FAQ

About the measurement

Which engines do you measure?

ChatGPT, Google AI Overviews, Google AI Mode and Perplexity, captured from their real consumer interfaces. Gemini is deferred. A small panel of grounded provider APIs runs alongside as calibration and never mixes into the same metric. See AI engines.

Do you use the APIs or the real product?

The real product. An API answer is not what a user sees — the consumer surfaces route between models, personalize, and render ads and shopping units the API never returns. Measuring the API tells you what a model would say in a laboratory.

How often is data collected?

Daily. Dispatch runs at 01:00 UTC with hourly passes for anything added since, and a reconciliation sweep every five minutes chases jobs that did not report back. Every prompt runs every day — there is no cadence quietly downgrading half your set to weekly.

Why do my numbers change when I did nothing?

Because AI answers are non-deterministic and source selection shifts. That is exactly why each prompt is asked several times a day per engine and every rate carries a confidence interval. Check whether the change cleared the noise floor before treating it as real. If it did and it was large, look at Sources — most unexplained movement is an engine changing which domains it reads.

What is the noise floor?

The threshold a change must clear to be called a change. Comparing two independent estimates carries about 1.4× the confidence half-width of either one, so the floor quoted beside a delta is wider than the interval on the endpoints. Moves inside it are not marked significant.

Do you ask the model which brands it mentioned?

No. Mentions are extracted from the captured answer text by deterministic matching. Asking a model to grade its own output requires modifying the prompt, which changes the answer being measured. See Mentions and sentiment.

What if my brand name is an ordinary word?

Handled specially. Names like On or Arc are matched with stricter case and context rules, because a naive matcher counts the preposition and inflates visibility toward 100%. See Mentions and sentiment.

What is the difference between a quoted and a listed source?

Quoted means the engine cited the page against a specific claim in the answer. Listed means it appeared in the sources panel. They barely overlap in practice, and being quoted is the one that moves an answer. Both are tracked separately in Sources.

Can I compare my visibility score to another tool's?

No, and be wary of anyone who says otherwise. Scores depend on the prompt set, the entity set, the engines, the locations and the formula — and on whether branded prompts were excluded, which inflates the number badly when they are not. Compare within one tool, over time, against your own competitors.

About using the product

How many prompts should I track?

Enough for each subject to produce a usable interval. Roughly 2 answers per prompt per day means one prompt over 30 days gives a band about 22 points wide; four prompts on the same subject gives about 11. The bounds are 150 prompts total and 10 products, and the split is yours to choose. See Prompts.

Why can't my visibility prompts contain my brand name?

Because a prompt naming you guarantees an answer naming you. Branded questions belong in the perception groups, where naming you is the point; mixing them into visibility measures your own name back at yourself.

How many competitors?

Three to six. Share of Voice divides among tracked entities, so a bloated set makes every share small and every movement look flat. See Competitors.

My visibility went up but Share of Voice went down. What does that mean?

The category is getting more coverage than you are capturing. More answers name entities; a smaller proportion name you. Usually a competitor gained a source. See Share of Voice.

Why does a value show a dash instead of zero?

Because it has not been measured. A product added today has no reading; a product measured across 200 answers and named in none has a reading of zero. Those are different claims, and the interface never conflates them.

I fixed a page. When will it show up?

Days, sometimes longer. The engine has to re-crawl the page and then select it as a source. Collection is fast; search infrastructure is not.

Which screens are showing my data today?

Overview, Sources, Query Fan-out, Ads & Shopping, Brand Perception, Prompts, Competitors, Products and PR Opportunities. Recommendations, Pages Optimization, GEO Copilot, Pages and AI Tracker are previews on sample data. See Product status.

About GEO in general

Does GEO replace SEO?

No. They overlap heavily — the engines search the web, so authority and crawlability still matter — but the targets differ. SEO optimizes for a ranked list; GEO optimizes for being extracted, named and recommended inside a synthesized answer.

If I block AI crawlers, do I disappear from ChatGPT?

It depends which crawler. Blocking a training crawler does not remove you from AI search. Blocking a search crawler does. Many sites block both by accident. Check with the AI Crawlability Checker.

Do ads and shopping units matter?

Increasingly. They render above or beside the answer on several surfaces and are frequently the first thing a buyer sees. Share of those slots is measured separately from prose visibility. See Ads & Shopping.

About accounts and billing

Who can manage billing?

Members with the billing.manage permission — the owner role by default. Billing belongs to the workspace, not to a person. See Teams and roles.

What happens to my data if I cancel?

It is retained; collection stops. Your history stays readable, no new runs are dispatched, and resubscribing resumes daily collection with a gap for the period you were not collecting.

Is my data isolated from other customers?

Yes, enforced by database row-level security with cross-tenant tests in the migration suite, not by application code. See Security and privacy.

Can I put confidential terms in a prompt?

Don't. Prompts are asked of third-party AI surfaces through a collection provider, so an unreleased product name in a prompt is an unreleased product name in someone else's logs.

Something not here

Use the contact page. Include the workspace, the screen and what you expected — it saves a round trip.