agentleFS
frontmost-aifrontmost-ai/aeo-skills/skills/question-drafting/SKILL.md

question-drafting

Skill1 starsChanged 3 days agoMITView on GitHub

Draft, audit, or expand the buyer questions to track in AI apps (AEO or GEO prompt tracking): branded and unbranded questions, intent clusters, and real buyer phrasing. Use for "which questions should I track in ChatGPT", a review of a question set, or new questions for a Frontmost AI project. Adds questions only after an explicit go-ahead. Works from what the person supplies and the web, or from a Frontmost AI project.

The file

Show its rules and text

We split this skill into the rules it gives an agent

15 rules. About 0 are common, 12 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈12

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use this when someone wants the buyer questions to track in AI apps, wants the current set reviewed, or asks whether their questions are …

    line 8
  2. Very specific:

    If a Frontmost AI tool returns setup_required, give the person the url from the result.

    line 14
  3. Distinctive:

    Before you draft, read references/examples.md.

    line 18
  4. Distinctive:

    Do not reword, pause or remove anything in this step.

    line 44
  5. Distinctive:

    Use, in this order. Items 1 to 3 need a Frontmost AI project:

    line 48
  6. Distinctive:

    Always include the plain head term in list format ("What are the best <category> tools right now?").

    line 63
  7. Distinctive:

    Write the way a buyer would type it.

    line 71
  8. Distinctive:

    Never ask the AI to pick among a fixed list of rivals in an unbranded question.

    line 75
  9. Distinctive:

    No year in the question, no leading claims, keep it under about 25 words.

    line 76
  10. Distinctive:

    No accidental near-duplicates.

    line 77
  11. Very specific:

    Check every draft against the examples in references/examples.md before presenting it.

    line 79
  12. Distinctive:

    Do not add anything yet. A question about whether you can add them is not permission.

    line 90
  13. Very specific:

    Add in one edit_questions batch with a topic for each.

    line 94
  14. Distinctive:

    Look for what the new questions reveal beyond the score: brand confusion with a similarly named company, trust gaps, new cited pages that …

    line 105
  15. Distinctive:

    If unbranded visibility is still zero, say so plainly and point to the leading measures: cited pages that list the brand, and answers that …

    line 106

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Question drafting

Use this when someone wants the buyer questions to track in AI apps, wants the current set reviewed, or asks whether their questions are good. The goal is a question set that measures discovery honestly and gives a realistic chance of a first unbranded mention.

With or without Frontmost AI

  • With a Frontmost AI project, do all six steps. The saved answers show which questions are weak, and the cited pages show how real people ask.
  • With no Frontmost AI project, do steps 2 to 4 with what the person supplies and what you can read on the web. Skip steps 1, 5, and 6. Return the table, and say that no AI answers are measured yet.
  • If a Frontmost AI tool returns setup_required, give the person the url from the result. If it returns plan_required, say that the data needs a paid plan. In both cases, continue with the no-project path.

Read the examples first

Before you draft, read references/examples.md. If the file is not available, stop and say so.

What the research says (keep these in mind throughout)

  • The same prompt almost never returns the same brand list twice. How often a brand appears is stable. Its rank is not. (SparkToro and Gumshoe, Jan 2026)
  • One run of one prompt is close to uninformative. Use several wordings per intent and read results over a rolling 2 to 4 week window. (University of St. Gallen preprint, Apr 2026)
  • Rewording barely changes results while the meaning stays close. Format matters more: list-style and short keyword prompts surface roughly 20 to 25 percent more brands. A changed qualifier (budget, location, team size, named rival) is a new intent. (Peec AI study, Jun 2026, vendor-sponsored)
  • Judge clusters of prompts, not single prompts, and keep branded and unbranded separate. (Ahrefs guide)
  • Public data on what people ask AI apps is thin. Say so. Never present a drafted question as a known real query.

Step 1: Read the current set

This step needs a Frontmost AI project. Call list_projects, then get_questions for the project (and get_scan with questions: "unbranded" for the unbranded split). For each existing question, note:

  • Scope: branded (names the project's brand) or unbranded.
  • Buying stage: early research, comparing options, or ready to buy.
  • Format: list ("best X"), short keyword, "what do X recommend", problem-first, persona, comparison, or alternatives-to.
  • Qualifiers: budget, team size, location, platform, named rival.

Flag weak questions, with evidence from saved answers:

  • Ambiguous: answers drift into another product category. Check which business each answer named first.
  • Unwinnable: the question asks the AI to choose among a closed list of named rivals, so the brand can never be named. Useful for rival intelligence, but it adds guaranteed zeros to the unbranded count.
  • Off-target: aimed at a buyer the business does not sell to.
  • Monoculture: most questions share one format or one stage.

Do not reword, pause or remove anything in this step.

Step 2: Gather real phrasing

Use, in this order. Items 1 to 3 need a Frontmost AI project:

  1. get_sources with questions: "unbranded": titles of the cited list articles and the cited Reddit threads. Thread titles are real people asking.
  2. get_rivals: the most-named rivals, for "alternatives to" and comparison questions.
  3. get_queries: the web searches ChatGPT ran. Treat as weak evidence, because they mirror the questions already tracked.
  4. Anything the person supplies: Search Console questions, support chats, sales calls, forum threads.
  5. The business's own differentiators (price point, integrations, method). Verify each one before using it. Do not claim "cheapest" or similar unless checked.

Step 3: Draft

Defaults, unless the person says otherwise:

  • Ratio: about 3 unbranded to 1 branded.
  • Stage mix for unbranded: roughly 25 percent early research, 50 percent comparing options, 25 percent ready to buy.
  • Organize by intent cluster. Give important intents 2 to 4 wordings in different formats.
  • Always include the plain head term in list format ("What are the best <category> tools right now?"). It is the format that surfaces the most brands.
  • Include the category's own vocabulary (for AI visibility: AEO, GEO).
  • One "alternatives to <rival>" question per top rival not already covered.
  • A few questions built on verified differentiators. These are where a first mention is most plausible, but do not promise it.
  • Branded set: what it is and who it is for, one comparison per top rival, a trust question ("is it legit, are there reviews"), "alternatives to <brand>", and pricing.

Wording rules:

  • Write the way a buyer would type it. Plain, first person where natural.
  • Unbranded questions never contain the brand name.
  • State the category unambiguously. If a phrase could mean another product category, add the words that pin it down.
  • One intent and at most one qualifier per question.
  • Never ask the AI to pick among a fixed list of rivals in an unbranded question.
  • No year in the question, no leading claims, keep it under about 25 words.
  • No accidental near-duplicates. A deliberate cluster variant must differ in format, not just in a word or two.

Check every draft against the examples in references/examples.md before presenting it.

Step 4: Present for approval

Show one table: question, scope, topic, stage, format, cluster, and what it is based on (cited thread, cited article, rival data, differentiator, or judgment). Then state:

  • With a project: how many questions the project and the plan will have after the change.
  • That the unbranded denominator changes, so day-over-day comparisons break on the day they start.
  • Which drafts rest on evidence and which are judgment.
  • Any weak existing questions found in step 1, as a separate recommendation.

Do not add anything yet. A question about whether you can add them is not permission.

Step 5: Add, only on an explicit go-ahead

  • Add in one edit_questions batch with a topic for each.
  • Rewording an existing question pauses the old wording and starts new history. Only do it when asked, and say what it costs.
  • Tell the person to ask again after the next scan. Then confirm that each new question was sorted as branded or unbranded correctly. Fix a wrong one with branded_override.
  • Report the new totals and when the first results arrive.

Step 6: Reading the first results

  • Report total, branded and unbranded visibility together, each with counts.
  • Compare the original questions separately from the new ones for at least a week. A drop in branded visibility after adding branded questions is usually the mix changing, not a decline.
  • Compare formats separately (list against list, persona against persona).
  • One day is a first read. Draw conclusions from a rolling 2 to 4 week window.
  • Look for what the new questions reveal beyond the score: brand confusion with a similarly named company, trust gaps, new cited pages that do not list the brand.
  • If unbranded visibility is still zero, say so plainly and point to the leading measures: cited pages that list the brand, and answers that cite those pages.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.