Here is the most important number in AI visibility: 12%.
That is how often AI engines cite a brand's own website in answers to category questions, the "best", "which", "how do I" and "compare" questions where new customers find brands. When the question names the brand, the figure is 73%.
The data comes from the GEO Compass benchmark: 900,000+ answers from ChatGPT, Gemini, Perplexity, Claude, Copilot and Google's AI Overviews and AI Mode, across 11 industries and 24 markets. Each study was built around a focus brand, so we can see how that brand fares in questions that do and don't name it.
Named, but described by someone else
GEO Compass benchmark, all markets. "Named" is a share of all answers; "own site cited" is a share of answers that use sources.
In unbranded questions, the focus brand is named in 27% of answers, but its own website is cited in only 12% of the answers that list sources. So in many of the answers that recommend a brand, the engine is taking the description from somewhere else: a review, a news story, a comparison site, a forum thread or Wikipedia.
That has two consequences:
- You don't control most of your AI description. What third parties publish about you is what AI engines repeat, in category questions and increasingly in branded ones too. Even in branded answers, earned media supplies a quarter of citations, and Wikipedia appears in 8%.
- Your own site has to work harder for the answers it does appear in. In branded answers, your site is the default fact base. It needs to state prices, specifications, policies and differentiators plainly enough to be quoted.
The gap by question type
Not every category question is equal. The intent behind it changes how often the brand is named and cited.
| Question theme (unbranded) | Brand named | Own site cited |
|---|---|---|
| Product & service details | 37% | 28% |
| Local & places | 42% | 14% |
| Corporate & financial | 55% | 13% |
| Price & value | 33% | 13% |
| How-to & advice | 14% | 13% |
| Recommendations & "best of" | 31% | 10% |
| Brand trust & reputation | 38% | 10% |
| Health & medical | 10% | 8% |
| Comparisons & alternatives | 25% | 8% |
GEO Compass benchmark, unbranded questions. Own site cited is a share of answers with sources.
The lowest own-site citation rates are in the questions closest to a purchase decision: comparisons, recommendations and trust. Those are exactly where AI engines go to third-party reviewers and publishers. Product-detail questions are the exception: where a brand publishes clear specifications, engines quote them.
The gap by engine
| Engine | Brand named | Own site cited |
|---|---|---|
| Google AI Mode | 34% | 15% |
| Microsoft Copilot | 23% | 14% |
| ChatGPT | 27% | 13% |
| Google AI Overviews | 27% | 13% |
| Gemini | 35% | 10% |
| Claude | 36% | 9% |
| Perplexity | 18% | 8% |
Unbranded questions.
Gemini and Claude name brands often but cite their own sites rarely. Perplexity names brands least. See how each engine cites for more.
The gap by market
The gap is wider in several Asia-Pacific markets. Focus brands are named in more than half of unbranded answers in Indonesia (56%) and Malaysia (53%), and in 43% in India. But their own sites are cited in only 9–16% of those answers. AI engines in these markets know the brands; local news, aggregators, Wikipedia and travel or review platforms supply the description.
In Europe, own-site citation in unbranded answers is lower still, under 7%, with institutions, consumer bodies and national press doing more of the work.
What to do about it
Close the gap on your own site.
- Publish the facts AI engines look for in each question type: specifications, prices, comparisons, eligibility, locations, opening hours, policies.
- Answer the question in the first sentence of each section, under a heading phrased as the question.
- Add structured data (Organization, Product, FAQ, LocalBusiness as relevant) and keep names, addresses and descriptions consistent everywhere.
Win the sources that describe you.
- Find which third-party sites AI engines cite for your category in each market. The explorer shows the most-cited sites by industry, engine and market.
- Earn coverage, reviews and listings on those sites, and correct anything inaccurate.
- Maintain accurate Wikipedia and reference entries where they exist.
- Treat creators, YouTube and community platforms as part of the source base, not just as social channels.
Measure both numbers. Track how often you are named and how often your own site is cited, separately for branded and unbranded questions, by engine and market. The gap between the two is where your description is being written by someone else.
Compare branded and unbranded questions in your category on the Branded vs unbranded page.
Frequently asked questions
- What is the difference between branded and unbranded prompts?
- A branded prompt names a brand or one of its products, such as 'Is [brand] reliable?'. An unbranded prompt asks about a need or a category, such as 'best family SUV under $50,000'. Unbranded prompts are where new customers discover brands.
- Why doesn't ChatGPT cite my website?
- In category questions, AI engines mostly cite third-party sources such as reviews, news, comparison sites, forums and reference sites. In the GEO Compass benchmark, a brand's own site is cited in only 12% of unbranded answers that use sources, even though the brand is named in 27%.
- How can a brand get cited more in unbranded AI answers?
- Publish the specific facts and comparisons engines need on your own site in an extractable format, and earn coverage in the publications, review sites, creators and communities that AI engines cite for your category in each market.
Figures marked as GEO Compass data come from the anonymised GEO Compass benchmark of AI answers collected by We AI & Data Advisors. Related: AI search in Australia · AI search in Singapore · AI search in Malaysia · AI search in Indonesia · AI search in India · AI search in Europe