Technical

How does AI search visibility tracking actually work?

By Seene Team · 5 min read

Published 23 September 2026

AI search visibility tracking works by asking AI assistants the same questions your customers ask, on a schedule, and recording what comes back. A tracking tool sends a set of realistic customer questions to ChatGPT, Claude, Gemini and Perplexity, reads every answer, and records whether your business was mentioned, recommended, cited as a source, and which competitors showed up instead. Those results roll up into a score you can watch move over time.

There's no Search Console for ChatGPT. OpenAI, Anthropic and Google don't publish how often they name your business, so the only way to measure it is to ask and count. Below is how that works in practice, step by step, using Seene's own pipeline as the example.

How does a tracking tool know what your business is?

It reads your website first. Before any questions get asked, the tool needs to know what you do, where you are, and what a customer would actually search for. Seene pulls your business name, category, location, and two to five specific things you're known for, described the way a customer would search for them rather than your own branded product names.

This step matters more than it sounds. A cricket gear brand's customers search "best cricket box for fast bowling," not "sports protection." If the tool gets your category wrong, every question after it tests the wrong thing.

What questions does it actually ask the AI?

It asks the questions a stranger would ask before they've heard of you. Good tracking uses a portfolio of question types, not one keyword. Seene's discovery questions fall into four groups:

Unbranded: broad searches for what you sell ("where can I get a decent cricket box online?")
Local: searches anchored to your city or region ("good electrician in Newcastle for a switchboard upgrade")
Attribute: searches filtered by something buyers care about, like price or quality
Problem: the customer describes their problem, not your category ("my tap's leaking, who do I call?")

None of these questions includes your business name, and that's deliberate. Asking "is [your business] any good?" only proves the AI recognises you. It doesn't prove the AI would ever bring you up unprompted, which is what actually wins customers.

Branded questions still get asked, but separately. Seene keeps them out of the headline score, because a question that names you is almost guaranteed to mention you and would inflate every business's number by the same amount.

How are the answers collected?

Each question is sent through the AI platform's search-grounded mode, the same mode that looks things up on the web when a real person asks for a recommendation. Plain, offline model answers only reflect old training data, while search-grounded answers reflect what the AI finds today.

A few details separate useful tracking from noise:

Clean sessions. No login, no chat history, no memory. Your own ChatGPT account knows you, which skews the answer.
The right location. The question is grounded in the customer's likely location. Without this, most AI search defaults to US results, which is useless for a business in Brisbane.
Multiple platforms. ChatGPT, Claude, Gemini and Perplexity use different search sources and often name different businesses for the same question.

How does the tool decide if you were "mentioned" or "recommended"?

A separate model reads each answer and grades it, so no platform grades its own answer. Seene uses one fixed classifier model for every platform, which means ChatGPT's answers aren't judged by ChatGPT. It starts with a plain text match on your business name and known variations, then the classifier decides:

Were you mentioned at all?
Were you recommended, not just listed?
What position were you in the list?
Was the tone positive, neutral or negative?
Which competitors were named?
Did the answer cite your website as a source?

What metrics come out the other end?

Metric	What it tells you
Mention rate	Percentage of customer questions where your name appears
Recommendation rate	Percentage where the AI actively suggests you
Share of voice	Your mentions as a share of all business mentions, yours plus competitors
Average position	Where you sit in the list when you do appear
Citation rate	How often the AI links to your own website as a source
Competitor frequency	Who gets named most often instead of you

Seene's score is a mention rate with a confidence range attached (a Wilson score interval) that shows how solid the number is given how many answers were checked. A score of 30% from 10 answers is a rough read. The same 30% from 80 answers is much firmer.

Why does AI visibility need to be tracked over time?

Because AI answers move. The same question asked on Monday and Thursday can name different businesses, since the model, the search index behind it, and the pages it reads all change constantly. One check is a snapshot. A weekly check across the same set of questions shows you whether you're actually gaining ground, and whether a change you made to your site did anything.

That's also why tracking is only half the job. Once you know where you're missing, the fixes are covered in our guide to what determines AI search visibility, and what AI search optimisation costs covers what it's worth paying for.

How Seene helps

Seene's free scan runs 10 realistic customer questions through ChatGPT right now and shows you each answer, who got named instead of you, and your score. A paid plan tracks the same questions weekly across ChatGPT, Claude, Gemini and Perplexity, starting at $39 AUD/mo.

Check your AI visibility for free →

Frequently asked questions

Do I need a tool to check AI search visibility, or can I just ask ChatGPT myself?

You can ask manually, and it's worth doing at least once, but a single check without a clean session, the right location context, and a fixed set of questions isn't repeatable or comparable over time. A tool matters most for tracking the same questions on a schedule, across multiple platforms, so you can see whether you're actually gaining ground.

Why don't tracking tools just ask "is [my business] any good?"

Because that question already tells the AI your name, so it's almost guaranteed to mention you. It proves the AI recognises your business, not that the AI would ever bring you up to a customer who hasn't heard of you yet, which is the thing that actually drives new customers.

Why does the AI give different answers to the same question from one day to the next?

Search-grounded AI answers depend on the model, the live search index behind it, and the specific pages it reads at that moment, all of which change constantly. A single check is a snapshot of that moment, not a stable measurement, which is why tracking the same questions weekly matters more than any one result.

Does my own ChatGPT account affect the results if I check myself?

Yes. If you're logged in, or have chat history and memory enabled, the model can use what it already knows about you, which skews the answer toward mentioning you. A genuine test needs a clean, logged-out session with no prior context, the same way a real stranger would ask.