Ansar Bhatti
Traditional SEO has mature tools: rank trackers, Search Console, analytics. AI search does not, at least not yet. Customers are asking ChatGPT, Gemini, Perplexity, Copilot and Google AI Overviews for recommendations, and most businesses have no idea whether they are named, described correctly or ignored. This guide shows how to track brand visibility in AI search with a method you can run today, using free tools and a simple spreadsheet, and how to read the results without fooling yourself.
The goal is not a perfect score. AI answers are variable by design. The goal is a consistent process that shows direction over time and reveals specific problems you can fix.
What “brand visibility in AI search” means
Visibility is not one number. It is a set of observations about how AI systems treat your brand when people ask relevant questions.
- Mention rate. How often your brand is named in answers to relevant prompts.
- Citation rate. How often your website is linked or cited as a source.
- Position and context. Whether you are recommended first, listed among many, or mentioned only as a comparison.
- Accuracy. Whether the facts, such as services, prices and location, are correct.
- Sentiment. Whether the description is positive, neutral or negative.
- Share of voice. How you compare with competitors for the same prompts.
Step 1: Build a prompt set
Your prompt set is the list of questions you will ask every month. Keep it fixed so results are comparable. Aim for 20 to 40 prompts across these groups:
| Prompt type | Example | What it tells you |
|---|---|---|
| Brand | “What is [brand] and what does it do?” | Whether you are known and described correctly |
| Category | “Best [service] companies for [audience]” | Whether you are recommended for the main topic |
| Problem | “How do I fix [problem your service solves]?” | Whether your content is used to answer questions |
| Comparison | “[Brand] vs [competitor]” | How you are positioned against rivals |
| Local | “[Service] near [city]” | Local recommendation presence |
| Buying | “How much does [service] cost?” | Whether your pricing information is found and accurate |
Write prompts the way customers speak, not the way marketers write. Use search data, sales calls and support questions for ideas. Our keyword research guide shows how to find real questions.
Step 2: Choose which assistants to test
Start with the ones your customers actually use. For most businesses that means ChatGPT, Google AI Overviews and AI Mode, Perplexity and Gemini, and possibly Microsoft Copilot. Record which product and which mode (with or without web search) you used, since results differ.
Step 3: Run the tests carefully
AI answers change between runs, between accounts and between locations. A few habits reduce the noise:
- Use a clean session or a logged-out window where possible, so personal history does not skew answers.
- Run each prompt two or three times and record the overall pattern, not a single result.
- Test on the same day each month and note the date.
- For local prompts, set the location you actually serve.
- Save screenshots or copy the text of important answers as evidence.
Step 4: Log results in a simple sheet
A spreadsheet is enough. Create one row per prompt per assistant per date, with these columns:
- Date, assistant and mode
- Prompt and prompt group
- Brand mentioned (yes or no)
- Your website cited (yes or no)
- Position or role (recommended, listed, compared, absent)
- Competitors named
- Accuracy notes and any wrong facts
- Sources cited in the answer
From this you can calculate mention rate (answers that name you divided by answers tested) and compare it month to month. Treat small changes as noise and focus on sustained moves.
Step 5: Use analytics and Search Console too
Manual testing shows what AI says. Your analytics show what people do next.
- Referral traffic. In Google Analytics, look at traffic sources for visits that come from AI platforms. Names and referrer formats change over time, so check your source and referral reports periodically for new entries.
- Branded search. Rising searches for your brand name in Search Console often follow growing AI and media exposure.
- Landing pages. See which pages receive AI referrals and whether those visitors convert.
- Server logs. Logs show visits from AI crawlers such as OAI-SearchBot, ChatGPT-User, PerplexityBot and others, which tells you whether you are being fetched at all.
Remember that Google counts AI Overview and AI Mode traffic within overall Search data, so you cannot isolate it perfectly. Treat Search Console as a directional signal.
Step 6: Consider dedicated tracking tools
A growing number of paid platforms automate prompt tracking across assistants, store history and benchmark competitors. They can save time if you have many prompts or locations. Before paying, check how they collect data (real interface runs or API calls), how often they refresh, how they handle location, and whether you can export your data. Many small businesses do fine with a manual sheet for the first few months.
Step 7: Turn findings into fixes
Tracking is only useful if it changes what you do. Use the patterns you see:
| What you find | Likely cause | What to do |
|---|---|---|
| Not mentioned at all | Weak footprint or unclear category | Clarify what you do, build third-party mentions, add answer-first pages |
| Wrong facts | Old or inconsistent information online | Update your site, profiles and directories to match |
| Competitors cited, not you | They have clearer, better-supported pages | Study their cited pages and create something more useful |
| Mentioned but never linked | Model memory without strong retrievable pages | Publish quotable, well-structured pages for those topics |
| No AI crawler visits | Crawlers blocked or site hard to fetch | Review robots.txt, firewall rules and site speed |
If you need the background on how to improve these signals, read our GEO guide and our practical article on getting mentioned in ChatGPT.
Common tracking mistakes
- Changing the prompt set every month, which makes trends meaningless.
- Judging from one answer instead of repeated runs.
- Counting mentions without checking accuracy and sentiment.
- Ignoring conversions. Visibility only matters if it brings the right customers.
- Expecting instant movement. Improvements usually take months.
Frequently asked questions
Is there a free way to track AI visibility?
Yes. A fixed prompt list, a spreadsheet and monthly manual tests cost nothing but time. Combine them with Search Console and analytics for context.
How often should I check?
Monthly is enough for most businesses. Check more often only after a major change or if you are in a fast-moving market.
Why do I get different answers each time?
AI systems generate answers probabilistically and may use live search, personalization and location. That is why you repeat prompts and look at patterns.
Can I see how many people saw my brand in AI answers?
Not directly. No assistant provides complete impression data to site owners. You estimate through sampling, referral traffic and branded search trends.
What is a good mention rate?
There is no universal benchmark. Compare against your direct competitors for the same prompts and track your own trend upward.
Also read: checking whether AI bots crawl your site.
Also read: AI visibility audit.
Also read: llms.txt explained.
Also read: how to get cited in Google Gemini.
For dental practices, see GEO for dentists.
Want an AI visibility baseline for your business?
We run this process as part of our GEO services: a prompt set built around your customers, a competitor benchmark and a prioritized list of fixes. Contact us with your website address and we will tell you where you stand.
