AI visibility tracking: see what ChatGPT, Claude and Perplexity say about you

1. 1

   Buyer questions from your site

   When you add a website, GEOHATS reads it and suggests 30 to 38 questions your buyers ask AI assistants: alternatives, comparisons, questions about competitors, and advice questions. You can star the ones that matter most or pause any of them.
2. 2

   Each assistant gets asked, as a buyer would

   Every active question goes to each assistant you picked, through its API with web search on and no system prompt, so the answer is as close to the consumer experience as an API allows. Scans use each provider’s lower-cost model to keep your bill small.
3. 3

   The answer is parsed

   GEOHATS checks whether your brand is named (and its position if the answer is a numbered list), collects the pages the assistant cited as sources, and marks which are yours and which belong to competitors.
4. 4

   Failed calls are left out

   A rate limit or a dead key is missing data, not a lost recommendation, so errored answers never count against your score. A scan where every call failed doesn’t use up a run.
5. 5

   Score, map and changes update

   The visibility score, the question map and the “what changed since last scan” list update. Any page you published through GEOHATS that is now cited moves to “cited” on its own.

Costs are the estimates the app shows before a scan; your provider bills the real amount, which GEOHATS logs per website. Full cost examples are on the pricing page.

• Named or not.Whether the reply text names your brand. Matching handles variations like “AI Render” vs “AI-Render”, and very short brand names are matched case-sensitively to avoid false hits on common words.
• Position. If the answer is a numbered list, the number of the first item that names you.
• Cited pages. The URLs the assistant cited as sources, marked as yours or not. Being named and being cited are separate signals: either can happen without the other.
• Competitors cited instead. The other domains in the sources, which become the pages to beat and, later, the input for the content writer.
• The answer itself. A snippet of what each assistant said, with your brand highlighted, so you can read the context rather than trust a number.

One number per scan, built from three parts of the same answers:

The score comes with a plain label (Invisible under 20, Barely visible, Emerging, Competitive, Dominant from 80) and a history chart. If the set of assistants that answered or the question list changed between two scans, the change is flagged as not comparable instead of being shown as a gain or a loss.

Every buyer question is one node on a map, grouped into four themes: alternatives, comparisons, competitor questions, and advice and how-to. Click a node to see, per assistant, who it quotes today and what it said. Each node shows where you stand:

If the site optimizer has crawled your site and you already have a page that covers a question, the plan says so, so you improve that page instead of writing a competing one.

Add your competitors’ domains and GEOHATS counts, over the same answers, how often each one is named or cited compared with you, and charts your share of voice against the strongest rival from scan to scan. Competitor names are matched with a heuristic based on their domain; cited counts are exact domain matches. Questions that name a competitor in the question itself are excluded from the benchmark, since the answer will name them anyway.

AI answers vary. Ask the same question twice and you can get a different list. By default each scan asks every question once per assistant, so a single answer is one sample, not a verdict. Scans also go through each provider’s API with a lower-cost model, which is close to, but not the same as, what a signed-in user sees in the ChatGPT or Claude app.

Read the trend across several scans, not one result. We wrote up the limits of every tracker, ours included, in how accurate AI visibility trackers really are.

Automatic re-checks

Each website can be re-checked daily or weekly, or only when you click scan (the default, since scans run on your keys). After each scan you get a list of the sources you won and lost since the previous one, and published pages that start getting cited are marked automatically. That is how you see whether a page worked, usually 2 to 8 weeks after publishing.

A report link for clients

Turn on a read-only share link for a website and send it to a client or your team. It opens without a login and shows the current score, the change since the last scan, the sources won and lost, and which published pages are being cited.

• How accurate are AI visibility trackers? What the numbers mean
• How to track brand mentions in AI answers
• Profound vs Peec AI vs Otterly.AI compared
• Affordable AI visibility tools