AI Search
Is my website visible to Gemini?
Gemini answers inside Google, inside the Gemini app, and inside Workspace, and it names companies while it does it. We measured which ones, and which websites it read first. Of the four assistants we tested, Gemini turned out to be the one where an ordinary business has the best chance of being named.
Key takeaways
- Gemini has the most room in it. Across 50 buyer questions it named 9.1 businesses per answer and cited 8.3 sources, more of both than any other assistant we tested.
- It is the least directory-dependent. About 13% of what it cited was a listing site, against ChatGPT's 26%, so it was roughly twice as likely to cite a business's own website.
- Being cited by Gemini says little about the rest. 62% of its sources were used by no other assistant on the same question.
- Blocking Google-Extended does not remove you from AI Overviews, and blocking Googlebot to "fix" that removes you from Google Search. This is the costliest mistake in this area.
What Gemini reads, measured
We measured this across 50 buying questions put to ChatGPT, Claude, Gemini and Perplexity in July 2026, ten questions in each of five industries. The methodology, including the prompts, the run dates and a run-to-run variance control, is published with the dataset rather than summarised here. Nothing below is modelled. Every figure is counted from the recorded answers.
| Assistant | Sources cited | Businesses named | Directory share |
|---|---|---|---|
| Gemini | 8.3 | 9.1 | 13% |
| ChatGPT | 4.9 | 8.9 | 26% |
| Claude | 6.4 | 5.9 | 18% |
| Perplexity | 7.8 | 4.5 | 18% |
Averages per answer over 50 questions in five industries, July 2026. Treat the directory figures loosely: they move depending on where you draw the line between a directory and a publisher. The gap between 13% and 26% is the point, not the decimal.
Why Gemini is the winnable one
Directory shareis the proportion of an assistant's cited sources that are listing or review sites rather than the website of a business it named. It matters because a directory entry is a page you do not control, and two numbers in the table decide the rest.
A worked example. Ask any of the four "best digital marketing agency in Mumbai" and you get a shortlist. In our run of that exact question the four assistants read 25 websites between them, named 24 businesses, and not one website was used by all four. About a third of what they read was a directory. Winning that question is not one job, it is four.
There are more slots. Gemini named 9.1 businesses per answer. Perplexity named 4.5. If an assistant only ever recommends four companies, you are competing for one of four places; at nine, the same market has twice the room. Gemini also gave longer answers than Claude or Perplexity, around 3,400 characters, and length is where extra names fit.
More of those slots point at real businesses.Gemini drew about 13% of its citations from directories. ChatGPT drew 26%. Put plainly, when ChatGPT went looking for who is good it reached for a listing site twice as often, and a listing site is a page you do not control. Gemini was the most likely of the four to read the business's own website and name it from there.
It also never declined. Across all 50 questions Gemini hedged zero times, where Claude declined to recommend anyone on three. An assistant that always answers is an assistant that always has a shortlist, and a shortlist is something you can be on.
Being read by Gemini tells you nothing about the others
62% of the sources Gemini cited were used by no other assistant on the same question. It found its own reading list far more often than it shared one.
That is the finding most likely to mislead somebody checking their visibility. Asking Gemini one question, seeing your domain in the sources, and concluding you are visible to AI is a conclusion the data does not support. On the same question, the odds are better than even that ChatGPT, Claude and Perplexity all read something else entirely. The full study lists every source each assistant used, question by question.
The Google-Extended mistake that costs you Google
This is the one worth reading twice, because acting on it wrongly does real damage. Two names get confused, and they control different things:
- Googlebot is the crawler that builds the Google Search index. Google AI Overviews are generated from that index, so Googlebot governs both your ordinary search listings and your AI Overviews eligibility. Google publishes the full list of its agents in its common crawlers documentation.
- Google-Extended is a robots.txt token rather than a crawler. It controls whether your content may be used to improve Gemini and Vertex AI, and it has no effect on Search.
- Groundingrefers to an assistant fetching live sources to support an answer instead of relying on training data alone. Gemini's is documented in the Gemini API grounding guide, and it is the path that lets a page published today be cited today.
In Google's own words, Google-Extended lets publishers "manage whether their sites help improve Bard and Vertex AI generative APIs" ( Google, September 2023; Bard is now Gemini). It governs Gemini app grounding and model training.
It does not govern Google Search, and it does not govern AI Overviews. AI Overviews are built from the Search index, which follows Googlebot. A site can disallow Google-Extended and still appear in AI Overviews every day.
Which makes the failure mode expensive. A scanner reports "Google-Extended: Disallow", someone reads it as "invisible to Google's AI", and goes looking for the Google crawler to unblock. Touch Googlebot and you do not fix AI Overviews. You leave Google Search.
Blocking Google-Extended is a legitimate choice, and plenty of publishers make it deliberately. Just make it knowing what it costs, which is Gemini app grounding, and what it does not, which is anything in Search. You can see exactly which agents your robots.txt currently admits with our free crawler checker, which keeps those two rows separate for this reason.
How to check whether Gemini names you
Do it the way a buyer would, not the way an owner would.
- Ask a buying question, not your brand name."Best [what you sell] in [your city]" is the question that decides a sale. "What is [your company]?" only tells you whether Gemini has heard of you.
- Read the sources, not just the answer. Gemini lists what it drew on. Your domain being absent while a directory that lists you is present is a specific, fixable problem, and a different one from being absent entirely.
- Ask five to ten questions, not one. Answers move between runs. One question answered once is an anecdote.
- Repeat on the other three.Given that 62% figure, Gemini's answer is close to no evidence about ChatGPT's.
What actually moves it
Nothing on this list is Gemini-specific, which is the useful part: the work that gets you named by Gemini is the work that gets you named anywhere.
- Let the crawlers in. Googlebot for Search and AI Overviews, Google-Extended for Gemini grounding. Decide each one deliberately.
- Publish structured data. Organization and Product or Service markup is how you are read as an entity rather than as prose to be interpreted. Worked examples are in our schema reference.
- Answer the question in the first sentence. A passage that stands alone is a passage that can be quoted.
- Keep dates in the markup. An undated page gives an assistant nothing to weigh on freshness.
- Get onto the sites it already reads.Even at 13%, directories were part of Gemini's reading. In your category, the study shows which ones.
Questions people ask
Does blocking Google-Extended remove me from Google AI Overviews?
No. Google-Extended governs whether your content can be used to improve Gemini and Vertex AI. AI Overviews inside Google Search are built from the regular Search index, which is crawled by Googlebot. A site can disallow Google-Extended and still appear in AI Overviews. Blocking Googlebot to resolve a Google-Extended finding would remove the site from Google Search entirely, which is almost never the intention.
How do I check if Gemini knows my website?
Ask Gemini the questions a buyer would ask before purchasing what you sell, not your brand name, and check whether your site appears among the sources it lists. Repeat across five to ten real buyer questions, because answers vary between runs. Checking one question once tells you very little.
Is Gemini easier to get cited by than ChatGPT?
On our measurements, yes, on two counts. Across 50 buyer questions Gemini named an average of 9.1 businesses per answer against ChatGPT's 8.9 while citing 8.3 sources against ChatGPT's 4.9, so there are more slots and more room. Gemini also drew about 13 percent of its citations from directories against ChatGPT's 26 percent, meaning it was roughly twice as likely to cite a business's own website.
If Gemini cites my site, will the other assistants?
Not reliably. In our study 62 percent of the sources Gemini cited were used by no other assistant on the same question. Being read by one assistant is weak evidence about the other three, which is why visibility has to be measured per engine rather than inferred from a single check.
Measure it rather than guess
Gemini is the most winnable of the four, which makes it the worst one to be invisible on. Greater Than Services puts your buyer questions to Gemini, ChatGPT, Claude, Perplexity and Google AI Overviews every week, records which answers named you and what was cited instead, and generates the code to close the gap.
The figures in this article are our own. The study sets out the questions, the assistants, the dates and the variance control, and the raw dataset is published under CC BY so you can recount them. If something here does not match what you see, tell us and we will correct it.