Why can the same question give different answers?

Generated answers are probabilistic and shift with wording, conversation history, model version, geography and live web access. That is why we look for repeated patterns across a defined set of observations rather than treating any single run as a result.

Is a mention always good news?

No. An answer can name a company while placing it in the wrong category, attaching it to a weak third-party source, or listing competitors ahead of it. We review what the answer says, not only whether the name is present.

Can you guarantee we will be recommended?

No, and any provider who does is describing something they do not control. What observation gives you is a clearer picture of where the public evidence is missing, so the work goes where the gap is.

How does this connect to the technical work?

Directly. A page that cannot be fetched, or an entity that is described inconsistently, is a plausible reason for its absence from an answer. Observation tells you where to look; crawlability and schema work is usually what changes it.

AI Visibility Measurement

AI Citation Tracking for B2B Companies

Repeated, structured observation of how a company appears in generated answers — whether it is named, which of its pages is cited, and whether the description is right. Not a live score, and not a guarantee.

Get a Demo Audit See What We Check

Illustrative interface

Category definitionChatGPTVisible
Shortlist requestPerplexitySource present
AlternativesGoogle AI OverviewsNot observed

Generic values. Not a live dashboard and not client results.

Public example

What an observation looks like

A single record answers four questions: what the buyer was asking about, where the answer came from, whether the company appeared, and which public page was reachable.

Illustrative interface with generic values. Not a live dashboard, not client results, and not a report of any specific run.

Illustrative citation overview: query theme, AI platform, brand visibility state and the public source page.
Query theme	AI platform	Brand visibility	Public source page
Category definition	ChatGPT	Visible	/generative-engine-optimization
Shortlist request	Perplexity	Source present	/services
Alternatives	Google AI Overviews	Not observed	/comparison-hubs
Implementation question	Claude	Visible	/schema-markup-for-ai-search
Problem discovery	Gemini	Review required	/geo-audit

Category definition
ChatGPT
Visible
Public source page
/generative-engine-optimization
Shortlist request
Perplexity
Source present
Public source page
/services
Alternatives
Google AI Overviews
Not observed
Public source page
/comparison-hubs
Implementation question
Claude
Visible
Public source page
/schema-markup-for-ai-search
Problem discovery
Gemini
Review required
Public source page
/geo-audit

Repeated observation over time

Each column is one observation cycle for one query theme. Filled marks are cycles where the company appeared; hollow marks are cycles where it did not. The shape matters, not a number.

Four different things

Mentioned, cited, recommended and correct are not the same

Collapsing them into one number is how teams end up celebrating visibility that does not convert. Each row is observed separately.

Mention: The company name appears somewhere in the answer; Says nothing about whether the description was correct
Citation: A page is named as the source behind a statement; Tells you which page is doing the work — and which is not
Recommendation: The company appears in a shortlist for a buying question; The stage where absence is most expensive
Accuracy: The category and capability described match reality; A confident wrong description is worse than no mention

How it runs

What an observation cycle looks like

A high-level view. Prompt sets, evaluation rules and the reporting cadence are agreed before the first cycle and kept stable so cycles stay comparable.

Agree the question themes
Themes are chosen by decision stage — problem discovery, category definition, shortlists, alternatives — so a gap can be read against how close the buyer is to choosing.
Observe across platforms
The same themes are put to several answer engines under conditions kept as stable as we can, because platforms differ in what they read and what they surface.
Record what the answer said
Presence, the page named as a source and the accuracy of the description are recorded separately, so one strong signal cannot hide a weak one.
Repeat and compare cycles
A single run is an anecdote. Cycles are repeated with the same themes and rules so a pattern — or its disappearance — becomes visible.

What you get

Observations your team can act on

The deliverable is a comparable record across cycles, paired with the pages and entity signals that plausibly explain each gap — ordered by how close the underlying question is to a buying decision.

A baseline you can compare later cycles against
Where competitors appear and your company does not
Which of your pages is being used as a source
A prioritised list of gaps worth closing first

FAQ

Frequently asked questions

Related GEO systems

Authoritative references

Next step

Ready to make your site citable by AI answer engines?

Tell us about your company and website. We will review the request and contact you to discuss the most relevant GEO and AI Visibility audit scope.

Get a Demo Audit Request GEO Audit