How to Track Brand Mentions in AI Search (Free to Enterprise, 2026)

Tracking brand mentions in AI search means running your real buyer questions against ChatGPT, Claude, Gemini and Perplexity on a repeatable schedule, then recording whether you're mentioned, where you rank against competitors, and which sources the answer actually cites. A single spot-check tells you almost nothing, because the same prompt can return a different answer an hour later. What follows is a method that scales from a free five-minute check to enterprise tooling, in that order.

Why does checking ChatGPT once tell you almost nothing?

Because the answer you get depends on more than your brand's actual visibility. It depends on phrasing, timing, your account history and sometimes your location, and AI engines are not shy about proving that.

A Washington State University study found that across ten identical prompts, ChatGPT rated the same statements as accurate only 73% of the time, flipping between true and false on repeat runs of the exact same question. If a model can't stay consistent on its own answers, it won't stay consistent on whether it mentions your brand either.

Profound's research goes further on why. Analysing real query behaviour, it found that 91% of the search queries ChatGPT generates behind the scenes are unique even when the user's visible prompt barely changes, and those generated queries share only around 13% of their wording with what the user actually typed. ChatGPT isn't rerunning your question. It's generating new retrieval paths each time, chasing the same intent through different vocabulary. Perplexity behaves differently again, repeating its own search queries at a much higher rate, which is one reason a tool built for one engine doesn't always transfer cleanly to another.

Then there's context. Personalisation and memory mean two people asking ChatGPT the identical question can get different answers on the same day: conversation history changes what the model references, location shifts region-specific recommendations, and account type (free, Plus, a custom GPT) changes the underlying system prompt. Logged-out and incognito checks strip most of that away, which is exactly why they're the more honest starting point, but even then, phrasing and the moment you ask still move the answer.

None of this makes tracking pointless. It makes single checks misleading and repeated, structured sampling the only method that produces a number you can trust.

How do you do a manual spot-check properly?

Ask three to five phrasings of the same buyer question, in an incognito or logged-out browser window, across at least three engines, and read what's cited alongside any mention.

Vary the phrasing deliberately: a short version ("best CRM for small teams"), a longer natural-language version a real buyer would actually type, and a comparison version ("X vs Y for small teams"). Run each across ChatGPT, Gemini and Perplexity at minimum, since coverage and citation habits differ enough between engines that one platform's answer doesn't predict another's.

Check the market setting too. Region and language change the answer, sometimes sharply, so if you sell in the UK and the US, run the same prompt set in both rather than assuming a US-based check represents your UK buyers.

Use incognito or a logged-out session so account history and memory don't skew what you see, and note the date and time you ran each check. A spot-check has real value as an early warning signal or a "did that page fix actually move anything" gut check. It has no value as a scorecard, because the sample size is one phrasing, one moment, one context. Treat it accordingly.

Which free tools can check brand mentions in AI search right now?

Ahrefs' AI Visibility Checker and Semrush's AI Search Visibility Checker are both genuinely free, need no signup, and return a result in under a minute. Both are a legitimate first move, not a full tracking system.

Ahrefs' tool queries six platforms (ChatGPT, Gemini, Perplexity, Microsoft Copilot, Google AI Overviews and Google AI Mode) with search-backed prompts and returns total mentions, a platform breakdown, the topics that trigger mentions, and the top domains and pages cited alongside your brand. It shows a limited preview (the top five results per section); full history and custom prompts sit behind the paid Brand Radar product.

Semrush's checker covers ChatGPT, Google AI Overviews, Gemini, SearchGPT and Perplexity, and returns a 0 to 100 AI Visibility Score alongside mentions, cited sources, competitor comparisons, and the actual prompts and search volumes behind each mention. Smaller free scanners exist too. Prominara offers three free scans a day across ChatGPT, Perplexity and Google AI Overviews with no signup, useful as a quick secondary read, though as a newer vendor it's worth treating as a bonus check rather than your main source.

The honest limitation of every free checker is the same one that applies to any spot-check: it's one snapshot, run against a fixed prompt set that isn't necessarily your buyer's actual language, on one day. Run it monthly rather than once, and read it alongside a manual check on your own phrasing, not instead of one.

How do you build a proper prompt-set spreadsheet for AI visibility tracking?

Build a spreadsheet of 25 to 50 real buyer questions, run each one across at least three engines on a fixed schedule, and log mention, position, sentiment and cited sources every time. This is the point where tracking stops being a spot-check and starts being data.

Step 1: Build the prompt set. Skip your brand name. Write the questions your buyer actually asks before they know your name: category questions ("best project management tool for a 10-person agency"), comparison questions ("X vs Y for remote teams"), and decision questions ("is X worth it for a small business", "what's the cheapest alternative to X"). Aim for a spread across the funnel: roughly a third awareness-stage category questions, a third comparison questions naming you or a rival, and a third bottom-of-funnel decision questions. Twenty five prompts is a workable minimum; fifty gives you enough volume that one odd answer doesn't distort the whole read.

Step 2: Set the cadence. Weekly is the realistic minimum for a category that moves (new competitors, funding news, product launches); monthly is the floor for anything you're tracking at all. Given how much a single answer varies by phrasing and moment, a one-off run tells you almost nothing about trend, only about that day.

Step 3: Record it consistently. For each prompt, each engine, each run, log:

Step 4: Look for the pattern, not the single result. After four to six runs you'll start seeing which prompts you consistently lose, which engine barely knows you exist, and which third-party sites keep showing up in the citations. That pattern is the actionable part. One run in isolation is not.

This is genuinely manual work: 50 prompts across four engines, run weekly, is 200 checks a week if you do it by hand. Most teams that try this seriously either automate the prompt runs with a script hitting each engine's interface, or move to a dedicated tool once the spreadsheet proves the method is worth paying to speed up.

What do dedicated AI visibility tools add over a spreadsheet?

Dedicated tools automate the prompt runs, add historical trend lines, and break out share of voice against named competitors, for a monthly fee instead of your own time.

Pricing runs from roughly $20 to $29/month at the entry level up to custom enterprise contracts, and coverage varies sharply on which engines are included at which tier, particularly Claude, which several tools gate behind enterprise plans. Rather than duplicate that comparison here, the full breakdown with verified July 2026 pricing for eleven tools, including Peec AI, Profound and Otterly.AI, is in the best AI search visibility tools for 2026.

What do mention rate, share of voice, position, sentiment and citation actually mean?

These five terms get used loosely across vendor sites. Here's what each one actually measures.

Mention rate and share of voice measure genuinely different things, and the gap between them is worth watching on its own. A brand can be mentioned in a smaller share of total answers than several rivals (lower visibility) while still dominating the conversation whenever it does appear (higher share of voice). Breadth and depth are separate problems, and a tracking setup that only reports one of them is telling you half the story.

Retrieval and citation are the pair most people conflate. A domain can be retrieved constantly and cited rarely, or the other way round. DataForSEO's analysis of 100,249 fanout queries found Wikipedia ranked 106th by raw search appearances but first by citations, while Pinterest appeared in search results far more often than it ever got cited. Citation is also more concentrated than retrieval: the top ten cited domains accounted for about 52.7% of all citations in that dataset, versus the top ten domains making up only 29.9% of search appearances. Appearing in an engine's research pass and getting named in its final answer are not the same achievement, and a good tracking process reports both.

What should you actually do with the tracking data once you have it?

Route every low-position or missing mention to a specific fix: rewrite the page behind that gap, close a gap on the third-party sites engines actually retrieve from, then recheck the same prompt after the fix ships. Measurement without an owner produces a spreadsheet, not visibility.

The biggest gap most tracking setups miss is that "your site" isn't where most citations come from. Our own 30-day tracking of AI engine citations across five engines (June 5 to July 5, 2026) found Reddit was the most-retrieved third-party domain for AI-search-category prompts at 276 retrievals, ahead of LinkedIn at 140, Semrush at 93, Search Engine Land at 77, Medium at 70 and G2 at 51. If your tracking spreadsheet only ever checks whether your own pages get cited, you'll miss most of where the actual citation activity is happening.

The same tracking window showed just how uneven visibility can be engine by engine. Across 603 tracked answers, our own brand, Schmitdy, ranged from 6.15% visibility on Claude down to a flat 0.00% on Google AI Overviews, essentially invisible on one engine while genuinely present on others. That's not a measurement glitch. It's the same pattern behind why a brand can rank well in classic Google and still be invisible in ChatGPT, because the two systems select and reward sources on different rules. The same tracking also had us ranked 5th of 13 tracked competitors on visibility (breadth: how often it's mentioned at all) while ranking 2nd on share of voice at 15.2% (depth: how much of the conversation it owns once it does appear). Both readings are correct. They're measuring different things, which is exactly why the metrics table above matters more than it looks.

That's the pattern worth building a process around: pull the data, find where you're actually retrieved from versus where you wish you were, and put someone in charge of closing the specific gap the data points at. A dashboard that nobody acts on is a subscription, not a strategy, whichever rung of this ladder you're standing on.

If you'd rather see where your own brand currently stands across five engines before building any of this yourself, the free AI search audit runs the check for you.