Does the AI mention you?

Our earlier guides cover how to get cited by AI search and a checklist for your website. Both are about what to build. This one is about what to measure first, because you cannot judge any of that work without a starting point.

The test is simple. Ask the assistants what your customers would ask, and write down what comes back. A dental clinic in Columbus, Ohio might ask "Which dentists near Columbus take new patients on weekends?" A Manchester accountant might ask "Who can help a small online shop with VAT?" What the assistants say is what a growing share of your customers see.

Two honest cautions before you start. First, nobody can guarantee placement in these answers. Google's own page says there are no additional requirements to appear in AI Overviews or AI Mode beyond being indexed and eligible to show a snippet, and it promises nothing more. Second, the measurement is noisy. Microsoft's Azure OpenAI documentation puts it plainly: if you ask a model the same question several times, you are likely to get a different response. Treat each answer as one sample.

Write them the way customers talk, not the way your website does. Use five kinds of question, four of each. Here is a template, with the Ohio dental clinic as the example.

• Find me one (4 questions)"Best family dentist in Columbus, Ohio." "Dentist near me open on Saturday." Include your area and your main service.
• Compare (4 questions)"Dentist A or Dentist B for braces?" Use your real competitors by name, and also the generic "X vs Y" questions your customers ask.
• Price and process (4 questions)"How much does a dental implant cost in Ohio?" "How long does a root canal take?" Assistants often name providers when they answer these.
• Trust and problems (4 questions)"Is [your business] any good?" "Are there complaints about [your business]?" "How do I choose a dentist?"
• Your name (4 questions)"What does [your business] do?" "Where is [your business] and what are its hours?" This is where wrong details show up fastest.

Get the wording from real life: your search queries, the questions your front desk answers every week, the phrases in your reviews. Keep the list fixed, so you can compare month to month.

Use five places where customers now get answers. Each has a slightly different way of working, so keep notes of what you used.

• ChatGPT. Start a new chat for each question with search on, so it can use the web. A fresh chat avoids earlier conversation shaping the answer.
• Gemini. The same method. Google documents that its Gemini apps can use content from Google Search to ground answers.
• Perplexity. It shows its sources by default, which makes the citation score easy.
• Google AI Overviews and AI Mode. Search the question on Google in a private window. An AI Overview does not appear for every search, so write "none shown" when it does not. Try AI Mode on the same question. Google says both can run multiple related searches across subtopics to build a response, so they may link to a wider set of pages than a classic search.
• Microsoft Copilot. Ask in a new conversation. Microsoft's Bing Webmaster Tools also has an AI Performance report, in public preview since February 2026, that counts how often your pages are cited as sources in Copilot and Bing's AI summaries. That is a rare case of a vendor giving you your own numbers.

Twenty questions across five assistants is 100 answers. That is a long afternoon, which is why many people start with the "find me one" and "your name" groups.

For every answer, ask four things. Give one point for each yes, so each answer scores 0 to 4.

• MentionIs your business named at all? This is the first gate. If not, nothing else matters for that question.
• PositionAre you among the first three named? Assistants usually list a handful of options, and people rarely go further.
• AccuracyAre the name, address, hours, prices and services right? A wrong phone number or an old price counts as a no, and it is the most useful thing to fix.
• CitationDoes the answer link to your own website? A mention without a link still builds awareness, but a link sends visitors.

Add the points per assistant and per question group. You will quickly see a pattern: perhaps you are named for your branded questions but absent from "find me one", or Perplexity cites you and Gemini does not. That tells you where to look.

How often to repeat

Because answers vary, one run proves little. A practical rhythm:

1. First roundAsk each question three times, on three different days. Count a mention only if you see it in at least two of three, and note when you appear only once.
2. After a changeRepeat after you fix a robots.txt rule or publish an important page. OpenAI and Perplexity say crawler changes can take about 24 hours to apply, and content changes can take much longer. No company publishes a timeline.
3. MonthlyRun the full list again and compare totals with last month.

Start with whatever your scorecard shows. Here is the order we would work in. None of it guarantees a result.

1. Crawler access

If you appear nowhere, make sure the bots that feed search answers are not blocked in your robots.txt (a plain text file at the root of your site that tells crawlers what they may read). The companies separate bots by job, so blocking one does not block the others:

• OpenAI. OAI-SearchBot is for search and surfaces sites in ChatGPT's search features. GPTBot is for training. ChatGPT-User visits pages when a person asks, and OpenAI says robots.txt rules may not apply to it. Opting out of OAI-SearchBot keeps you out of ChatGPT search answers, though you can still appear as a navigational link.
• Perplexity. PerplexityBot surfaces and links sites in Perplexity results and is not used for training. Perplexity-User fetches pages when a person asks and generally ignores robots.txt.
• Anthropic. ClaudeBot collects content that could contribute to training. Claude-SearchBot improves the quality of search results, and Claude-User fetches pages when a Claude user asks. Anthropic says disabling the last two may reduce your visibility.
• Google. No special bot for AI Overviews or AI Mode: the page must be indexed and eligible to show a snippet. Google-Extended is a robots.txt token that governs use for Gemini training and for grounding in Gemini apps. It does not affect inclusion in Google Search.
• Microsoft. Bing says it respects robots.txt preferences. Its AI Performance report is where you check Copilot citations.

Whether to allow training bots is a separate business decision. OpenAI and Anthropic each list their training and search bots separately, so you can treat them differently. Also check your hosting or CDN is not blocking crawlers by default. Our website checklist goes through this in more detail.

2. Answers on your own site

If you appear but with wrong details, fix the source. Put your address, hours, prices and services as plain text on the page, not only in images or PDFs. Google says no special markup is needed for AI features, but structured data is fine as long as it matches what visitors can see.

3. Consistent details everywhere

Bing's guidance says to keep text, images and video consistent about the same products or concepts, and local businesses can keep address, hours and contact details current through Bing Places for Business. Do the same on your Google Business Profile, directories and social pages. Mismatched details across the web are a likely source of wrong answers, though no vendor has said how much weight any one source gets.

4. Mentions elsewhere

Assistants that search the web draw on more than your site. Reviews, local directories, industry roundups and partner pages are places where a customer's answer may come from. This is a sensible bet rather than a documented formula, so treat it as that.

5. llms.txt

llmstxt.org describes it as a proposed Markdown file that gives AI agents a short guide to your site. It remains a proposal, and Google's guidance says Search does not need new machine-readable files. It is cheap to add and unproven. Do it last, if at all.

The manual test works well for one business and a few dozen questions. It gets tedious when you track competitors, run questions repeatedly or want a trend line. That is the problem Statbricks is meant to solve. It is our tool in development: it chooses the questions your customers ask in your market, samples the assistants' answers repeatedly so scores are steadier than a single reading, and lists the content changes to make first. It is not live yet and the waitlist is open.

Until then, the spreadsheet above does the same job by hand. Run it once this week, fix the single biggest gap, and run it again next month.