domain authority and position | Spearman -0.19 (weak) |
| Correlation, referring domains and position | Spearman -0.13 (very weak) |
| Top-10 positions with zero backlinks | 38 percent (26 of 68) |
| Median referring domains, positions 1–3 | 9 |
| Median referring domains, position 20 and below | 1 |
| Median DR, positions 1–3 | 67 |
| Median DR, position 20 and below | 53 |
Source: own analysis of Ahrefs SERP Overview data, Google Netherlands, retrieved 24 August 2026.
Two numbers stand out. Thirty-eight percent of all top-10 positions across these eight queries have zero referring domains. No backlink at all, and still on page one. The median top-3 position has nine, a number you can reach with a handful of good guest contributions.
38%
Top-10 positions with zero backlinks
26 of the 68 measured top-10 positions have zero referring domains.
Median number of referring domains
The gap between the top three and the bottom tiers is small: median DR 67 versus 53. Fourteen points on a logarithmic scale of a hundred. If authority were decisive, that gap would be larger.
Why domain authority predicts less than you think
Behind that -0.19 sit large differences. The correlation varies sharply by query.
Correlation between domain authority and position, per query
Source: own analysis of Ahrefs SERP Overview data, Google Netherlands, 24 August 2026. Negative means: higher authority correlates with a better position.
For ai act, authority counts heavily. At the top sit the European Commission, Wikipedia, the Dutch data protection authority and the national government. On an established legal topic, the institution wins.
For ai workshop, the correlation is +0.07 — essentially zero. The number-one result for that query has DR 8. Number three has DR 7 and zero referring domains. Meanwhile a university with DR 84 sits at position 2, and a university of applied sciences with DR 76 at position 7. The authority order is completely scrambled.
For ai training the picture is similar: -0.11. A provider with DR 51 sits at position 1, a Microsoft page (DR 96) at 25, and LinkedIn (DR 99) at 32.
The newer and more commercial the topic, the less authority predicts. For established informational topics, positions are locked in and there is little to gain. For new topics the pecking order does not exist yet, and relevance decides.
For an organisation starting now, that is the opening. Pick a topic where the order is still unsettled and put the best answer there.
What I notice at the boardroom table is that this runs against intuition. The reflex is: we're a small brand, we'll never break in there. For AI Act, that's true. For AI for accountants or AI visibility for notaries, the data shows something different, and that window will not stay open forever.
How AI answers choose their sources
Inside the AI Overview you can see exactly which sources Google cites. That's where citation and authority diverge most clearly.
For the query AI visibility, the AI Overview cites three Dutch sources. The first has DR 20, one referring domain and five visitors a month. The second has DR 30, one referring domain and zero measured visitors. The third has DR 40 and zero referring domains. None of the three is a recognisable name.
For AI literacy training, the picture is the same. Alongside the Dutch data protection authority (DR 92), the AI Overview cites three commercial training providers with DR 53, DR 38 and DR 34, all with zero or one referring domain and zero measured organic traffic.
Pages with no measurable search traffic at all get cited in the AI Overview. They draw no visitors from Google, yet appear in the answer shown above it. Your traffic dashboards show none of this.
- 1
Someone asks a question
- 2
AI Overview or assistant answer
- 3
Model selects citable sources
- 4
Your page gets mentioned or ignored
What the cited pages share is their shape: a clear question as the heading, a direct answer in the first sentences beneath it, and self-contained blocks you can lift out of the page without losing their meaning. Their authority they do not share.
This connects to what I wrote earlier about what an AI Overview means for your website and about who is liable when an AI Overview takes your traffic.
What this changes on your website today
Four changes you can make today, with no link-building budget.
Write one complete answer per section. An AI assistant does not cite pages, it cites paragraphs. Every H2 on your site should be a concrete question, followed by an answer that still holds true if lifted straight out of the page. References like "as described above" make a block useless to cite.
Attach a source to every claim. Figures without a source get skipped. Figures with a direct link to a primary source get reused, because the model can carry that link into its answer.
Make your recency visible. A "last updated" date in the visible text and in structured data helps you beat an outdated competing article. The best-ranking Dutch piece on this topic is fifteen months old, and it shows in the content.
Ship a clean text version. AI agents parse clean markdown more reliably than HTML full of components. A `.md` variant of your key pages, announced in the head with a ``, raises the odds a model cites you correctly.
Self-check
- Every H2 is a real question someone would type
- The answer appears in the first two sentences under the heading
- Every figure has a direct source link
- A visible last-updated date appears in the text
- Definitions are short, self-contained sentences of 15 to 20 words
- There is an FAQ using literal question phrasings
- The key pages are also available as clean markdown
How to measure AI visibility
Your regular analytics tell you nothing here. A mention inside a ChatGPT answer generates no visit, so your dashboard stays flat while your visibility grows or collapses.
There are three measurement layers, from coarse to fine.
| Layer | What you measure | How |
|---|---|---|
| Citations | How often AI platforms link to your domain | Ahrefs Brand Radar or similar AI-citation monitoring |
| Mentions | Whether your brand is named without a link | Repeat a fixed set of prompts periodically and log the answers |
| Referral traffic | Visits that actually originate from AI platforms | Referrer segment in your analytics for chatgpt.com, perplexity.ai and gemini.google.com |
The second layer is the cheapest and the one most often skipped. Write twenty prompts a customer would genuinely ask, run them through three or four assistants every month, and log verbatim who gets named. It's manual work, and it's the most reliable way to see whether you appear in the answer and your competitor does not.
What I notice here: most organisations start at layer three because it fits an existing dashboard, then conclude that "barely any traffic comes from AI." That's also true. It's just the wrong number. The real question is how often your name appears in the answer that made the click unnecessary.
The six biggest misconceptions about AI visibility
Misconception: AI visibility is just SEO with a new name. They overlap partly. Classic SEO optimises for a position in a list; GEO optimises for reuse inside an answer. A page can be highly citable while barely ranking, as the cited sources with zero organic traffic show.
Misconception: you need domain authority first. Of the measured top-10 positions, 38 percent have zero referring domains, and the AI Overview cites sources with DR 20. For new topics, authority is not an entry requirement.
Misconception: more content means more visibility. Volume without structure does not raise your odds. One current, sourced piece beats twenty generic blog posts, because a model needs one block it can lift.
Misconception: structured data is the deciding factor. Schema.org helps with comprehension and rich results, but it does not force citation. The cited pages in my measurement stand out mainly through their text structure.
Misconception: if AI takes my traffic, there's nothing I can do. Value shifts from click to mention. Being named in the answer a customer reads has commercial value, you just won't see it in your old dashboard.
Misconception: this is something for a year or two from now. The window is open precisely now, because the pecking order is not yet fixed. Once established players claim these topics, ordinary authority logic applies again.
How to tackle AI visibility (step-by-step plan)
- Pick topics where the pecking order is not yet fixed. Measure, per topic, whether authority explains the positions. If it explains them strongly, pick a different topic or a narrower angle. What you don't do: compete with the European Commission on "AI Act".
- Phrase the questions your customer actually asks. Use the sentence someone types or puts to an assistant, not your service names. Those sentences become your headings.
- Build one citable block per question. Question as heading, answer in the first two sentences, supporting detail beneath, source attached. What you don't do: build a narrative where the answer only arrives in paragraph six.
- Add the layer only you can supply. Your own numbers, your own practice, a sector you know from the inside. Without that layer, your piece is interchangeable with what a model generates itself, and then the model doesn't cite you, it replaces you.
- Publish a clean text version too. A markdown variant of your core pages, announced in the head. What you don't do: hide everything in JavaScript components a crawler can't read.
- Measure from day one with a fixed prompt set. Twenty prompts, monthly, multiple assistants, log the results. Without a baseline you won't know in six months whether anything worked.
- Revise instead of only adding. Recency is a ranking factor on this topic. One piece you revise every quarter is worth more than four new pieces that go stale.
AI visibility and the rules
Two things worth watching.
The first is content liability. If an AI assistant cites your content and distorts it, your name stays attached underneath. For regulated sectors, insurance and financial advice first among them, that is not a theoretical risk. A misquoted policy term or premium is a statement a supervisor can look into. I wrote about this earlier in the piece on liability around AI Overviews.
The second is the wider body of AI regulation. The EU AI Act (Regulation 2024/1689) does not touch AI visibility directly: optimising your own content is not an AI system you are placing on the market. But the moment you use AI yourself to generate that content or to answer customer questions, you fall within the scope of the regulation and its transparency obligations. What applies to your organisation then is covered in the EU AI Act hub and in the AI Act Impact Scanner.
For content in sensitive categories, money and health first among them, search engines and models also apply stricter quality standards. What that means is covered on the YMYL definitions page.
Frequently asked questions
How do you get found in AI?
By building pages that can be cited verbatim. That means: every section answers one concrete question, the answer sits in the first two sentences, every claim has a source, and there's a visible last-updated date. Authority helps, but for new topics it isn't a requirement: in my measurement of 136 Dutch search results, 38 percent of top-10 positions have zero backlinks.
What is the difference between SEO and GEO?
SEO optimises for a position in a list of search results. GEO, Generative Engine Optimization, optimises for reuse inside a generated answer. They overlap in technique and quality, but differ in goal. A page can be cited in an AI Overview without ranking meaningfully in Google, and vice versa.
Do I need backlinks to be mentioned in ChatGPT?
Not necessarily. The AI Overview for the query "AI visibility" cites sources with one or zero referring domains and almost no organic traffic. For established informational topics it's different: there, institutions and encyclopedias dominate.
How do I measure whether AI mentions my business?
Using three layers. Citation monitoring through a tool like Ahrefs Brand Radar, your own prompt set of roughly twenty customer questions run monthly through multiple assistants and logged, and a referrer segment in your analytics for visits from chatgpt.com, perplexity.ai and gemini.google.com. The middle layer is the cheapest and most often forgotten.
Does AI visibility cost me search traffic?
Yes, measured in clicks, probably. An answer that resolves the question makes the click unnecessary. Value shifts from visit to mention: being named in the answer your customer reads is commercially relevant, even without a visit. That does require a different dashboard.
Is an llms.txt file necessary?
It's a proposed file format for stating which content AI systems may use, similar to robots.txt. Support is not yet universal and it is no substitute for citable content. Treat it as cheap hygiene, not the core of your approach.
Does AI visibility work for local or profession-specific services too?
Especially there. Queries like "AI visibility for accountants" or "AI for notaries" have low volume, but competition is scarce and search intent is sharp. This is exactly the type of topic where the pecking order is not yet fixed.
How often do I need to update this kind of content?
Every quarter for the core pages. Recency weighs heavily in AI citation, and content on this topic visibly ages within a year. The best-ranking Dutch article on AI visibility dates from May 2025 and misses everything that has changed since.
Go deeper
What AI does to search and websites
- What an AI Overview means for your website — the basics of how AI Overviews affect your search traffic.
- AI Overviews, liability and the search traffic that returns — who is responsible when your content is reused in distorted form.
- Do we still need websites in 2026? — the fundamental question behind this whole theme.
- Stack Overflow is dead — what happens to a platform once AI takes over the answers.
How AI changes the buying process
- Agentic commerce in 2026 — if an agent chooses on the customer's behalf, who does it even see?
- Why OpenAI is becoming your competitor — the platform layer stepping between you and your customer.
- Four AI winners in the Google consumer market — who is redividing the search market.
- Recent data decides which search engine wins — why recency is the deciding factor.
Related guides
- EU AI Act — what the regulation means for your organisation.
- AI governance — how to set up oversight of AI use.
- YMYL — why content about money and health is judged more strictly.
About Marc Diks
Marc Diks has worked in insurance for more than 25 years and focuses on AI strategy at board level. He also builds AI applications himself, without a formal coding background. That combination means he writes about how these systems actually work, from practice. More about his background on the about page.