In measured numbers
Every figure on this page is read from a code path, not written by hand — the same discipline the product applies to your own data.
What gets measured
| Engine | Model | Whose key |
|---|---|---|
| Claude | Anthropic | Included in the plan |
| ChatGPT | OpenAI | Your own key, like any other connection |
| Gemini | Your own key, like any other connection |
Each engine is asked the questions a user would ask — "what is the best dream interpretation app" — and the answer is scanned for your name and for the names of the competitors already known from your rank measurements. What each engine said is reported separately; we do not average them, because one engine is not a measure of another.
How it works
- Questions are set for your app and its category, in the language of the storefront.
- Each engine answers from its own knowledge. We do not make the assistant search the web — what we measure is the answer most people actually get.
- The answer text is scanned for names. Your app, and the rivals known from your measurements.
- The result is stored with its date so that the trend, not the single reading, becomes the evidence.
What you do with it
Find out if you exist there
For most apps the first honest answer is “not yet”, and that is worth knowing before you plan around it.
Read what AI visibility isSee who is named instead
The apps assistants recommend in your category are a different competitive set — and often a more durable one.
See what a report containsWatch the trend after coverage
Assistants name what the open web writes about. A review shows up here before it shows up anywhere else.
Pull the same numbers into your assistantWhat this does not tell you
One reading is not proof. The trend across readings is.
The scoring is not done by the model either — we search the answer text for names rather than asking the model where an app ranks. A model grading its own output is an opinion, not a measurement.
- A model's training data freezes on a date. A new app missing from the answer is not doing badly — it is not yet known.
- The answer is not identical every time. That is why the result is stored with its date.
- We do not make the assistant browse. That would measure a retrieval pipeline, not the answer most users get.
- It is not a ranking you can optimise directly. There is no keyword field for an assistant.
What each plan includes
| Free | Pro | Studio | |
|---|---|---|---|
| AI visibility measurement | — | — | Yes |
| Runs per day | — | — | 3 |
| Claude engine | — | — | Included |
| ChatGPT and Gemini | — | — | Your own key |
AI visibility is a Studio feature (60 $/month, 54 $/month billed yearly).
Frequently asked
Which assistants do you measure?
Claude, ChatGPT and Gemini. Claude runs on our key and is included in the plan; ChatGPT and Gemini run on your own key, like any other connection.
Do you make the assistant search the web first?
No. We measure the answer the model gives from its own knowledge, because that is the answer most people see.
My app is not mentioned at all. Is that bad?
Not necessarily. Training data freezes on a date, so a recent app is usually not yet known rather than badly regarded. What matters is whether that changes over time.
Do you average the three engines into one score?
No. Each engine is reported separately. One engine is not a measure of another, and an average would hide the only interesting case — where they disagree.
See this for your own app
Paste your store link, add your keywords. Rank, the apps above you and demand are measured every day.
Start free →