Docs / Tutorials
Measure AI visibility
Track whether ChatGPT, Claude, Gemini and Perplexity mention and cite your brand for the questions buyers ask.
AI visibility asks the questions your customers ask, across the assistants they use, and records whether each answer mentions your brand, whether it cites your site, and where you come compared with your competitors.
Set it up
Describe your brand
Open AI visibility and go to Setup. The brand name starts as your project name and the domain as its website. Add other names people use: abbreviations, product names, common misspellings.
Add competitors
Up to 20. They cost nothing extra: they are looked for in the same answers, which gives you share of voice. A domain lets OpenSEO count their citations too.
Choose the models
The default set has one current model per assistant people use: ChatGPT, Gemini, Claude and Perplexity. Grok, Meta AI and DeepSeek are one click away, and All models lists more than 120. Each model answers every prompt, so each one multiplies the price of a run. The exact model ID is stored with every answer, so a new model version shows up in your trend instead of silently changing it.
Keep web search on
With web search, models look things up and list their sources, which is what makes citations possible. It adds roughly a cent or more per answer. Without it, models answer from memory and cite nothing.
Save
Click Save setup. Saving spends nothing. You can also set a daily, weekly or monthly schedule here.
Add prompts
On the Prompts tab, add the questions a buyer might type into an assistant, one per line, for example What is the best invoicing software for plumbers?. Leave your brand name out: the point is to see whether the assistant brings you up by itself.
No ideas? Suggest prompts drafts them from a sentence about the business, for about a cent. Nothing is added until you tick the ones you want. Pause a prompt to stop paying for it while keeping its history.
Run it
The bar at the top shows one run (prompts × models answers), the minimum price ($0.01 per answer), and, once every model has answered before, the expected price from your history. Click Run now and confirm. You can cancel a run that is going; answers already collected are kept.
Read the results
| Measure | How it is counted |
|---|---|
| Mentioned | The answer’s text contains your brand name, one of your other names, or your domain, as a whole word in any letter case. Links, email addresses and citation markers such as [1] do not count: a link to your site is a citation, not a mention. |
| Cited | One of the sources the model listed is on your domain or a subdomain of it. |
| Position | The order in which tracked brands are first mentioned in the answer: 1 means you came up first. |
| Mention rate, citation rate | Share of answers in a run that mention or cite you, counting only prompts that do not name your brand. |
| Share of voice | How many answers mention each tracked brand, yours highlighted, with how often each came first or was cited. |
The Overview tab has the latest run in detail: rates by model, share of voice, the domains models cite most (good link-building targets), and a grid of prompts by model. Click a prompt to read each model's full answer and its sources. After a second run, a trend chart appears.
What is left out, and when a change is shown
- Prompts that name your brand ("Is Acme any good?") nearly always get an answer that mentions it, so they would flatter your rates. They are left out of the rates, share of voice and trend, and marked names your brand in the grid.
- Failed or empty answers are left out too, rather than counted as answers that missed you. They show as No answer in the grid.
- The change since the previous run is shown only when both runs answered at least 90% of what they planned, and only over the prompt, model and brand combinations both runs asked. If a run fell short, or you changed your prompts, models or brand name in between, the page says why there is no change instead of showing a misleading one. Trend points from runs under 90% are marked as less reliable.
- Cited links are stored without tracking parameters (
utm_,gclidand similar), so the same page is not counted twice.
Try a question first
The Explorer tab asks one prompt to up to 8 models side by side, without adding it to your tracked prompts or your trend. It is a quick way to test a question before you track it. It is priced by tokens, with a $0.01 minimum per model.