Home / Resources / Tracking AI search ranking
GEO · MeasurementHow to track your AI search ranking and visibility
There's no rank #1 in an AI answer — you're either mentioned or you're not, described accurately or not. Traditional rank trackers can't measure this at all. Here's the exact method we use, a scoring rubric you can copy today, and the mistakes that quietly waste people's first attempt at this.
Why rank trackers don't work here
A rank tracker checks your position on a results page for a fixed keyword — "position 3 for 'best CRM software'," refreshed daily. That model assumes a stable page, a stable position, and a stable query. None of that holds for AI search.
Ask ChatGPT "best CRM software for a 20-person team" twice in the same week and you can get two differently worded answers, citing a different mix of sources, in a different order. There is no page, no position, and no fixed query to track — the answer is synthesized fresh from whatever the model retrieves and weighs as credible at that moment. A tool built to watch a static SERP is structurally blind to this. You need a measurement unit that survives the answer changing every time: not "where do I rank," but "was I mentioned, and was I described the way I'd want to be."
What "visibility" actually means in AI search
Before you can track it, you need to define it precisely enough to score consistently. We break every AI response down into four separate yes/no questions, because a brand can pass some and fail others in the same answer:
Mentioned
Is your brand named anywhere in the response at all — even in passing, even without detail?
Recommended
When the question implies a choice ("which one should I pick"), does the engine put you forward as an option, not just acknowledge you exist?
Accurately described
Are your positioning, pricing and category correct — or is the model hallucinating a feature you don't have or a price you don't charge?
Cited over competitors
When a named competitor is also a plausible answer, did the engine choose to surface you, them, or both?
Score all four separately for every question in your panel. A brand that's "mentioned" in 80% of responses but "recommended" in only 10% has a very specific, fixable problem — one a single aggregate score would hide completely.
The prompt-panel method
Four steps, repeated on a schedule. Each one matters more than it looks — most DIY attempts fail on step 1 or step 4, not the parts people worry about.
Build a query panel
30–40 real questions your buyers would actually type — not just your brand name, but category questions, comparison questions and buying-intent phrasing.
"best project management tool for a remote agency" · "[Competitor] vs [Your Brand] for small teams" · "is [Your Brand] good for construction project tracking"
Run it across engines
Ask every question in ChatGPT, Gemini, Claude and Perplexity — same wording, fresh conversation each time so prior context doesn't bias the answer. Each engine has different training data and browsing behavior, so expect real variance between them.
Score each response
Use the four-question rubric above for every single answer. Log it in a spreadsheet with the exact question, the engine, and the date — you'll need this history later.
Track monthly, not once
Re-run the identical panel on the same schedule. A single snapshot tells you where you stand today; the trend line is what tells you whether what you're doing is working.
A scoring rubric you can copy
Turn the four dimensions into a simple weighted score per response, then average across your panel for a monthly number you can actually compare month over month:
| Signal | What it means | Points |
|---|---|---|
| Mentioned | Named anywhere in the response | +1 |
| Recommended | Put forward as an option when a choice was implied | +2 |
| Accurately described | Positioning, pricing and category stated correctly | +1 |
| Cited over a named competitor | Chosen instead of, or alongside, a direct competitor | +2 |
| Hallucinated detail | Wrong price, wrong feature, or confused with another brand | −2 |
Six points is the maximum per response. Average your panel's scores monthly, and split the average by engine — a brand can be a 5/6 on Perplexity and a 1/6 on Gemini, and averaging them together would hide exactly the gap you need to close first.
Common mistakes when tracking this
- ✕Only testing branded queriesAsking "tell me about [Your Brand]" confirms the model has heard of you — it tells you nothing about whether you're recommended when a buyer doesn't already know your name.
- ✕Testing one engine and calling it doneChatGPT and Gemini pull from different sources and weigh them differently. Strong visibility in one is not evidence of visibility anywhere else.
- ✕Running the panel once and stoppingA single check is a snapshot, not a trend. Without a monthly re-run on the same panel, you can't tell whether a change you made actually moved anything.
- ✕Ignoring hallucinations because you were "mentioned"Being named with the wrong price or a feature you dropped two years ago is often worse than not being mentioned at all — it actively costs you the sale.
- ✕Changing the panel between runsSwapping in "better" questions each month feels like an improvement but destroys your ability to compare month over month. Lock the panel; only add to it deliberately, and note when you did.
Free vs proper tracking
You can run this manually for free — open each engine, type your panel of questions, log the results in a spreadsheet. It works, but it's slow, and it's easy to introduce inconsistency between sessions (different wording, different order, someone forgetting to log a hallucination). A proper program automates the panel, standardizes scoring against the rubric above, and benchmarks against named competitors automatically — which is where a monthly GEO audit earns its keep.
Start free, upgrade when it matters. A manual quarterly check is enough to know if you have a problem. A monthly automated panel, scored consistently and benchmarked to competitors, is what you need once AI visibility is a real revenue channel.
Try it yourself, free
Our free AI rank tracker generates a personalized prompt panel for your brand and category in seconds — no email required — so you can run this exact method manually before committing to anything paid.
The full build
The automated, monthly version of this method is the core of our GEO Audit & AI Visibility Score — a 40-prompt matrix across four engines, scored against the rubric above, with competitor benchmarking, refreshed monthly.
Common questions
How often should I check my AI search visibility?
Quarterly is enough to know whether you have a problem. Monthly is what you need once AI-driven discovery is contributing real revenue, since AI answers shift as engines retrain and as competitors publish new citable content.
Which AI engines actually matter to track?
At minimum ChatGPT, Gemini, Perplexity and Claude — they have different training data, different browsing/retrieval behavior, and different audiences, so a brand can be strong in one and invisible in another. Google AI Overviews is worth adding once the core four are covered.
Is there a free way to track ChatGPT visibility?
Yes — run the prompt-panel method manually: ask your panel of questions in each engine and log mentions, recommendations and accuracy in a spreadsheet. It's slower and less consistent than an automated panel, but it's genuinely free and works. Our free AI rank tracker helps you build the panel in seconds.
How is this different from SEO rank tracking?
A rank tracker checks a fixed position on a results page for a fixed keyword. AI answers are generated fresh each time with no stable position to track, so the unit of measurement changes from "rank" to "was I mentioned, recommended and described accurately" across a panel of real questions.
What's a good AI visibility score to aim for?
There's no universal number — it's relative to your category and competitive set. The useful benchmark is your own baseline: are you mentioned in more of the panel this month than last month, and is your share of voice growing relative to your named competitors?
See how visible you are to AI today
A free GEO audit baselines your brand across four AI engines vs your top competitor — 40 prompts, a real number, no pitch unless you ask.
Get my free GEO audit →