AI Visibility of Dutch SMEs: Research Across 310 Websites

But how well prepared are Dutch SME websites for that shift? To measure it, we scanned the websites of 310 companies in September and October 2026. We did not only look at whether AI mentions them, but above all at whether their website provides the foundation to be mentioned. Below are all the findings, by topic and by sector, each set alongside what scientific research says about it.

Short answer: Most Dutch SME websites are technically in good shape for Google, but not ready for AI visibility. On one in five, ChatGPT's crawler cannot get in, half write too vaguely to be cited, 41 per cent show no readable date and nine in ten do not tell a computer where they are located.

What are the key figures on the AI visibility of SME websites?

Eight figures summarise the research. Each figure stands for a concrete point at which AI assistants can skip a website.

Which websites did we study, and what did we measure?

We analysed 310 business websites, mainly of Dutch SMEs. The largest group consists of physiotherapy practices (209 websites), followed by engineering companies (30) and installation companies (27). The remaining 44 websites come from a range of industries.

We assessed each website against five questions that together determine AI visibility:

1. Access: can an AI assistant's crawler open the website?
2. Recognisability: does the website tell a computer who the company is and where it is located?
3. Content: does it contain concrete facts that an AI can pick up?
4. Freshness: can a computer see how recent the information is?
5. Foundation in Google: is the technical health for search engines in order?

The measurements were taken with the AIO Clicks audit tool: a technical scan, public data and AI-assisted content analysis. This is a snapshot; AI answers and websites change constantly.

Can AI actually get into these websites?

Before ChatGPT can say anything about a company, it has to be able to read the website. OpenAI does this with a crawler called GPTBot. Other AI companies have similar programs. A website can deliberately block such crawlers via robots.txt, a small file of house rules for visiting programs.

Deliberate blocking turned out to be rare. Of the 291 websites whose robots.txt we could read, exactly one excluded GPTBot. Yet GPTBot got no or insufficient access on 60 of the 296 websites tested. An ordinary browser opened the same page without any problem. In most cases the server responded with error code 403: access denied.

For AI visibility this is the first hurdle, and the gap between those two figures is telling. Nobody made a choice here; the block arises somewhere else. The pattern points to security: firewalls and bot protection from hosting providers that automatically stop 'suspicious' visitors, shutting out AI crawlers in the process.

The problem is not limited to outdated websites. Among technically strong websites the share was 21 per cent, barely different from the average. By sector, engineering companies stand out: there GPTBot could not get in on 27 per cent, against 19 per cent for physiotherapists and installers.

Google itself says that no special files or tricks are needed to appear in its AI answers (Google Search Central). But a website that does not let an AI crawler in cannot be read either, however good the content.

How to check it yourself: ask your hosting provider whether their firewall or bot protection stops visitors named GPTBot, OAI-SearchBot or ClaudeBot. If you have access to the server logs, you can also search for GPTBot requests that were answered with a 403.

Do AI systems know who these companies are?

An AI assistant not only has to be able to read pages, but also to understand that those pages belong to one recognisable company, with a fixed address and a clear industry. That is what structured data is for: invisible labels in a website's code that tell a computer exactly 'this is the company name, this is the address, this is the phone number'. The common form is called Organization or LocalBusiness schema.

Here the data shows a striking contradiction. Of the companies, 88 per cent are on Google Maps; among installation companies, every single one is. So Google usually knows perfectly well where these companies are. But their own website does not tell a computer:

• 89 per cent do not state their address in machine-readable form;
• 91 per cent do the same with their phone number;
• 52 per cent have no Organization or LocalBusiness schema at all;
• 81 per cent do not refer anywhere to profiles elsewhere, such as LinkedIn, the Google Business Profile or an industry association. Among installation companies that figure is 96 per cent.

Those references to other profiles matter more for AI visibility than they seem. An AI system builds its picture of a company from multiple places on the web. An Ahrefs analysis of 75,000 brands showed that the number of times a brand is mentioned elsewhere on the web correlates more strongly with visibility in AI answers than the authority score based on links to a website. A review study of ranking factors sets that finding alongside other research and reaches the same conclusion: authority is shifting from links to recognisability (Kargaev, 2026).

For SMEs there is good news here. Research from the University of Toronto shows that, for well-known brands, AI models rely mainly on what they already learned during training. For smaller and specialised companies, the information the model finds online directly determines who makes it into the answer (Chen et al., 2026). What you publish about yourself as a small business therefore carries relatively heavy weight.

How to check it yourself: enter your homepage URL into Google's Rich Results Test. If no Organization or LocalBusiness item appears with a name, address and phone number, your website is missing this foundation.

Is there anything on these websites that AI can pick up?

This is the area with the biggest gap in AI visibility. An AI assistant looks for building blocks for its answer: prices, dimensions, quantities, delivery times and explanations. Generic sentences give it nothing.

We had each website's copy assessed on three points: how concrete it is, whether it explains concepts and compares options, and how completely it covers the field. The result:

• Half of the websites score weakly on concreteness; only 12 per cent score strongly.
• Half also score weakly on explanation and comparison; only 7 per cent strongly.
• Just 9 per cent cover their own field completely.
• A typical website contains fewer than three concrete data points per 1,000 words, such as a price, a dimension or a lead time. On 22 per cent of websites we found none at all.

The differences between sectors are large. Among installation companies 86 per cent score weakly on concreteness, among engineering companies 80 per cent and among physiotherapists 45 per cent. On half of the installation companies' websites we found not a single concrete data point in the copy.

The examples from the research show what makes the difference. We came across sentences like these a lot:

'High quality at a fair price.'

'Our own skilled professionals give targeted, expert advice.'

They sound fine, but an AI can do nothing with them: they fit any website in any industry. Compare that with these sentences from other companies in the research:

'Standard physiotherapy session: € 35.00. First treatment, including intake and assessment: € 45.00.'

'Urinary incontinence: 9 treatments covered by basic health insurance (with a referral from your doctor).'

'In anodising, the aluminium surface is treated electrochemically, creating a hard, protective oxide layer that is fully bonded to the metal.'

The first two literally answer the questions people ask an AI assistant: what does it cost and what is covered? The third explains a concept, exactly the kind of text AI likes to pick up. This also explains why physiotherapists are ahead: many practices put their rates and reimbursements per treatment on their website.

Why source citations are so rare, and why that matters

On 354 pages with figures, prices or other claims, we found a total of three links to an independent source, such as Statistics Netherlands (CBS), the government or scientific research. That is less than one per cent.

This is precisely where there is an opportunity for better AI visibility. In controlled experiments, adding sources, statistics and quotations increased a text's visibility in AI answers by up to 40 per cent (Aggarwal et al., 2024). An analysis of more than 21,000 citations in AI answers showed that pages with figures, definitions or comparisons carry through 55 to 62 per cent more strongly into the text the AI ultimately writes (Zhang et al., 2026).

Does an FAQ block help?

An FAQ block is often recommended as a quick fix for AI visibility. Only 6 per cent of the websites had an FAQ block that was also marked up as such for computers. Yet that need not be a concern: in the same analysis of 21,000 citations, a question-and-answer format offered no advantage on its own (Zhang et al., 2026). An FAQ block only works if the answers themselves contain concrete facts, explanations or comparisons.

How to check it yourself: take your three most important pages and count the concrete data points. Do they include prices or price indications, delivery times, dimensions, quantities or reimbursements? Do you explain at least one concept and compare at least two options? If not, an AI has little to pick up.

Do these websites show how up to date they are?

AI search engines clearly favour recent information more often. Research from the University of Toronto found that half of the pages AI assistants cite in consumer electronics are younger than 62 to 90 days, against 130 days for Google. In the automotive sector the difference was even greater: 148 to 217 days against 493 days (Chen et al., 2026).

Here too AI visibility is lost: an AI can only see that a page is recent if there is a date a computer can read. On 41 per cent of the websites we found none at all. Among engineering companies that was 64 per cent, among installation companies 57 per cent and among physiotherapists 37 per cent.

What stands out is that the problem usually does not lie in the content. Of the websites that do have a readable date, 77 per cent were updated in the past year. So many companies do keep their website up to date, but nobody can see it.

How to check it yourself: do your most important pages show a publication or update date? Ask your web developer whether that date is also in the code, for example as dateModified in the page's structured data.

How healthy is the technical foundation for Google?

Here the news is mostly good. The websites' technical search health has a median score of 87 out of 100, and 72 per cent score 80 or higher. There is hardly any difference between sectors. Virtually every website has a secure connection (99 per cent) and most have a sitemap (86 per cent).

There are two points of attention:

• Mobile speed: the median score is 62 out of 100, and 18 per cent of websites fall below 50.
• Too little visitor data: for 71 per cent of the websites, Google has insufficient data from real visitors to assess the user experience (Core Web Vitals). That is typical of smaller websites, but it does mean Google has fewer signals to rely on.

That technical foundation matters, for AI visibility too: it is the base on which everything else rests. According to Google, good SEO is the foundation for appearing in its AI answers (Google Search Central). A review study of ranking factors accordingly describes AI search as an extra layer on top of the organic search results, not a replacement for them (Kargaev, 2026).

Google knows the name, but not the trade

A technically healthy website is not yet a findable website. Almost half of the companies (46 per cent) do not rank in Google's top 10 for any search term except their own company name. Among installation companies that figure is 78 per cent. Anyone who already knows the company's name will find it. Anyone searching for the service the company provides will not. That also affects AI visibility, because AI answers in Google build on the regular search results.

What does llms.txt do?

An llms.txt file is meant to tell AI programs briefly what a website contains. It is often recommended in advice on AI visibility. 22 per cent of the websites in our research had one. Google, however, says that Google Search does not use the file: it neither helps nor harms (Google Search Central). A telling detail: nine websites with an llms.txt were keeping GPTBot out at the same time. The welcome sign was hanging on a locked door.

How do physiotherapy, installation and engineering differ?

The three largest sectors in the research each show their own pattern of AI visibility.

Physiotherapy: concrete, thanks to the rates. Physiotherapists score best on concrete content. Many practices publish their rates, reimbursements and treatments per complaint: exactly the information people ask about. Their weak point is recognisability: four in five do not refer anywhere to external profiles.

Installation: easy to find on the map, not in the copy. Every installation company in the research is on Google Maps. But their copy is the least concrete of all sectors, and 78 per cent are found in Google only by their own name. An AI looking for 'how much does a heat pump cost' finds little to go on with these companies.

Engineering: the door and the date. Engineering companies most often have a website GPTBot cannot get into (27 per cent) and most often have no readable date (64 per cent). Their copy is also predominantly vague, while this is precisely the sector with a great deal of specialist knowledge to share.

The installation and engineering groups are smaller than physiotherapy. So treat the differences as an indication, not an exact measure.

Why does a good technical score say so little about AI visibility?

Many companies assume that a technically sound website will automatically be easy for AI to find as well. Our data contradicts that. We found no relationship whatsoever between technical search health and the concreteness of the copy.

Take the technically strong websites, with a score of 80 or higher. Of the 171 websites for which we had all the data, 77 per cent fell short on at least two of the five AI points: access for GPTBot, structured data, references to profiles, concrete copy and a readable date. Only 9 per cent had all five in order.

So many SMEs have two report grades without knowing it: a pass for Google and a fail for AI.

There is one caveat about measuring AI visibility. If you check your AI visibility by asking ChatGPT a question once, you see a snapshot. Researchers at the University of St. Gallen asked four AI search engines the same questions every day for six weeks. Of the sources mentioned on one day, on average only a third came back the next day (Schulte et al., 2026). In another study, how often a brand was mentioned fluctuated by up to 40 percentage points within two weeks (Luther & Touboul-Cohen, 2026). What counts is the trend across many questions and several days.

What can you do as a business now?

The five points from the research are also a to-do list for better AI visibility. None of the steps requires a new website.

1. Let AI in. Ask your hosting provider whether GPTBot and other AI crawlers are being stopped by security or a firewall. This is usually a single setting.
2. Say who you are. Put your company name, address and phone number in Organization or LocalBusiness schema, and refer in it to your profiles on LinkedIn, Google and with your industry association.
3. Write concretely. Replace generic promises with prices or price indications, delivery times, dimensions, quantities and explanations. Where possible, refer to an independent source, such as Statistics Netherlands (CBS) or the government.
4. Show that you are up to date. Give your most important pages a date that both visitors and computers can read.
5. Be found for your trade, not just your name. Create a page for each service that answers customers' questions, so that Google and AI also find you when someone does not yet know your name.

Rather not work it all out yourself? The free analysis from AIO Clicks checks these five points for your own website in one go.

How does AIO Clicks help businesses with AI visibility?

AIO Clicks is a digital visibility agency based in Haaksbergen. We help businesses in the Netherlands, Belgium and Germany to be found in Google and to be mentioned by AI assistants such as ChatGPT, Gemini and Perplexity.

The five points from this research form the core of our approach to AI search and GEO: we make sure AI crawlers can read your website, that AI systems know who you are, that your copy contains concrete and verifiable information and that it is visible how up to date it is. We also work on the technical foundation and findability in Google, on content that converts and on reputation and trust in the places where AI gets its information.

Want to know how your website scores on the points from this research? Request the free analysis and receive a report on your visibility in Google and AI within minutes.

What questions do we get asked most about AI visibility?

How many Dutch websites keep ChatGPT out?

In our research, ChatGPT's crawler got no or insufficient access on one in five websites. Only one website did so deliberately via robots.txt. For the rest, the hosting provider's security appears to be the cause.

What is GPTBot?

GPTBot is the program OpenAI uses to read websites. What GPTBot cannot read, ChatGPT cannot use to tell people about your business either. Other AI companies use similar crawlers, such as ClaudeBot and PerplexityBot.

Does an llms.txt file help AI visibility?

Not for Google: Google Search does not use the file, and it neither helps nor harms your position. Some other AI services may read it. It is little work, but first make sure AI crawlers can open your website at all.

Is an FAQ block worthwhile for AI?

Only if the answers contain concrete information. Research shows that a question-and-answer format offers no advantage on its own; facts, definitions and comparisons do. So an FAQ block with generic answers does little to help.

How often should you measure your AI visibility?

Not once, but repeatedly. AI answers change by the day and even by the minute. Researchers advise asking the same question at least seven times and measuring over several days before drawing conclusions.

How do I know whether my website is ready for AI?

Go through the five points from this report: access, recognisability, concrete content, freshness and the foundation in Google. It is quicker with the free analysis from AIO Clicks, which summarises these points for your website in a single report.

What does this research mean for Dutch SMEs?

The SME websites in our research are technically healthy, but do not yet speak the language of AI. The good news is that the fixes for better AI visibility are small: a setting with the hosting provider, a few lines of structured data, more concrete copy and a visible date. No new website, no big budget.

Putting it off is riskier than many businesses think, though. People who see an AI summary in Google click through to a website far less often: in 8 per cent of cases, against 15 per cent without a summary (Pew Research Center, 2025). What is more, people trust AI answers with source citations more, even when those sources are wrong (Li & Aral, 2025). The AI's answer thus becomes the first impression of your business. At the same time, the traffic is not disappearing: people who start using AI assistants actually visit more different websites (Gholami et al., 2026). Whoever appears in that answer stands the best chance of winning those visitors.

For more background on how AI search engines choose their sources, read our complete guide to AI search optimisation and our guide to answer engine optimisation.

Using these figures

You may use the figures from this research in your own articles and presentations, with attribution: AIO Clicks (2026), AI Visibility of Dutch SMEs: Research Across 310 Websites, with a link to this page.

About this research

We analysed 310 business websites, scanned between 15 September and 20 October 2026 with the AIO Clicks audit tool: a technical scan, public data and AI-assisted content analysis. All results have been anonymised and are reported only in aggregate. This is a snapshot; AI answers and websites change constantly.

Sources

• Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., Narasimhan, K., & Deshpande, A. (2024). GEO: Generative engine optimization. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (pp. 5–16). ACM. https://doi.org/10.1145/3637528.3671900
• Aral, S., Li, H., & Zuo, R. (2026). The rise of AI search: Implications for information markets and human judgement at scale (arXiv:2602.13415). arxiv.org/abs/2602.13415
• Chen, M., Wang, X., Chen, K., & Koudas, N. (2026). Navigating the shift: A comparative analysis of web search and generative AI response generation. In Proceedings of the Workshops of the EDBT/ICDT 2026 Joint Conference. CEUR Workshop Proceedings. ceur-ws.org/Vol-4192/DARLIAP-paper4.pdf
• Gholami, S., Firullo, C., Cheyre, C., & Acquisti, A. (2026). Beyond search: LLM adoption and web traffic concentration [Working paper]. kgi.georgetown.edu
• Google Search Central. (2026). Optimizing your website for generative AI features on Google Search. developers.google.com
• Kargaev, D. (2026). The SEO-to-GEO gap: Quantifying ranking factor divergence between traditional and generative search (SSRN 6476021). papers.ssrn.com
• Li, H., & Aral, S. (2025). Human trust in AI search: A large-scale experiment (arXiv:2504.06435). arxiv.org/abs/2504.06435
• Luther, V., & Touboul-Cohen, O. (2026). Brand visibility in AI search: A longitudinal analysis of AI visibility metrics in the U.S. tea industry [Book chapter].
• Pew Research Center. (2025, July 22). Google users are less likely to click on links when an AI summary appears in the results. pewresearch.org
• Schulte, J., Bleeker, M., & Kaufmann, P. (2026). Don't measure once: Measuring visibility in AI search (GEO) (arXiv:2604.07585). arxiv.org/abs/2604.07585
• Zhang, K., He, X., & Yao, J. (2026). From citation selection to citation absorption: A measurement framework for generative engine optimization across AI search platforms (arXiv:2604.25707). arxiv.org/abs/2604.25707