How to check whether AI tools mention your brand, without a paid tool

Do AI tools mention your brand when people ask about your topic, and can you find out without paying for a tracking tool? You can get a rough, honest picture with a spreadsheet, a monthly session of asking questions, your server logs and two dashboards from Google and Microsoft. This guide shows the routine, and it says plainly what the routine can't tell you.

Why one answer proves little

If you ask an AI tool "who are the best [anything] providers?" once and your brand isn't in the answer, you've learned very little. Two pieces of research explain why.

SparkToro and Gumshoe.ai, January 2026. About 600 volunteers ran 12 prompts that asked for brand or product recommendations (for example headphones, chef's knives and digital marketing consultants) through ChatGPT, Claude and Google's AI features. That produced 2,961 prompt runs in November and December 2025, with each prompt run 60 to 100 times per platform. The authors found less than a 1 in 100 chance that ChatGPT or Google's AI would give the same list of brands if asked 100 times, and about a 1 in 1,000 chance of the same list in the same order. They also found that how often a brand appeared was steadier than where it was listed. The top brands appeared in 55% to 100% of answers, depending on the size of the category. Their own limits: the volunteers used their normal settings, the study isn't peer-reviewed, and the authors say they'd need far larger samples for firm conclusions.

Tow Center for Digital Journalism, March 2025. Researchers gave eight AI search tools 1,600 test queries in February 2025 (20 news publishers, 10 articles each, 8 tools). Each query contained a quote from an article, and the tool was asked to name the headline, publisher, date and URL. More than 60% of the answers were wrong. Perplexity was wrong 37% of the time and Grok-3 94% of the time. Over half of the Gemini and Grok-3 responses contained broken or made-up URLs. The test was about finding the source of a quote, which isn't the same as answering a buyer's question, and the tools have changed since. The lesson that carries over is that a link in an AI answer needs to be opened, because it may not say what the answer claims.

So the method below doesn't rely on any single answer. It relies on repeats, and on opening what the tool cites.

Step 1: Write ten prompts

Use questions a real buyer would ask. Don't write prompts that name your brand in every line, because that tells you only that the tool knows you exist. Mix the patterns below, replace the brackets with your own details, and keep the wording fixed from month to month.

1. What are the best [product or service category] for [type of buyer]?
2. Which [type of company] would you recommend in [city or country]?
3. How do I [problem your product solves], and which tools or services can help?
4. What should I look for when choosing a [service], and which providers are worth a look?
5. Give me a shortlist of [category] providers that work with [your niche or company size].
6. What are alternatives to [a competitor's name]?
7. [Competitor] or [your brand]: which is better for [use case]?
8. What is [your brand], and what is it known for?
9. What do customers say about [your brand]? Is it a good choice for [use case]?
10. The same question as number 1, written in the language of another market you serve.

Patterns 1 to 6 show whether you appear without being asked for by name. Patterns 7 to 9 check what the tools say when your name is in the question, including whether the facts are right. Pattern 10 matters if you work in more than one language, because answers differ by language.

Step 2: Set up a tracking sheet

You need two tabs.

Tab 1: every run. One row per answer.

(The row above is a made-up example of how to fill it in.)

Tab 2: the summary. One row per prompt and tool, with one number: in how many of your runs the brand was named. For example, "named in 1 of 3 runs".

Record whether you were logged in and whether web search was switched on, because both can change what a tool says. Which settings each tool uses by default changes over time, so note what you actually did rather than trusting a list.

Step 3: Run the prompts and read the answers

Open a new chat for every run. Run each prompt three times on a tool before you move on, ideally on different days. That gives you 30 answers per tool for ten prompts, which is a fair monthly session for two or three tools.

For each answer, check these things:

Reading one answer

• Named?Does the answer name your brand at all
• Right facts?Are what it says about you and your prices accurate
• Links open?Does every cited link load
• Links match?Does the page say what the answer claims
• Who else?Which competitors and sites are named
• Is it new?Same as last month, or different

Then fill in the summary tab. A simple way to read it, and this is our rule of thumb, not a standard:

• Named in none of your runs for a pattern: you're not showing up for that wording right now.
• Named in some runs: you're in the pool of possible answers, and it varies. This is the common middle result.
• Named in most runs: you're consistently present for that question.

Our view: the useful output is the share of runs where you appear and the list of sources the tools keep citing, not any single answer. If the same few sites show up again and again in answers about your topic, that's a list of places where your market is being written about.

Step 4: Check AI crawler visits in your server logs

Your server log records every request to your site, including requests from the AI companies' systems. It can show whether those systems visit you. It can't show whether they used what they found.

The names below come from each company's own documentation, which we opened on 6 October 2026. Don't rely on any list of user agents from a third-party site, because the names change.

A few notes on the table:

• Google's AI features use the normal crawler. Google's documentation says crawling preferences for Googlebot affect Google Search, including all Search features. So a Googlebot visit can't tell you whether the page was used for an AI summary.
• Google-Extended won't appear in your logs. Google says it has no separate user agent string; it's a robots.txt token that controls whether content may be used to train Gemini models.
• Microsoft Copilot: in the Microsoft pages we could open, we didn't find a separate user agent named for Copilot. Bing's crawler is bingbot, and Microsoft's citation data is covered in the next section.
• User-triggered visits are different. OpenAI says ChatGPT-User isn't used to crawl the web automatically, and that because these actions are initiated by a user, robots.txt rules may not apply. Perplexity says Perplexity-User generally ignores robots.txt rules for the same reason. A visit from one of these suggests a person's question led a tool to fetch your page. The log doesn't show the question.

How to pull the numbers from your logs

Ask your host, your developer or your content delivery network (CDN) for access logs. If a CDN serves your pages from its own cache, the log that matters may be the CDN's, not your server's. A Linux server with a standard log file can count visits by agent name with one command:

grep -o -E -i "OAI-SearchBot|GPTBot|ChatGPT-User|ClaudeBot|Claude-SearchBot|Claude-User|PerplexityBot|Perplexity-User|Google-Agent|bingbot" access.log | sort | uniq -c | sort -rn

That assumes your log records the user agent, as the common "combined" format does. If you can't run it, send the sentence above to whoever manages your hosting and ask for last month's counts per agent and the pages visited.

If you've published an llms.txt file, the same log will show whether anything requests it. What that file does, and whether it matters, is covered in our post on llms.txt.

Check that the visitor is who it says it is

Anyone can put GPTBot in a user agent string. The companies publish ways to check:

• OpenAI and Perplexity publish IP address lists for each bot, and Anthropic says that a crawler with a source address on its published list is coming from Anthropic.
• Google says to run a reverse DNS lookup and check that the name ends in googlebot.com, google.com or googleusercontent.com, or to match the address against its published IP range files.
• Bing's verification guidance (a 2012 post) says a reverse DNS lookup should resolve to a name ending in search.msn.com, and that Bing doesn't publish a fixed list of addresses. Check Bing's current help pages before you rely on that.

Make sure you aren't blocking them

If your robots.txt blocks a bot, it can't read your pages. OpenAI says each of its bot settings is independent, so you can allow OAI-SearchBot while disallowing GPTBot, and that a change to robots.txt can take about 24 hours to be picked up by its search system. Whether to allow training crawlers is a business decision, and it's separate from whether you want to appear in search answers.

What each source can and can't tell you

Manual prompts

• Whether you were named in your runs
• What the answer said about you
• Which sources were cited

Server logs

• Whether the named bots visit your pages
• Which pages they fetch
• How often

None of them show

• Why you were or weren't named
• How many people asked
• Whether anyone clicked or bought

What Search Console and Bing Webmaster Tools show

Google Search Console. Google's documentation says traffic from AI features such as AI Overviews and AI Mode is included in the Performance report under the Web search type. It also added a generative AI performance report. Per Google's help page, it shows how many times links to your site were shown to a user in a generative AI feature, by page, country, date and device, and Google says that as of August 31, 2026 it has rolled it out to all websites worldwide, though not every site may see it yet. The page describes impressions only. It doesn't list the questions people asked and it doesn't include clicks. Our guide to judging SEO when clicks fall explains how to use it.

Bing Webmaster Tools. Microsoft launched an AI Performance report in public preview on February 10, 2026. Microsoft's announcement says it covers Microsoft Copilot, AI-generated summaries in Bing and select partner integrations. It shows total citations (how often your pages were displayed as sources), the average number of unique pages cited per day, a sample of "grounding queries" (key phrases the AI used when retrieving your content) and citation counts by page. Microsoft says the metrics don't show placement within an answer or page importance, and the report doesn't measure clicks or traffic. Both reports come from the companies that run the AI features, which makes them better evidence than a third party's guess.

Your own analytics. Look at referral sources in your analytics for visits that arrive from AI tools. We couldn't open the vendors' own documentation on how they tag outgoing links, so look at what actually appears in your reports and don't assume a particular tag.

What not to conclude

• "We weren't named once, so AI ignores us." One answer is a coin flip. Use runs, not single results.
• "We were named once, so we're visible." Same reason.
• "A bot visited, so we were cited." A visit shows access. It doesn't show that anything was used in an answer.
• "These links or mentions made the difference." No study we've read shows that links or mentions cause an AI tool to name a brand. The Ahrefs study of 75,000 brands we describe in our post on the future of link building found a correlation between brand mentions and AI visibility, and it's a correlation. Our post on brand mentions versus backlinks goes deeper.
• "We can fix this with a trick." Google's page on AI features says there are no additional requirements to appear in AI Overviews or AI Mode, and no special optimizations necessary. Nobody can promise you a mention.
• "Our sheet is our market share." Your ten prompts are a small sample of the questions people ask.

What this method cannot tell you

Be upfront about this with anyone who reads your report.

• It can't tell you what a different person sees. Location, language, account history and settings can all change answers.
• It can't tell you why a tool named a brand or left it out.
• It can't tell you how many people asked, or whether any of them visited or bought.
• It samples the tools on days you choose, and tools change often.
• Paid trackers don't escape these limits. They run prompts for you at scale, but they face the same variation, and SparkToro's authors advise buyers to ask for stats-backed, publicly reviewable research before paying for one.

What to do

1. Write your ten prompts once and save them.
2. Build the two-tab sheet.
3. Run each prompt three times on two tools. Record every run.
4. Open every cited link and check the facts about your brand.
5. Ask for last month's AI crawler counts from your logs, and check robots.txt isn't blocking the bots you want to allow.
6. Check Bing Webmaster Tools for the AI Performance report and Search Console for the generative AI report.
7. Repeat monthly with the same wording, and look at the trend over a few months.
8. If the same sources keep appearing in answers about your topic, see whether you belong on any of them, and whether the editors there would care about what you know. Earned coverage and relevant placements, whether through guest posts or other services, are slow work with no guaranteed result. Google's spam policies treat links that are bought or sold to pass ranking signals as link spam unless the link is marked rel="sponsored" or rel="nofollow", and some placements carry a label. Our post on how AI assistants pick their sources explains what's known about that.

Sources

• SparkToro: AIs are highly inconsistent when recommending brands or products (January 28, 2026; about 600 volunteers, 12 prompts, 2,961 runs)
• Columbia Journalism Review, Tow Center: We compared eight AI search engines. They're all bad at citing news (March 2025; 1,600 queries, February 2025)
• OpenAI: Overview of OpenAI crawlers (OAI-SearchBot, GPTBot, ChatGPT-User; checked October 6, 2026)
• Anthropic: Does Anthropic crawl data from the web, and how can site owners block the crawler? (April 7, 2026)
• Perplexity: Perplexity crawlers (PerplexityBot and Perplexity-User; checked October 6, 2026)
• Google Search Central: Google's common crawlers (updated July 14, 2026)
• Google Search Central: Google's user-triggered fetchers (updated August 19, 2026)
• Google Search Central: Verifying Googlebot and other Google crawlers (updated March 20, 2026)
• Bing Webmaster Blog: How to verify that Bingbot is Bingbot (August 2012)
• Bing Webmaster Blog: Introducing AI Performance in Bing Webmaster Tools public preview (February 10, 2026)
• Google Search Central: AI features and your website (updated December 10, 2025)
• Google Search Console Help: Generative AI performance report (Search) (rollout to all websites as of August 31, 2026)
• Google Search Central: Spam policies for Google web search (updated August 28, 2026)

Checked on 6 October 2026.