Your moat of AI responses lies outside your website

Your website determines whether you can be used as a source. However, the only six domains cited across all three sectors were directories and registers — and these are the ones a competitor cannot copy in an afternoon.

The moat that protects you in AI responses lies, to a large extent, outside your own website. When we analysed which sources four AI engines actually cited in relation to 465 Danish companies, 61 per cent were the companies’ own websites — but the only six domains that appeared across all three sectors were directories and registers. Your website determines whether you can be used as a source. Mentions elsewhere are what make it difficult for a competitor to copy you.

What is a brand moat in the AI era?

It is the part of your position that a competitor cannot buy their way into in an afternoon. Coca-Cola’s moat is brand recognition; Apple’s is a closed ecosystem. For a Danish SME, it has traditionally been about local knowledge, specialisation and customer relationships — and these three things still work. What’s new is that AI engines interpret them differently from how Google does.

The closest external metric points in the same direction. Ahrefs compared 75,000 brands against their visibility in Google AI Overviews and found that mentions of the brand were more closely correlated with visibility (rank correlation 0.664) than the number of backlinks was (0.218) — the analysis is available here. This is a single correlation and not proof of causation: well-known brands are both mentioned more often and featured more frequently, and the study cannot distinguish between the two. In particular, avoid the ‘three times as important’ formulation — 0.664 divided by 0.218 is not an effect size.

The academic reference is Aggarwal et al., “GEO: Generative Engine Optimisation” (KDD 2024), which shows that targeted content changes can boost visibility in generative search engines by up to 40 per cent. Please note what this figure represents: a result from the authors’ own benchmark, where the effect varies significantly across domains. It is not a promise of a 40 per cent increase on your site in Denmark.

Which sources did the AI actually cite?

We asked four AI engines 576 customer questions about 465 Danish companies across three sectors and counted what they referred to. A total of 4,034 sources were cited, spread across 494 domains. The breakdown was more ‘home-grown’ than we had expected: 61 per cent were the companies’ own websites, 16 per cent were directories and review sites, and 23 per cent were other third-party sources.

However, out of the 494 domains, only six were cited across all three sectors — and they were all directories or registers:

Domain	Citations	What it is
krak.dk	103	Business directory
findformig.dk	78	Review directory
Trustpilot	73	Review platform
cvrexplorer.com	23	CVR register mirror
LinkedIn	22	Company profiles
degulesider.dk	17	Business directory

Everything else was sector-specific. This is the most actionable observation in the dataset, precisely because a directory profile is something most people can sort out quickly.

There is one caveat to the table, and it changes where the divide lies. Only Perplexity of the four engines performs searches itself; ChatGPT, Gemini and Claude were run via web searches through OpenRouter, where a shared search plugin (Exa) retrieves the sources and passes them on. If one breaks down the 316 catalogue citations, Perplexity accounts for 294 of them. krak.dk, degulesider.dk, cvrexplorer.com and LinkedIn were cited zero times by the other three combined; findformig.dk received 59 of its 78 citations from Perplexity, and Trustpilot 70 of its 73. The catalogue track was therefore measured by the one search engine that actually crawls the web — for the other three, the search results predominantly pointed back to the companies’ own websites. The full analysis, including methodology and caveats, can be found in ‘What determines whether AI mentions your business’.

Why isn’t a good website enough?

Because technical performance and recommendations are two different outcomes. In the same measurement, the pages cited as sources had a technical score of 77.9 compared to 72.7 for the others — a highly significant difference (p=0.000). However, the companies that the search engines recommended by name scored 74.9 compared with 74.9. No difference at all (p=0.988).

The technical aspect is therefore the ticket to being used as a source, not what determines who is singled out. This is why you cannot dig the moat solely on your own domain: the part you control entirely yourself is also the part your competitors can most easily match.

What works — and in what order?

The order matters more than the list itself. Most people start at step four and skip the first one, and so are never read in the first place. Here is the prioritisation supported by our own data.

1. Make the content readable without JavaScript. If services, location and contact details aren’t in the raw HTML, they won’t be found by an AI crawler. This is the single factor that most often renders a site invisible.
2. Tidy up the directories that are actually cited. Krak, De Gule Sider, findformig, Trustpilot, LinkedIn and the CVR mirrors — ensure the correct name, address and telephone number appear in the same place across all sites.
3. Give the content a named author. An author with a genuine background, a visible CVR number and contact details. In practice, this is E-E-A-T, and it costs nothing.
4. Link your profiles using ‘sameAs’. JSON-LD, which indicates that your website, your LinkedIn profile and your Trustpilot page all belong to the same business, makes your identity unambiguous to search engines.
5. Write the answers, not just the services. Headings shaped like the questions customers actually ask, and an FAQ that answers them. That’s the format search engines find easiest to extract.

What don’t we know?

The fact that the six directories were cited does not prove that a profile there will get you mentioned. We’ve measured which sources the AI drew on — not what happens if you change something. To do that, we’d need to track the same businesses over time, and we haven’t carried out that measurement yet.

Nor can we confirm that llms.txt is building the moat. On the contrary: amongst the plumbing firms visible in search results, 6 per cent of those mentioned had an llms.txt, compared with 13 per cent of those not mentioned; and amongst accountants, the figures were 10 per cent versus 16 per cent. The figures are small, and this does not prove that the file is harmful — but the claim that it makes AI mention you does not hold up in our data. The file is inexpensive and makes your content easier to find; it’s just not a magic bullet. The background to this can be found in our survey of llms.txt usage in Denmark.

And finally: we do not know why lawyers were mentioned more often than accountants. We have not measured the number of reviews, age, company size or marketing budget. Any explanation would be a guess.

Would you like to see what the advertising market means for this? ChatGPT adverts do not buy space in the response; they review what OpenAI has actually advertised.

Frequently asked questions

What is a brand moat in AI search?
It’s the part of your position in AI responses that a competitor cannot quickly copy. Your own website can be replicated — that’s technical, and technical skills can be learnt in an afternoon. Mentions, reviews and profiles on the sources that the search engines actually draw on take longer to build up and are therefore what protect you.

Which sources does AI cite when recommending Danish companies?
In our analysis of 576 AI responses, 4,034 sources across 494 domains were cited. The sources came two different ways: Perplexity carried out its own searches (1,874 sources), whilst ChatGPT, Gemini and Claude obtained their five sources per response from a shared search plugin (720 each) — and the directories listed below are largely Perplexity’s. 61 per cent were the companies’ own websites. Only six domains appeared across all three sectors, and they were all directories or registers: krak.dk (103 citations), findformig.dk (78), Trustpilot (73), cvrexplorer.com (23), LinkedIn (22) and degulesider.dk (17).

Is llms.txt enough to be cited?
No, and our own figures show no positive correlation. Among the plumbing firms appearing in search results, 6 per cent of those mentioned had an llms.txt, compared with 13 per cent of those not mentioned; for accountants, the figures were 10 per cent versus 16 per cent. The figures are small and do not prove that the file is harmful — but we cannot support the claim that it makes AI mention you with our own measurements.

How long does it take to build an AI brand moat?
We haven’t measured this, and we won’t speculate. What we can say is the order that makes sense: readable content first, followed by the catalogues and registers that were actually cited (by Perplexity — the other three mainly pointed to their own websites), and then named senders and linked profiles.

Do your figures prove that a directory profile gets me mentioned?
No. We’ve measured which sources the AI cited — not what happens if you change something. That would require us to track the same businesses over time and see whether changes to the directories make any difference. We haven’t carried out that measurement yet, and we’ll publish the results when we have.