Storelift › Guides › Our own AI visibility, published
Our own AI visibility, published
Storelift measures whether an app gets named in AI assistant answers. It would be strange to sell that and never run it on ourselves — so we do, on a fixed prompt set, and we publish the result including the part where the answer is zero.
The result, 2026-08-18
| Engine | Model | Named in discovery prompts |
|---|---|---|
| Claude | claude-sonnet-4-5-20250929 | 0 of 5 |
| ChatGPT | — | Not measured — no key on this account |
| Gemini | — | Not measured — no key on this account |
Who gets named instead
This is the useful half. Across the discovery prompts, the assistant consistently names the same short list:
| Tool | Named in how many of 5 prompts |
|---|---|
| AppFollow | 5 / 5 |
| App Radar | 5 / 5 |
| Sensor Tower | 5 / 5 |
| AppTweak | 4 / 5 |
| Mobile Action | 4 / 5 |
Two things stand out. First, the list is entirely enterprise and mid-tier — the $9–39 indie tier that most solo developers actually buy from does not appear at all, in any prompt, in either language. Second, the Turkish prompts return the same names as the English ones, which suggests the answer is not localised so much as translated.
What a name match cannot tell you
Asked directly — What is Storelift? — the assistant produced a confident description of a Meta measurement product for ad lift testing. That is not us. It is not anything, as far as we can tell.
We designed the measurement to look for our name in the answer text rather than asking the model to score itself. That is the right call for did a competitor get listed. It is not sufficient for is this my product, and this run is the proof: a bare name match scored a hit on an answer describing something else entirely.
So brand prompts are excluded from the score and kept as raw text for a human to read. The alternative — counting that as recognition — would have made this page state something false about its own subject.
What this measurement is not
- The assistant is not searching. We measure what the model answers from its own knowledge, which is what most people see.
- Training data freezes at a date. Storelift is new; not appearing is not yet known, not bad.
- One measurement is not evidence. Answers vary between runs. What is worth anything is the trend over time.
- The parsing is not done by a model. We search the answer text for known names. Asking a model to grade its own output is an opinion, not a measurement.
The same limits are printed inside the product, on the screen that runs this measurement for your app.
Frequently asked
Why publish a zero?
Because the product's one real claim is that every number in it was measured. A visibility page that only appeared once the number turned favourable would be an advertisement, and the honest version of this measurement is the reason anyone should believe the ones we run for their app.
Does this mean AI visibility does not matter yet?
It means it is early, and that early is when the measurement is cheap. Store search still carries most installs, but the share moving to assistant answers is the reason this is measured at all — and a tool nobody names is a tool nobody installs from that surface.
How often is this rerun?
Monthly. Model knowledge does not change weekly, so running it more often produces noise and cost rather than information. The date at the top is the date of the last run.
See this for your own app
Paste your store link, add your keywords. Rank, the apps above you and demand are measured every night.
Start free →