target-prompts-discovery
Skill0 starsChanged 3 months agoMITView on GitHub
Discover and curate the target prompts the company should win in AI answers. Research external demand (community questions, competitor targeting, search surfaces), filter by product fit, and maintain the tracked prompt panel in strategy/target-prompts/.
In this repository12 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- llms.txtllms.txt
- channel-presenceSkill
- content-gap-analysisSkill
- content-generationSkill
- content-reworkSkill
- generate-hero-imageSkill
- listings-checkSkill
- retroSkill
- setupSkill
- site-readinessSkill
- visibility-measureSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 20
Start from the problems the product solves (listed in strategy/company-profile.md) and research what people actually ask when they have …
- Distinctive: line 49
Every prompt carries its source and evidence: six months later you must know why it earned its slot.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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.
Files
Target Prompts Discovery
Builds and maintains the panel of target prompts: the natural-language questions buyers ask AI engines (ChatGPT, Claude, Perplexity, Gemini) that the company should be cited in. The panel defines what the entire GEO system measures and optimizes for. /visibility-measure measures visibility for every active prompt, and /content-gap-analysis diagnoses the ones being lost.
Everything below is guidance with its reasoning, not a checklist. When the situation argues for a different call, make it and record the reasoning in the change log. The two hard constraints are the demand-first principle and the human approval gate.
Core principle: demand-first, not visibility-first
Prompts are discovered from external demand (what real people ask when they have a problem the product solves), never from your own visibility data. Choosing prompts from where you already rank is circular: it can only optimize what you already have, and stays blind exactly where the gaps are biggest. Your own data validates and refines candidates; it does not originate them.
Process
1. Research the market's questions
Start from the problems the product solves (listed in strategy/company-profile.md) and research what people actually ask when they have each problem:
- Community mining: real questions posted on Reddit, Hacker News, Stack Overflow, and social platforms. Use the Xpoz MCP tools to search Reddit, Twitter/X, TikTok, and Instagram for question phrasings at scale (
getRedditPostsByKeywords,getTwitterPostsByKeywords, subreddit and comment searches). Community questions are the closest thing to ground-truth buyer demand, and this is the discovery source most GEO programs skip because it is hard to search without a tool. - Competitor reverse-engineering: competitors' blog titles, landing pages, and FAQs reveal which queries they believe convert. Multiple competitors targeting the same query is validated demand.
- Search-demand surfaces: Google autocomplete, People Also Ask, and related searches for category terms. These reflect what everyone searches, not what you rank for.
- Asking the engines themselves: prompt AI engines for what users ask when looking for tools in the category. Synthetic, but captures natural AI-chat phrasing, which differs from search-keyword phrasing. Run these conversations through
scripts/ask-claude.pyandscripts/ask-openai.py(single message, or multi-turn via--thread); each run's transcript is saved underrecordings/chats/<engine>/for mining afterwards. The search queries recorded in the latestrecordings/target-prompt-visibility-runs/<run-folder>/visibility-summary.mdadditionally show the exact fan-out queries engines run for prompts already on the panel: a real sample of engine phrasing.
2. Filter by relevance and winnability
Raw demand is huge; the panel holds 30-50 active intents. Each candidate must pass: Can the product genuinely serve this intent? Is the persona one the company sells to? Is the query winnable, or owned by dominant established players on terms the company does not compete on?
For candidates that survive the filter, probe the engines with the candidate prompt itself (the ask scripts) to capture who is cited for it today. This serves two purposes: the citation landscape is the best winnability evidence a proposal can carry (an answer assembled from listicles is winnable; one dominated by entrenched documentation may not be), and it becomes the pre-panel baseline; without it, the first measurable before/after starts a full cycle late. Probes are single-sample per engine: enough to read the citation landscape, too noisy to treat one citation (or its absence) as fact. When a decision hinges on a borderline result, probe again rather than argue from one sample. In baselines and proposals, distinguish cited (a company URL in the answer's citations) from mentioned (named in the answer text only), per the panel header's definitions. Skip the probe when it genuinely adds nothing (e.g. a branded prompt), and say so in the proposal.
3. Validate against your own data (supporting role only)
- Demand validation: Google Search Console impressions confirm search volume where connected (
integrations.search_consoleinconfig.json), regardless of where you rank. - Blind-spot detection: what paying users actually do with the product (product analytics, support questions, sales calls) reveals intents that competitor content and communities would never surface. Ask the operator what usage signals exist.
- Sensor feedback: the latest visibility snapshot and deltas summary are the maintenance input. Prompts flagged as dead are the retirement candidates, and the engine fan-out queries (plus AI-referral landing pages where GA4 is connected) show the phrasings and intents through which engines already route people to you. These refine wording, inform retirement, and seed hypotheses, but per the demand-first principle they do not originate prompts on their own; a hypothesis they raise still needs external demand behind it.
4. Propose, review, update the panel
Present proposed candidates with evidence, existing prompts recommended for retirement, and coverage analysis across buckets. Candidates become active only after human approval: the panel steers everything downstream, so a bad prompt misdirects the whole system. Approved changes are applied to the panel and recorded in the change log with their reasoning.
Panel rules
- One record per intent, with a single canonical phrasing. Paraphrases of an existing intent are rejected; research shows semantically similar wordings produce the same visibility. Variants earn a slot only when semantically distant and high-value.
- Coverage across the four buckets: branded, category ("best X"), comparison ("alternatives to X"), problem-solution ("how do I..."). Add buckets (e.g. pricing) when the category calls for them.
- 30-50 active intents. Curation over growth; a bloated panel makes every downstream run slower and blurrier.
- Prompts are retired, never deleted; retired prompts keep their measurement history.
- Every prompt carries its source and evidence: six months later you must know why it earned its slot.
- Value is the agent's judgment of the asker's closeness to purchase. The anchor that keeps runs comparable: high, intents a buyer asks right before choosing a tool; medium, intents that shape the shortlist; low, broad informational presence-builders. Rate the intent, not the bucket: buckets classify phrasing, value rates the buyer.
End product
strategy/target-prompts/target-prompts.md: the updated panel, single source of truth for downstream skills.strategy/target-prompts/change-log.md: a dated change-log entry per run: what was proposed, the evidence, what was approved or retired, and bucket coverage.memory/competitors/: competitor knowledge the run produced anyway (reverse-engineering, baseline probes' who-is-cited landscape) deposited per that index's conventions, so the next session starts oriented instead of re-researching.
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
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.

