[ MEASUREMENT ]

How to track your AI visibility in ChatGPT, Perplexity and Google AI

A practical method for measuring how often AI answer engines name your product: building a prompt set, what to record, how to handle answers that change, and when to use an AI visibility tool.

By Omar Soliman·Updated September 24, 2026·3 min read

“Are we in ChatGPT?” is the wrong question, because the answer is different every time you ask. The useful question is: across the questions our buyers ask, how often are we named, where, how are we described, and why? Here’s how to measure that.

Step 1: Build the prompt set

Your prompt set is the list of questions you’ll test every time. It’s the most important decision, because a bad set produces a meaningless number.

Use buyer language, not your brand name. Asking “what is [your product]” only tests whether the model knows you exist. Ask what buyers ask: “best feature flag tool for a small team”.
Cover intents: category (“best X”), alternatives (“cheaper alternative to Y”), comparisons (“X vs Y”), problems (“how do I…”), and stack-specific (“X for Go/Rails/Kubernetes”).
Include prompts you currently lose. The point is to measure where you’re absent.
Start with 30 to 50. Enough to see a rate, small enough to run by hand.

Pull real phrasings from Reddit thread titles, your sales calls, support tickets and Search Console queries.

Step 2: Pick the engines

At minimum:

ChatGPT (with search on, since that’s how most buying questions get answered now)
Perplexity
Google AI Overviews / AI Mode (what shows at the top of Google)
Gemini

Add Claude and Copilot if your buyers use them. Enterprise developer audiences often do.

Step 3: Run it consistently

Use a clean session: logged out where possible, or memory and custom instructions off.
Keep location and language constant.
Run each prompt several times (three is a reasonable minimum) and record each run. A product named in two of three runs is in a very different position from one named in three of three.
Test on a fixed schedule, weekly or monthly, with the same prompts. Change the set rarely and note when you do.

Step 4: Record the right fields

A spreadsheet row per prompt, per engine, per run:

Field	Why it matters
Prompt, engine, date, run	So you can compare over time
Named? (yes/no)	The core visibility metric
Position in the answer	First named usually gets picked; fifth rarely
How you’re described	“Best for small teams” vs “expensive” vs outdated facts
Competitors named	Who you’re actually losing to
Cited sources (URLs)	Your to-do list: the pages deciding the answer
Full answer text	So you can re-read what was said, not just the score

Step 5: Turn it into metrics

Visibility rate: runs where you’re named ÷ total runs. Track overall and per engine.
Average position when named.
Share of voice: your mentions vs. each competitor’s across the same set.
Sentiment and accuracy: share of mentions that are positive and factually correct.
Top cited domains: which sites keep appearing as sources. For developer tools, expect Reddit near the top.

Watch the trend, not individual runs. A move from 12% to 30% over three months is real. A single prompt flipping between yes and no is noise.

Step 6: Act on the citations

The measurement only pays off when it tells you what to do. For each prompt where you’re absent or badly described, look at the cited sources:

A Reddit thread: can you add a genuinely useful comment? See Reddit marketing for developer tools.
A comparison article: is the information about you accurate? Contact the author.
Your own page: does it answer the question in its first paragraph?
A competitor’s page: publish your own honest comparison.

When to use a tool

Manual tracking works up to a few dozen prompts a month. Beyond that, dedicated AI visibility tools (Profound, Peec AI, Otterly and others) run prompts on a schedule across engines, keep history and compare competitors. When choosing one, check that it shows you the raw answers and cited URLs, not just a score, and that you control the prompt set. If you’re weighing one against the rest of your tooling budget, aistack.sh, another project of ours, lists tested AI tools for marketers with the workflow and real monthly cost of each stack.

What we report

Our monthly report for clients is built on this method: a fixed set of buyer prompts tested weekly across ChatGPT, Perplexity and Gemini, the full text of every answer, and the weeks you’re not in them. It sits next to every Reddit thread we published and its Google rank, so you can see which threads the engines start citing.

Questions people ask

What is AI visibility?

AI visibility is how often and how favorably AI systems like ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews mention your brand when people ask questions relevant to your product. It's usually measured as the share of a fixed set of buyer prompts where you're named, plus your position in the answer, how you're described and which sources are cited.

How do I check if ChatGPT mentions my brand?

Ask it the questions your buyers ask, not your brand name, in a fresh chat with memory off, and note whether you're named. Repeat each question several times, since answers vary between runs. Do the same in Perplexity, Gemini and Google. For more than a handful of prompts, use a spreadsheet or an AI visibility tool.

Why do AI answers change every time I ask?

Model outputs are probabilistic, search-enabled answers retrieve different pages on different runs, and results can vary by location, account, model version and exact wording. That's why AI visibility has to be measured as a rate across many prompts and runs, not as a single yes or no.

Do I need an AI visibility tool?

Not at first. A spreadsheet and 30 prompts tested monthly will tell you where you stand. Tools help once you want to test hundreds of prompts, run them frequently, track competitors and keep history without manual work.

[ KEEP READING ]