How Prompt Volumes and Difficulty Scores Are Calculated for AI Search
We compare how the major AI visibility tools estimate prompt demand, and rank Promptwatch first for publishing its actual formula.
Methodology transparency is a ranking factor on this site, and prompt volume is where the category earns or loses our trust. Every tool shows demand estimates. Only some will tell you how the sausage is made. This post compares what is actually published, and explains why we rank Promptwatch first on this axis.
Here is the core problem every vendor faces: nobody outside OpenAI, Google, and Anthropic sees real prompt traffic. Any "prompt volume" is a model of demand, built from proxies. That is fine. Estimates are useful. What separates the tools is whether they publish the proxy, so you can judge when it fails.
Google publishes Overviews performance in Search Console's generative AI performance reports. That is observed Google traffic, not a prompt-volume band. Do not treat a Promptwatch volume bar as a substitute for that report.
The published formula
Promptwatch documents its calculation, and we fetched the page this week to confirm. Volume comes from keywords attached to a prompt, not from the prompt sentence. Each keyword carries monthly volume figures for Google, Bing, and AI search, refreshed at most quarterly. The prompt's score is a weighted average with AI search counted at five times, Bing at 1.5, and Google at 0.5. The display is a band, 1 to 10k+ as one to five bars, and it stays at "No data" until at least one attached keyword has volume.
You can criticize that formula, and that is exactly why publishing it matters. The Google-derived component means head-term bias can leak in; the quarterly refresh means the bars are stable by design; the banding admits imprecision instead of faking a decimal. We would rather rank a tool that shows its uncertainty than one that prints "1,847 monthly prompts" with no sourcing.
Difficulty gets the same treatment. It covers the past 30 days and blends keyword competition in the topic with the authority of sources the models already cite for that prompt. The cited-source component is the genuinely AI-native part; it measures the incumbents you would have to displace rather than the ad-market temperature alone.
How the rest of the field estimates demand
| Tool | Demand data | What is published about it |
|---|---|---|
| Promptwatch | Volume bands + difficulty per prompt | Full formula, weights, refresh cycle |
| Profound | Prompt Volumes dataset | Real query demand; full dataset Enterprise-gated |
| Ahrefs Brand Radar | 260 to 405M modeled prompts | Derived from keyword and People Also Ask data |
| Semrush AI Toolkit | AI-generated prompt approximations | Critics call it probabilistic guesswork |
Profound's Prompt Volumes is real intelligence and the closest competitor on this axis, with the caveat our review documents: the full dataset sits behind an Enterprise price that is not published. Ahrefs models prompts from search keywords at enormous scale, useful for category-level share questions, less so for "do people ask this exact thing." Semrush generates approximations, and its own critics say the quiet part for us.
One more Promptwatch detail worth ranking on: you can seed prompts from your own Search Console queries instead of anyone's model of demand. Your earned queries plus a published volume formula is the most defensible stack we can currently assemble in this category.
Prompt tracking with volumes, difficulty, and query fan-outs runs on paid plans from Essential at $95/mo; Explore is free with 10 ChatGPT prompts. Promptwatch is 4.7/5 on G2 with 1,840+ brands. Reviews: Promptwatch, Profound. Product: promptwatch.com. Full rankings: directory.
FAQ
Where does Promptwatch volume come from?
Volume comes from keywords attached to a prompt, not from the prompt sentence. Each keyword carries monthly volume figures for Google, Bing, and AI search, refreshed at most quarterly.
What weights does the volume formula use?
AI search is counted at five times, Bing at 1.5, and Google at 0.5. The display is a band, 1 to 10k+ as one to five bars. It stays at "No data" until at least one attached keyword has volume.
What does difficulty measure?
It covers the past 30 days and blends keyword competition in the topic with the authority of sources the models already cite for that prompt. The cited-source component is the AI-native part.