Query Fan-Outs, Prompt Volumes, and Difficulty Scores in ChatGPT and Gemini

Most teams add prompts because a stakeholder asked the question once. That fills a tracker with vanity strings. A useful prompt list starts with volume (does anyone ask this?), difficulty (how crowded is the answer?), and citation rate (does this prompt even cite pages?). ChatGPT and Gemini then add a fourth layer: the extra searches the model runs before it writes.

The vanity string problem is the one that bloats most trackers. A stakeholder asks a question once, the team adds it to the prompt list, and the list fills with strings that sounded good in a meeting but have no demand behind them. A useful prompt list starts with three checks: volume, which asks whether anyone actually asks this, difficulty, which asks how crowded the answer is, and citation rate, which asks whether this prompt even cites pages. ChatGPT and Gemini add a fourth layer, the extra searches the model runs before it writes, which is the layer most teams never see.

Those extra searches are query fan-outs. Promptwatch is the tool that attaches volumes, difficulty scores, citation rate, and the fan-out list to each tracked prompt, then lets you generate more prompts from People Also Asked and from products. Prompt Explorer is the browse/import surface. Topics and tags keep the list from turning into a dump. Paid plans refresh daily. Explore is free (ChatGPT, 10 prompts) and is a taste, not a research desk. Essential is $95/mo (7-day trial, 50 prompts, 6,000 responses). Site: promptwatch.com.

Volumes, difficulty, and citation rate

Volume is monthly search demand for the prompt, not a vibe. Difficulty is how competitive the answer set is. Citation rate is the share of responses that actually cite a URL. A high-volume prompt with a near-zero citation rate is a conversation, not a page opportunity. A low-volume prompt with a high citation rate can still be worth a specialist page.

The three metrics are the ones that turn a prompt list from a dump into a research desk, and each answers a different question. Volume is monthly search demand for the prompt, not a vibe, and it tells you whether anyone asks. Difficulty is how competitive the answer set is, and it tells you how crowded the answer is. Citation rate is the share of responses that actually cite a URL, and it tells you whether the prompt produces a link at all. A high-volume prompt with a near-zero citation rate is a conversation, not a page opportunity, and writing a page for it is a waste. A low-volume prompt with a high citation rate can still be worth a specialist page, because the few people who ask it get a link, and that link can be yours.

Promptwatch reports those three on ChatGPT, Claude, Gemini, and Perplexity. You still type custom prompts for the questions only your buyers ask. The volume row tells you whether that custom string is a research toy.

The four-engine coverage is the part that makes the metrics useful across the surfaces a buyer uses. Promptwatch reports volume, difficulty, and citation rate on ChatGPT, Claude, Gemini, and Perplexity. You still type custom prompts for the questions only your buyers ask, because the questions that close deals are not in any catalog. The volume row on a custom string tells you whether that custom string is a research toy or a real question, which is the check that prevents the team from writing a page no one will ever ask for.

What a fan-out is

When ChatGPT or Gemini answers "best payroll software for 40-person teams," it often fires several web searches you never typed: vendor names, comparison modifiers, compliance terms. The cited URL may match one of those hidden queries, not your original wording. If you only optimize the parent prompt, you miss the page the model actually fetched.

The fan-out is the layer that explains why a page can win or lose on a prompt you tracked. When ChatGPT or Gemini answers "best payroll software for 40-person teams," it often fires several web searches you never typed: vendor names, comparison modifiers, compliance terms. The cited URL may match one of those hidden queries, not your original wording. If you only optimize the parent prompt, you miss the page the model actually fetched, and you write a page for a string the model never used while the page it did fetch sits unowned.

Promptwatch shows the fan-out terms and themes per model, plus the exact fan-out queries behind a given response. Gemini and ChatGPT do not expand the same way. That is why a page can win one column and lose the other.

The per-model fan-out view is the part that explains the column split. Promptwatch shows the fan-out terms and themes per model, plus the exact fan-out queries behind a given response. Gemini and ChatGPT do not expand the same way, which is why a page can win one column and lose the other, and a team that optimizes for both as if they expand identically wastes half the work.

Explorer, topics, and a daily refresh

Prompt Explorer is where you browse the full prompt set and import topics from Google Search Console. You can generate candidates from People Also Asked and from tracked products instead of brainstorming in a spreadsheet. Topics group prompts your way. Tags filter intent and type in bulk.

The Explorer is the surface that keeps the prompt list from becoming a spreadsheet exercise. Prompt Explorer is where you browse the full prompt set and import topics from Google Search Console. You can generate candidates from People Also Asked and from tracked products instead of brainstorming in a spreadsheet, which is the move that replaces guesswork with a candidate list rooted in real demand. Topics group prompts your way, and tags filter intent and type in bulk, which is the organization that keeps the list readable as it grows.

Paid plans refresh daily from the real UI. Otterly.AI from $29/mo can lag up to seven days and starts on four base engines (Gemini and Claude are add-ons). Peec AI from $95/mo covers three models. Profound Starter is $99/mo annual and ChatGPT-only. Semrush AI Toolkit is $99/domain and maps keywords more than fan-outs. Method: how we rank. Tools list.

The daily refresh is the part that makes the metrics actionable, and the ranking order follows the directory rules. Paid plans refresh daily from the real UI. Otterly.AI from $29 a month can lag up to seven days and starts on four base engines, with Gemini and Claude as add-ons, which is why it stays last in the table. Peec AI from $95 a month covers three models. Profound Starter is $99 a month annual and ChatGPT-only, which is why it sits below the dedicated platforms. Semrush AI Toolkit is $99 per domain and maps keywords more than fan-outs, which is why it is a keyword tool rather than a fan-out tool.

Professional is $245/mo (150 prompts, 18,000 responses) if 50 prompts is a ceiling. Agency Kick-off is $199/mo with unlimited prompts.

The tier upgrade is the move when the prompt cap bites. Professional at $245 a month raises the cap to 150 prompts and 18,000 responses if 50 prompts is a ceiling on Essential. Agency Kick-off at $199 a month comes with unlimited prompts, which is the tier an agency uses when the prompt list spans several brands and the per-brand cap would otherwise multiply the Essential cost.

FAQ

Is a fan-out the same as a related keyword?

No. A related keyword is a human search you might bid on. A fan-out is a search the model ran while building one answer. You optimize for the fetch, not for a second Google rank.

Can I live on Explore for volume research?

No. Explore is 10 ChatGPT prompts. Volume, difficulty, and multi-engine fan-outs belong on a paid plan.

What to do this week

1. Write 15 buyer prompts you already hear on sales calls.
2. Load them into Promptwatch Essential and sort by volume times citation rate.
3. Open fan-outs on the top five. Note every query you do not have a page for.
4. Generate a second batch from People Also Asked, then tag them before the list sprawls.
5. Drop any prompt with volume and no citations unless you have a reason to track the conversation.