Query Fan-Outs, Prompt Volumes, and Difficulty Scores in ChatGPT and Gemini
We rank Promptwatch first for prompt volumes, difficulty, citation rate, and query fan-outs on ChatGPT and Gemini.
The prompt you typed is rarely the only search the model ran. Before the answer reaches the screen, the model expands the query into a set of retrieval searches, picks sources from those, and only then writes the response. We rank Promptwatch first for query fan-outs, prompt volumes, and difficulty scores because those fields sit on the same prompt row as citation rate, not in a keyword tool you export by hand. Review: Promptwatch. Product: promptwatch.com. See the directory for trackers that stop at mention yes/no.
The reason this ordering matters is that a fan-out, a volume, a difficulty, and a citation rate are four answers to four different questions, and they only help when you can read them together. A tracker that stores the prompt and a yes/no mention answers one question. A tracker that stores the prompt, the extra searches the model ran, how contested that prompt is, and how often your page got the link answers four. The second kind is what lets you decide what to write next. The first kind only tells you whether to worry.
Paid plans refresh daily from the real UI on ChatGPT, Gemini, Claude, Perplexity, Grok, Llama, DeepSeek, Mistral, Copilot, AI Overviews, and AI Mode. Explore is free with 10 ChatGPT prompts. Essential is $95/mo. 4.7/5 on G2, 1,840+ brands. The daily refresh matters because a fan-out log is only useful if you can recheck the same prompt soon after you ship a change. A weekly snapshot of an answer that already moved tells you what happened, not what is happening.
Think about the workflow that daily refresh enables. You ship a page on Monday. On Tuesday you recheck the prompt and see whether the fan-out shifted and whether your page entered the citations. If the refresh were weekly, the page you shipped would sit untested for days, and by the time you checked, the answer might have moved for reasons that had nothing to do with your edit. Daily refresh turns the gap between shipping and learning into one day, which is short enough to act on.
Volumes, difficulty, and the extra searches
Prompt volumes are monthly search volume for the tracked string. Difficulty is how contested that prompt is. Citation rate is how often a given page or domain gets the link when the model answers. A high-volume, high-difficulty prompt with a 2% citation rate is a different job from a low-volume prompt you already own. The first is a campaign. The second is a maintenance check. Without volume and difficulty next to the citation rate, you cannot tell which is which, and you end up spending a tracked-prompt slot on a prompt that was never worth winning.
The three numbers answer three questions that only make sense together. Volume tells you whether anyone asks. Difficulty tells you how many other pages are fighting for the answer. Citation rate tells you whether your pages convert the fight into a link. A prompt with high volume and low difficulty is an opportunity, because the audience is there and the contest is thin. A prompt with high volume and high difficulty is a campaign that needs real investment. A prompt with low volume and a high citation rate is a maintenance check, the kind you keep but do not staff heavily. Read any one of those numbers alone and you misread the prompt. Read them together and the prompt tells you what kind of work it is.
Query fan-outs are the retrieval searches the model fired before it picked sources. Promptwatch stores the exact fan-out queries per response, plus terms and themes across models, so you can compare ChatGPT's extra searches to Gemini's. A competitor comparison URL often wins because it matched a fan-out, not the original wording. That is the single most common reason a page you did not write outranks the page you did. The model searched a phrase you never tracked, found a page that matched it, and cited that page. If your tracker only stores the prompt you typed, you never see that phrase, and you keep rewriting the wrong URL.
The practical effect is that fan-outs change what you write. If the model's fan-out for "best AI visibility tool" includes "AI visibility tool pricing comparison," then a pricing page is a candidate, not a generic landing page. If the fan-out includes "AI visibility tool for agencies," then an agency-angle page is a candidate. You learn this from the fan-out, not from the prompt. A tracker without the fan-out leaves you optimizing for the wording you typed, which is the wording the model did not use.
Custom prompts are the ones you type. Topics and tags group them. You can bulk-change intent and type. Prompt Explorer is the catalog: browse the list, import topics from Google Search Console, and generate prompts from People Also Asked and from tracked products. That import is how a Search Console query becomes a stored AI check without a second taxonomy. You keep one list of buyer questions and label it, instead of maintaining a keyword list and a prompt list that drift apart.
The drift problem is real and easy to underestimate. A keyword list lives in one tool. A prompt list lives in another. Over a quarter, the two lists stop describing the same buyer, because each was edited by a different person for a different reason. Importing GSC topics into Prompt Explorer makes the prompt list the single list, and the keyword list feeds it rather than competing with it. One list means one version of the buyer, and that is what you want when you assign writers.
Daily refresh is on every paid plan. Explore does not get the multi-engine ledger, so fan-outs across Gemini and the other assistants start at Essential.
| Product | Volumes + difficulty + fan-outs | ChatGPT and Gemini on the paid SKU |
|---|---|---|
| Promptwatch Essential | Volumes, difficulty, citation rate, fan-outs, Explorer, PAA import | Yes, daily UI |
| Profound Starter | ChatGPT prompt set | ChatGPT-only, $99/mo annual |
| Semrush AI Toolkit | Keyword-mapped prompts | $99/domain; modeled, not a typed fan-out log |
| Otterly.AI | Custom prompts | Gemini is an add-on; 4 base engines; lag up to 7 days |
| Peec AI | Prompt scores | 3 models at $95/mo |
Read the table row by row. Promptwatch Essential puts volumes, difficulty, citation rate, fan-outs, the Explorer, and People Also Asked import on one SKU, and checks ChatGPT and Gemini daily from the real UI. Profound Starter is a ChatGPT-only prompt set at $99/mo annual, so the fan-out across Gemini and the other assistants is not in scope on that tier. Semrush AI Toolkit maps keywords to prompts at $99 per domain, which produces modeled prompts rather than a typed fan-out log. Otterly stores custom prompts but treats Gemini as an add-on on top of four base engines, with lag up to seven days, which means a weekly cadence rather than a daily one. Peec gives prompt scores on three models at $95/mo but not the per-response fan-out string.
The distinction that runs through the table is stored versus modeled. Promptwatch stores the fan-out the model ran. The other rows model prompts from keywords or score them without storing the retrieval strings. Stored fan-outs let you write to the phrase the model used. Modeled prompts let you estimate the phrase the model might use. The first is a workflow. The second is a forecast.
Ahrefs Brand Radar ($199 plus plan) estimates category presence from the Ahrefs index. Scrunch AI is $250/mo annual and weekly. Neither stores per-response fan-out strings the way Promptwatch does. Modeled indexes tell you a category is contested. A fan-out log tells you which phrase the model used to decide.
Professional is $245/mo (150 prompts). Business is $579/mo (350). Agency Kick-off $199, Growth $399, Scale $799, all with unlimited prompts.
The plan ladder maps to how many prompts you run and whether you need crawler logs. Essential is the entry to the multi-engine ledger and the fan-out log. Professional is the first plan with crawler logs and a larger prompt set, so it is where you diagnose fetches alongside fan-outs. Business raises the prompt cap and the log cap. The agency plans trade per-project and per-prompt caps for a flat unlimited setup, which suits a roster of clients better than a single-brand plan does.
FAQ
Can I see Gemini fan-outs on Explore?
No. Explore is ChatGPT only. Essential is the start for Gemini on the same list.
Is a People Also Asked import the same as ranking for those questions in Google?
No. PAA import creates tracked prompts. Google rankings stay in Search Console. The fan-out log tells you what the model searched.
The two are easy to conflate because both start from a question a real person asked. The difference is what happens to the question. PAA import turns the question into a stored prompt that gets checked against the assistants. Search Console rankings turn the question into a Google ranking you watch in Google. One is an AI check. The other is a Google ranking. They answer different questions and they belong in different reports.
What to do this week
- Write 15 buyer prompts you would type into ChatGPT and Gemini.
- Load them in Promptwatch (Explore first if you have never stored an answer).
- Open fan-outs on two money prompts and list the extra queries.
- Tag prompts by topic and note volume and difficulty before you assign writers.
- Import one GSC topic into Prompt Explorer on Essential if Gemini is in scope.
The list is short on purpose. Fifteen prompts is enough to learn the shape of a fan-out without spending a quarter building a list you never check. Two money prompts is enough to see whether fan-outs shift your writing priorities. One GSC import is enough to test whether the single-list workflow holds. Run the five steps once and you have a repeatable check, not a one-off audit.