“What’s the monthly search volume of a prompt? Most likely, the number is one.” That is Conductor’s line, on the GEO Blueprint podcast, from a company whose business is search measurement. Tim Soulo of Ahrefs went further on the Ahrefs Podcast: “Anyone claiming accurate volume data for prompts is essentially lying to you.” Two of the most credible measurement voices in the field, on the record, calling the metric fiction.
The industry heard “the number is one” and shipped dashboards with four significant digits.
Why is there no prompt volume?
Prompts do not repeat, and a volume metric counts repetition. A search keyword is a compressed token that millions of people type the same way, which is what made keyword volume a real number for twenty years. A prompt is a paragraph: constraints, context, follow-ups, phrasing that varies per person and per session. Ethan Smith at Surfer Academy: “the tail is larger and longer in LLMs than in SEO… you need to answer questions, many of which have never been searched for before.” A question asked once has a volume of one. Conductor already told you that.
Panel data is the industry’s fallback, and its own advocates hedge it. Smith’s caveat in the same session: prompt volume can be estimated from panels, “directionally accurate but who knows what the absolute number is.” Dentsu, in the GEO Blueprint corpus, put the platform side in four words: “impression-level data doesn’t exist.” So the pipeline behind a vendor’s prompt-volume column is a small opt-in panel, extrapolated, with no ground truth to calibrate against. Directional at best. A database column either way.
What is the divide-by-25 heuristic?
Take a keyword’s search volume, divide by 25, and treat the result as your prompt-volume estimate. That is Ethan Smith’s heuristic from his podcast circuit this year, and it comes with a built-in fraud detector: the closer a vendor’s claimed prompt volume sits to the raw search volume, the more wrong it is. If the prompt number on the dashboard looks like the keyword number from Ahrefs, the dashboard copied the keyword number and relabeled the column.
A napkin estimate that admits it is a napkin estimate beats false precision. Divide-by-25 gets you a defensible order of magnitude for prioritization: enough to decide whether a topic deserves a page, which is the whole job the metric had.
What should you use for question research instead?
Sources you already own, converted into questions. Smith’s recipe from Lenny’s Podcast: take your own keyword data, organic and paid, plus the terms your competitors pay for, then run an LLM pass to turn keywords into the questions buyers ask. For the tail that never shows up in any keyword tool, mine conversations: sales calls and support tickets first, then the Reddit threads where buyers describe the problem in their own words. None of this requires a prompt-volume subscription. All of it is grounded in demand you can verify.
Scale down, too. Smith’s guidance puts 10 to 15 tracked prompts as sufficient for a small brand, a band Graphite itself sits in. The 1,000-prompt enterprise tracker is theater built on the fake volume column: multiply an invented frequency by thousands of prompts and the total is still invented.
Then measure the thing that is measurable: how often you appear when those questions get asked. Appearance rate over repeated runs holds up where a single-run rank does not, and the disciplined version is share of voice in AI answers. Our free scan runs 5 questions your buyers ask and shows the answers verbatim. The paid audit runs each question 7 times per engine and reports ranges with linked evidence. You will find no volume column in our reports. We cannot verify one, so we do not print one.
Pick your five questions from a sales call, ask an engine, and watch what it cites. That is question research with a sample size you can trust: start with the free pass.
Sources
- Ethan Smith at Surfer Academy: the ROI of improving AI visibility
- The Ahrefs Podcast, Tim Soulo with Ethan Smith (prompt volume data quote, divide-by-25 heuristic)
- The GEO Blueprint podcast: Conductor and Dentsu episodes (prompt volume and impression data quotes as stated on air)