Sagentel
Sign inBook a demoGet started
FeaturesProcessContentPricingCertificationChatGPT visibilityPerplexity visibilityGemini visibilityCopilot visibilityGoogle AI OverviewsSaaSFinanceAgenciesHospitalityTravelHealthKnowledge BaseBlogSign inGet started
ChatGPT Tracking

ChatGPT recommends one brand. Is it yours?

Sagentel replays the questions your customers ask, records the answers, and shows the pages behind them.

Start trial →Book a demo
ChatGPT visibility dashboard
What we measure

Being named and being chosen are two different outcomes

Counting appearances tells you almost nothing on its own. We keep four readings apart, because each one points at a different piece of work.

Appearance rateShare of a fixed question set where your name survives into the final answer, compared run against run so a change is provable.
PlacementOpening recommendation, one option among several, or a footnote after the advice has already been given.
CharacterisationThe adjectives attached to your name, plus stale prices, wrong markets and mixed-up product lines worth disputing.
Evidence trailEvery page quoted to build the answer, ranked by how often it turns up, with the ones missing your name flagged.
How it works

What the model reaches for before it answers

Nothing here is a ranking factor in the old sense. An answer is assembled from material the model can locate, believe and quote — and each of those is something your team can change.

01
A name the model is sure aboutIf listings, press and forum threads describe you three different ways, the model treats you as ambiguous — and ambiguity gets dropped rather than risked in a short answer.
02
An answer to the actual questionCategory awareness earns you nothing here. The passage has to resolve the specific question, in wording that can be lifted whole into a two-sentence reply.
03
Somebody else vouchingRound-ups, review threads and editorial picks that put you on a list. In most categories these carry more weight than anything published on your own domain.
04
Something recent to quoteThe pool of quotable material refreshes constantly. Keep adding to it and your place holds; stop, and the slot fills with whoever published last month.

Built for the people who own the number

In-house search teams
A number that survives the Monday report

One reading per engine and per question cluster, with the missing evidence listed underneath it. No manual collation before the meeting.

In-house search teams
Split by market, line and sub-brand

Keep every language and product family on its own track, and catch a wrong price or discontinued model while it is still only the model saying it.

Agencies
Something you can renew, not just present once

Separate workspaces per client, question sets that stay fixed between runs, and a before-and-after you can defend line by line.

Questions we get asked

What are you actually measuring?

Four readings, deliberately kept apart: how often your name reaches the final answer, where it lands inside it, the language wrapped around it, and the pages quoted to justify it. Appearing on the back of a rival's article is a different problem from not appearing at all, and the remedy is different too.

Which questions are worth watching first?

The ones where a decision gets made: what to buy, which of two to pick, what to use instead, what it costs, and anything pairing you with a rival. Around forty per category is enough to read a trend; widen the set once the obvious holes are closed.

Can any of this be changed?

Not by editing the answer, but yes by changing what it is built from. In practice one of three things is holding you back — thin coverage of the question, nobody outside your site vouching for you, or a name the model cannot pin down. Measurement's job is to say which one.

We rank well in Google. Isn't that the same thing?

It helps, and it guarantees nothing. Ahrefs found 28.3% of the pages ChatGPT cites most have no measurable organic visibility in Google at all. We routinely see small specialists appear more often than the category leader, because one channel rewards a ranked page and the other rewards a quotable one.

How stable are the answers?

Less stable than most teams expect. Wording and quoted pages drift as models refresh and new material is indexed, so a single audit is out of date within weeks. We re-run the same question set on a schedule and report the movement rather than the moment.

What does the first month look like?

Run one: a baseline across your starter question set, the names appearing in your place, and the pages behind those answers. Run two, a week later, shows what moved on its own. From there you get an ordered list of fixes, and a working session with a strategist if you want help sequencing it.

Somebody asked about your category this morning.

One domain, one question set, first results the same day — appearance rate, who filled your slot, and the pages that decided it.

Start trialBook a demo