Where AI really cites from
In the largest German B2B measurement so far, only around 5 % of AI citations led to the company's own domain. The other 95 % came from third parties. That turns the usual approach on its head: the market optimises the website — and the website is the five percent.
The answer in one sentence
Anyone who wants to appear in AI answers has to be where citations come from — and that is mostly not your own website but map services and industry directories. For local businesses Google Maps is by far the strongest channel; for B2B industry it is wlw.de.
The measured shares
| Share | Channel | Where the figure comes from | What you can do |
|---|---|---|---|
| 38 % | Google Maps / Business Profile | NeuRank, 16.06.2026: 38 % of ChatGPT source references for German local queries, collected from 800 queries in 20 cities.Confidence: Agency study | Fill the profile completely: categories, services, opening hours, photos. |
| meistzitiert | wlw (Wer liefert was) | morefire, 20.07.2026: in the DACH B2B measurement with 62 industrial companies and 12,250 citations, the most cited domain overall. No share value was published.Confidence: Agency study | Create a company profile; catalogue providers have a product data feed. |
| ~5 % | your own website | The same DACH measurement: only around 5 % of citations led to the company's own domain.Confidence: Agency study | Ensure fetchability — blocked means invisible, regardless of content. |
| 0,6 % | Review portals | NeuRank, 16.06.2026: 0.6 % of German ChatGPT source references.Confidence: Agency study | Maintain existing profiles — but do not sell it as a visibility measure. |
All four values come from agency studies, not from peer-reviewed research. They are usable as an order of magnitude and not as a promise. And importantly: what was measured is the effect — no provider states that it draws data from these sources.
The difference from your own website
That does not mean your own website is irrelevant. It is the precondition: whoever may not be read cannot be cited — no matter how good the text is. That is exactly why we check fetchability first and content only afterwards.
It does mean, though, that the distribution of effort is usually wrong. Whoever puts the whole budget into articles and leaves the industry-directory profile incomplete is working on the five percent.
Three claims that are demonstrably false
“ChatGPT draws 60–70 % of its local results from Foursquare.”
Checked and refuted (Steady Demand, 21.08.2026). Anyone selling Foursquare maintenance as an AI measure relies on a figure that does not survive scrutiny.
“An llms.txt makes your website readable for AI.”
Ahrefs examined 137,210 domains: 97 % of the published files never received a single request.
“A self-created Wikidata entry improves recognition.”
Not a single measurement could be found for this — only agency blogs without a data basis.
What follows from this
First check whether the searching AI crawlers are allowed to read your website at all — that is a yes/no question with an answer in seconds. After that it is worth looking at the third-party channels that fit your business. Only then at the content.
As of 1 September 2026 · Every figure with source and survey date