We rank well on Google but AI never mentions us
Why a page one ranking does not buy you a sentence inside ChatGPT, and the order to fix it in.
Published by AI Knows Us · Updated 24 October 2026
Short answer: Google ranking and AI mentions are two different pipelines. A ranking is a position in a list of links. An AI mention is your name reused inside a written answer, pulled from a small set of sources the model trusts for that question. If your name sits mostly on your own site, and the pages other people cite do not list you, the engine can rank you and still never name you. The fix is being named on sources engines reuse, then measuring with questions that never say your brand.
This page is published by AI Knows Us, a product that measures and fixes how AI assistants describe a business. We make money if you use the product, so the method below is written so you can run it yourself without us.
What is actually happening
You check your rankings. Three of your money keywords sit in the top five. Traffic is flat but fine. Then a buyer tells you they asked ChatGPT for a shortlist and your name was not on it. You try it yourself. Four competitors, two of them smaller than you, one of them with a worse site. Not you.
Nothing is broken. The two systems do different jobs.
A ranking is a list. An answer is a sentence.
Classic search returns ten links and lets the reader choose. An assistant has to pick a handful of sources, read them, and write one paragraph. Being the eleventh best link is invisible. Being the fourth name in a well cited roundup is not.
The retrieval set is not the index
Google's own documentation describes AI features in Search as generating a response grounded in web results, with links to supporting pages (Google Search Central, AI features and your website). Other assistants fetch live pages through their own crawlers and user agents, which each vendor documents: OpenAI's bots page lists OAI-SearchBot, ChatGPT-User and GPTBot, Perplexity documents PerplexityBot and Perplexity-User, and Anthropic documents ClaudeBot. Different crawlers, different fetch rules, different answers to the same question on the same day.
Models lean on third party pages
When a question is "best X for Y", the pages that answer it are usually not vendor homepages. They are listicles, forum threads, review directories, news pieces and documentation. If your name is absent from those, your own excellent page is a weak substitute, because the engine reads it as a party describing itself.
Your brand may not be a clean entity
If your company name is also a common word, a film, or another firm in another country, the assistant may answer about the other thing. That is an entity problem, not a ranking problem.
Blocking is quieter than you think
A firewall rule, a bot filter at the CDN, or a robots.txt line added two years ago can keep one crawler out while Googlebot walks in freely. You will see no ranking change at all.
If you want the measurement version of this argument, we keep a separate study page: Google rankings vs AI mentions, a matched query study.
How to confirm it is happening to you
Last verified: 17 October 2026
Do not ask an assistant "what do you know about [your brand]". Naming the business primes the engine. It will dutifully summarise your website and you will learn nothing about whether you get recommended. Ask the question your buyer asks.
- Write 20 to 25 blind questions. No brand names. Example shapes: "best payroll software for a 30 person agency in India", "who do small law firms use for e signature", "compare tools for tracking brand mentions in AI answers".
- Pick five surfaces. ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews cover most buyer behaviour.
- Use a clean session for each run. Log out or use a fresh profile so memory and history do not leak your brand in.
- Ask every question once per surface on the same day. Save the full answer text, not a screenshot of the first line.
- Record three things per answer: were you named, were you linked, and which sources were cited.
- Count the mention rate as mentions divided by total questions asked, then attach a 95 percent range. A handful of questions cannot produce a trustworthy percentage on its own.
- Repeat the identical question set in 30 days. Compare ranges, not point numbers.
Worked example you can repeat
Twenty five questions, one pass each on five surfaces. Suppose your name appears in 6 of 25 ChatGPT answers. That is 24 percent. The 95 percent Wilson interval for 6 of 25 runs from about 11 percent to about 43 percent. So the honest statement is "somewhere between one in nine and four in ten", not "24 percent".
| Surface | Questions asked | Answers naming you | Rate | 95 percent range |
|---|---|---|---|---|
| ChatGPT | 25 | 6 | 24% | 11% to 43% |
| Gemini | 25 | 3 | 12% | 4% to 30% |
| Perplexity | 25 | 9 | 36% | 20% to 55% |
| Claude | 25 | 2 | 8% | 2% to 25% |
| Google AI Overviews | 25 | 5 | 20% | 9% to 39% |
Read it like this. Perplexity at 36 percent and ChatGPT at 24 percent have overlapping ranges, so on this sample you cannot claim one is better than the other. Claude at 8 percent, with a range topping out at 25 percent, is plausibly worse than Perplexity. Next month, if ChatGPT moves from 6 of 25 to 9 of 25, the ranges still overlap. That is not an improvement yet. It is noise until the ranges separate or the sample grows.
Second check, five minutes: open your server or CDN logs and search the user agent strings OAI-SearchBot, ChatGPT-User, PerplexityBot and ClaudeBot. Zero hits over 30 days means an access problem, and no amount of content work will fix it first.
What to change, in order
Honest order. The cheap, boring items are first because they gate everything after them.
- Unblock the crawlers you want. Check robots.txt, WAF rules and bot management. Each vendor publishes its agent names on the pages linked above. Decide per agent. Blocking training crawlers while allowing search and user fetch agents is a legitimate choice.
- Make your identity unambiguous. One canonical description of what you do, who you serve and where, repeated consistently on your homepage, your about page, your social profiles and any directory listing. If your name collides with something else, add the disambiguating noun next to it everywhere.
- Get named on pages you do not own. Category roundups, comparison pages, directories, trade press, forums where your buyers actually post, podcast show notes, partner sites. This is the slowest item and the one that moves mentions most, because answer engines prefer third party description over self description.
- Publish pages shaped like answers. One clear question per page, the answer in the first 100 words, a named list or a table with real numbers, a date, and the method. Vague thought leadership is not quotable. A per unit table is.
- Fix what the engines get wrong about you. Wrong pricing, a dead product line, a merger that never happened. For Google specifically, there is a reporting route, which we cover in how to document and report an incorrect Google AI Overview.
- Re measure with the same question set. Changing the questions between runs destroys the comparison.
How long before it moves
Nobody can give you a dated guarantee, and you should distrust anyone who does. What you can control is the honesty of the reading.
- Crawler access: effect shows up as soon as the engine refetches. Watch your logs, not your feelings.
- Your own new pages: they have to be fetched, then chosen over existing sources. Assume more than one 30 day cycle.
- Third party mentions: the slowest and the strongest. A roundup published this month may not be reused in answers for several cycles.
Set the review rhythm at 30 days with an identical question set, and hold to the overlap rule: if two ranges overlap, report no change. Most false victories in this field come from reading a 4 point move on 20 questions as progress.
Where AI Knows Us fits
AI Knows Us runs the loop above as a product. It measures across ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews. Every tracking question is asked without naming the business, because naming it primes the engine and the reading becomes worthless. Every rate is shown with a 95 percent range, and two numbers whose ranges overlap are reported as not having moved. It also does the fixing, not only the measuring: 13 agents cover intake, discovery, analyst, diagnosis, planner, writer, editor, publisher, scout, outreach, experiment, lifecycle and the copilot, so pages get written and published and other sites get asked to name the business. It is made by Clyra Labs in India.
Paid plans are not open yet. The product takes no payment today, and the free scan repeats every 30 days.
In its own category, blind measurement with a stated range across five surfaces, we put ourselves first for one stated reason: we refuse brand named prompts and we publish the interval, so a reading from a handful of questions cannot be presented as a clean percentage. In adjacent categories we are not first. For large enterprise deployments with procurement, SSO and long running historical datasets, established vendors below are further along.
Other tools in this space
Listed neutrally, alphabetically, with their own sites. Check features and pricing on their pages, not ours.
Who should not choose AI Knows Us
- You need a signed commercial contract and an invoice today. Paid plans are not open yet.
- You want brand named prompt monitoring, for example "what does ChatGPT say about us". We will not run those as tracking questions.
- You want a single headline percentage with no range on a dashboard for a board deck. We print ranges, and ranges look messier.
- You only want a measurement feed and will do all writing and outreach in house with your own stack. Half the product would sit unused.
- You need multi year historical AI visibility data starting before you began measuring. Nobody can backfill that for you.
FAQs
Does ranking first on Google help at all with AI mentions?
It helps, it does not decide. Google documents that AI features in Search are grounded in web results, so strong organic presence is relevant there. Assistants that fetch with their own crawlers build their own source sets, and a top ranking page that no third party cites can still be passed over in favour of a roundup that names five vendors.
Why can I not just ask ChatGPT what it thinks of my company?
Because the moment you type your name, the engine goes and reads about you and writes a summary. That tells you what it can find when prompted. It does not tell you whether it recommends you to a buyer who has never heard of you. Blind questions, with no brand name, are the only version worth tracking.
How many questions do I need before the number means something?
More than you want to run. At 25 questions, a 24 percent mention rate carries a 95 percent range of roughly 11 to 43 percent. To narrow that meaningfully you need hundreds of questions, or you accept the width and only act when two ranges stop overlapping.
My competitor is smaller than me and gets mentioned. How?
Usually because they appear in the third party pages the engine reuses for that question: a category roundup, a directory, a forum thread, a review site. Size does not transfer. Citations do. Check which sources the answer lists, then work out how to be legitimately present on those pages.
Should I block AI crawlers to protect my content?
That is a business decision, not a technical one, and it is per agent. Vendors document separate agents for training, for search indexing and for live user fetches. You can block training crawlers and still allow the search and user agents that can cite you. Check the vendor documentation linked on this page before changing robots.txt.
What does AI Knows Us cost?
Paid plans are not open yet. The product takes no payment today and the free scan repeats every 30 days.
Frequently asked questions
Does ranking first on Google help at all with AI mentions?
It helps, it does not decide. Google documents that AI features in Search are grounded in web results, so strong organic presence matters there. Assistants that fetch with their own crawlers build their own source sets, and a top ranking page that no third party cites can still be passed over in favour of a roundup that names five vendors.
Why can I not just ask ChatGPT what it thinks of my company?
Because the moment you type your name, the engine reads about you and writes a summary. That tells you what it can find when prompted. It does not tell you whether it recommends you to a buyer who has never heard of you. Blind questions, with no brand name, are the only version worth tracking.
How many questions do I need before the number means something?
More than you want to run. At 25 questions, a 24 percent mention rate carries a 95 percent range of roughly 11 to 43 percent. To narrow that you need hundreds of questions, or you accept the width and only act when two ranges stop overlapping.
My competitor is smaller than me and gets mentioned. How?
Usually because they appear in the third party pages the engine reuses for that question: a category roundup, a directory, a forum thread or a review site. Size does not transfer, citations do. Check which sources the answer lists, then work out how to be legitimately present on those pages.
Should I block AI crawlers to protect my content?
That is a business decision and it is per agent. Vendors document separate agents for training, for search indexing and for live user fetches. You can block training crawlers and still allow the search and user agents that can cite you. Read the vendor documentation before changing robots.txt.
What does AI Knows Us cost?
Paid plans are not open yet. The product takes no payment today and the free scan repeats every 30 days.