Method
AI prompt volume does not exist
Every AI visibility tool ends up printing a number next to each question: "2,400 searches a month". Nobody measures that. Here is where the figure actually comes from, what it allows, and what it does not.
Nicolas Framery Founder of Etalo, SEO and generative visibility consultant
Collected on 16 October 2026. Every source carries its date in the text.
The essentials in thirty seconds
No assistant vendor — not OpenAI, not Anthropic, not Google — publishes how often a question is put to it. Prompt volumes shown by tools are therefore built, in one of two ways: from a consented user panel whose conversations are read and then statistically corrected, or by recycling the Google search volume of a nearby keyword. Both give a useful order of magnitude for ranking questions against each other, and neither gives a number you can defend line by line.
- There is no first-party prompt feed. Scrunch AI writes it in its own help centre: "No AI platform provides a native, first-party prompt feed today".
- A panel does not see everything: it misses mobile, native apps, Safari and browsers that block extensions. What it does see is then extrapolated to the whole population.
- A keyword is not a prompt. Across 10,000 prompts tested in March and April 2026, ChatGPT produced 91% unique queries, sharing only 13% of their words with the original prompt.
- A number without its provenance becomes a measurement within six months. That is the real risk: not the figure, but forgetting how it was built.
- Weighting does not require an absolute volume. A relative index works just as well mathematically — as long as it is never displayed as a count of people.
Why this number cannot be measured
Google search volume exists because Google counts its own queries and resells part of them, aggregated, through its ad platform. The chain is short: the vendor measures, the vendor publishes.
For an assistant, that chain does not exist. Conversations are private, no vendor publishes the aggregate, and there is no ad platform forcing them to. This is not a jealous competitor's claim: Scrunch AI, which sells an AI visibility tool, writes it in its own public documentation.
Anything shown as "prompt volume" is therefore a reconstruction. That does not make it useless — an honest reconstruction beats nothing — but it radically changes what you are entitled to conclude from it.
The two ways this number gets built
Every method on the market reduces to one of the two, sometimes to both mixed together without the interface saying so.
| Method | What it actually observes | What it misses | What it is good for |
|---|---|---|---|
| Consented panel browser extension, clickstream | The conversations of a small group of users who agreed to share them. Commercial-looking prompts are extracted, clustered by theme, then extrapolated to the whole population. | Mobile, native apps, Safari, browsers that block extensions. And the make-up of the panel, which does not look like the population — hence a statistical correction, which is one more assumption. | Comparing themes against each other. The theme level survives extrapolation; the individual line does not. |
| Recycled Google volume nearby keyword | The number of searches on a keyword judged equivalent to the question. It is a real measurement — of something else: people typing into a search engine, not writing to an assistant. | Everything that is not phrased as a keyword, which is most assistant use: writing, summarising, code, multi-turn conversation. | Ranking questions by how much the market cares. That is what Etalo does, and it writes the measured query next to the figure. |
Panel method and coverage limits: Scrunch AI help centre, 2026. The "good for" column is our reading, not the cited vendors'.
Three numbers that sum up the problem
unique queries produced by ChatGPT from 10,000 tested prompts
Profound study, March-April 2026
word overlap between the prompt asked and the queries the assistant actually ran
Same study; 88% for Perplexity
assistant vendors publishing a first-party prompt feed
Scrunch AI help centre, 2026
The 10,000-prompt study is published by an employee of Profound, a vendor that sells prompt volume. It has had no external validation. We cite it because it cuts against the commercial interest of whoever published it: it shows that a keyword does not predict a prompt.
What the 13% actually means
When you ask ChatGPT a question, it does not search your sentence. It breaks it into several queries run in parallel — query fan-out, which Google documents officially for its own generative features. The queries actually run share 13% of their words with your question.
The direct consequence: the Google volume of the keyword "best travel agency" describes neither the demand put to assistants, nor even the queries assistants run when asked that question. It describes a third thing, correlated with both, and that is all.
Does that make Google volume useless for a GEO audit? No. An imperfect correlation is still information when the goal is to rank fifty questions rather than forecast traffic. But it rules out the sentence "this question is asked 2,400 times a month on ChatGPT" — which is exactly what the interfaces let you read.
How to read a prompt volume without being fooled
Five questions to put to any tool that displays such a number — ours included.
Where does the figure come from?
Panel, recycled search volume, or an internal model? If the interface does not say and the documentation does not either, the matter is settled: this number cannot be defended in front of a board.
How old is it?
A volume collected eight months ago in a moving market is worth nothing, and nothing on screen ever says so. Ask for the collection date, not the display date.
On exactly which query?
If the figure comes from search volume, it was collected on a query derived from your question. Which one? Without that line there is no way to judge whether the approximation is reasonable or absurd.
Zero, or no data?
The most expensive confusion. A source that does not know a query has not observed zero demand: it has observed nothing. A tool that shows 0 in both cases drags your averages down for missing data.
Is the figure used to rank or to size?
Ranking fifty questions against each other: acceptable. Deriving visits, revenue or a budget from it: no. That is the one use that turns an approximation into an error.
Weighting without pretending to measure
An unweighted GEO audit treats the fifty questions of a corpus as equals. That is wrong: a question the whole market asks and a niche question are not worth the same to someone allocating a budget. So you have to weight, and so you need a weight per question.
Here is the part almost nobody says: the weight does not have to be an absolute volume. A weighted presence rate is the sum of (weight × citations) divided by the sum of (weight × observations). Multiply every weight by a constant and the ratio does not move. A relative index — "this question weighs three times that one" — therefore weights exactly as well as a search count, without claiming to count anyone.
That is Etalo's approach. The tool derives a search query from each question on its own, collects its demand, and writes the exact query that produced the figure next to it. It never calls that a prompt volume, and a question whose query the source does not know stays without a figure rather than dropping to zero.
Frequently asked questions
Are the prompt volumes shown by tools wrong?
Why not just use Google volume?
Isn't a user panel a real measurement?
How does Etalo get its demand figures?
Does a question with no figure mean nobody cares about it?
Should we wait until real data exists?
Sources
Absence of a first-party prompt feed, panel method, coverage limits and warnings on extrapolation: Scrunch AI, Understanding Prompt Volume in AI Search, help centre, 2026.
The 10,000-prompt study, unique queries and word overlap: Nick Lafferty, Prompt Volume Explained, March-April 2026. The author works at Profound and says so himself.
Query fan-out mechanism: Google's official guide for generative AI features, May 2026.
Read next
- The business law firms AI assistants cite most
When a company asks ChatGPT or Google which law firm to call, which names come back? We put 26 questions to five assistants and counted the citations of seven…
- Why my brand doesn't appear in ChatGPT
You asked the question, and a competitor came out instead of you. Before rewriting anything, work out which of the four causes applies: they share neither the…
- How much does a GEO audit cost
Two of the three most-cited tools no longer show any price at all. Here are the ones still public, what the API calls behind them really cost, and why the gap…
- Getting into Google AI Overviews
This is the generative surface most people see without ever looking for it: it sits above the results, in front of users who have never opened an assistant…
- When AI gets you wrong
A brand that is highly visible in assistant answers, with a wrong price, does not have a visibility problem. It is the one case where a good score is worse…
- Building your question corpus
It is the part of a GEO audit nobody wants to do, and it decides everything else. A badly written corpus produces impeccable figures about questions nobody…
- GEO audit: what it measures, how, and what makes an audit wrong
Public figures on generative search contradict each other: AI Overviews appear on "more than 20%" of searches by one measurement, on 75.24% by another. Both…
- How to get cited by ChatGPT
There is no button. There is a technical door that many sites close without knowing it, three editorial levers that have actually been measured, and a source…
- GEO or SEO: what really changes
Three serious studies measured the same thing — the share of AI citations coming from pages that rank well on Google. They found 16.7%, 52% and a correlation…
- AI visibility tracking tools in 2026
Seven tools, read on the same day from their public pages. Two publish no price at all. And once you divide published prices by the number of answers actually…
- robots.txt and llms.txt: letting AI read your site
It is the one generative-visibility job that can be settled in an afternoon, and the one that makes every other job pointless if you get it wrong. Three files…
- Etalo.ai review
You are looking for reviews of Etalo and you cannot find any. That is normal: the tool is young and has no clients to parade. Rather than manufacture…
- Best Peec.ai alternative
Peec.ai does one thing well: it tracks your visibility in assistant answers, daily, in a dashboard that is easy to read. If that is what you need, it is a good…
- Best Promptwatch alternative
Promptwatch has one real strength few tools have: crawler logs, that is, what AI robots actually come and read on your site. If that is your subject, keep it…
Weighting that says where it comes from
Etalo collects demand for every question on its own, and writes the measured query next to the figure. You can read the line back.