Understand how AI recommends your brand

TrackGeo measures how AI assistants answer the buying questions in your category. It tracks whether you get mentioned, which competitors get recommended instead, which sources the answer leaned on, and what to change to improve your position — with the underlying response kept as evidence.

AI visibility is how often, and how favourably, your brand appears when an assistant answers a question a buyer would actually ask. It is not a keyword ranking. There is no position one, and no single answer everyone sees — which is why it has to be measured by repeated sampling rather than a lookup.

TrackGeo measures 6 surfaces: ChatGPT via the OpenAI API, Perplexity via its sonar API, Gemini, Claude via the Anthropic API, Azure OpenAI, and Gemini with Google Search grounding. Importantly, these are provider APIs — the models behind the assistants — not scrapes of the consumer apps. We do not read Google AI Overviews from a search page, and we say so rather than implying otherwise.

You can write prompts yourself, import them, or start from suggestions generated for your brand and category. The prompts that matter are the ones with commercial intent, so they are classified by intent and funnel stage — a comparison question and a definition question are not worth the same effort.

Language models are non-deterministic: ask the same question twice and you can get two different answers, with different brands named. One response tells you almost nothing. Running a prompt repeatedly, over time and across engines, is what separates a real pattern from a coincidence.

Confidence reflects sample size and consistency — how many runs a finding is based on, across how many engines, over what window, and how stable the result was. Every finding shows its sample so you can judge it yourself. A result from a handful of runs is labelled as such rather than presented as fact.

No. Nobody can, and you should be sceptical of anyone who says otherwise. Assistants change models, prompts and grounding without notice, and their outputs vary run to run. TrackGeo shows you what is happening, what changed, and what evidence supports a recommended action. The outcome is not ours to promise.

Competitors are detected from the answers themselves — the brands assistants actually name in response to your prompts — combined with the competitor set you define. That means you find rivals you did not know you had, not just the ones already on your list.

Yes. Each client gets an isolated workspace with its own brands, prompts and competitors. Workspace count is the plan lever: Professional includes 2, Growth includes 5, and Enterprise is unlimited. White-label reporting is available on Growth and Enterprise.

Not yet. The ingestion is built and the entitlement is reserved for Professional and above, but there is no way for you to connect an account today — it is in development, and we would rather say so than let you buy a plan for it. What does work now is TrackLink attribution for assistant referral traffic. Ask us on the contact page if this is a deciding factor and we will tell you where it actually stands.

One credit is one answer check: a single prompt, on a single engine, in a single region. So 10 prompts across 3 engines in 1 region costs 30 credits per run. Monthly allowances are Essentials 500, Professional 18,000 and Growth 135,000, which is sized for daily scanning at each plan's prompt and engine limits. Unused plan credits do not roll over; purchased top-up credits never expire and are only spent once the monthly allowance is gone.

Yes, with no signup and no card. The audit on this page fetches your homepage and scores whether it is structured to be understood and cited by an assistant, and shows the buyer questions we would track for you. It is a readiness check of your site — it does not query the assistants themselves. That happens inside your workspace, where runs are attributed to your plan.

Authentication runs through Zitadel OIDC with httpOnly session cookies rather than tokens in browser storage. Workspaces and brands are isolated, access is role-based, and you control how long prompts and responses are retained. Prompts are sent to the model providers listed on our engines page in order to run them; we hold no security certifications today and do not claim any.