Get Mentioned By AI

get-mentioned-by-ai

Context load

Run rainmaker context --check first. It prints what exists, what is stale, and exits 1 if anything this skill requires is missing.

Then read, in this order:

1. context/business.md in full. If absent, stop: "No business context. Run rainmaker audit, then the know-my-buyer skill."
2. context/voice.md if this skill writes prose. If absent, stop and say so.
3. context/glossary.md if this skill names products, features or competitors.
4. data/strategy.json if this skill reads or writes strategy.
5. Only the data/ files listed in this skill's Consumes table. Never crawl or call an API the core already covers.

If strategy.json.context_hash does not match the current hash of context/business.md, say exactly:

"Business context was edited after the strategy was written. Re-run know-my-buyer, or run rainmaker context --sync to accept the prose as authoritative."

Then stop.

If context/business.md carries confidence: stub, continue, and stamp every output with: "Built on a stub context. Nothing in it came from a buyer. Run know-my-buyer to replace it."

Consumes

Produces

data/snapshots/<ts>/citations.json.

Probe design

12 prompts minimum, in buyer language rather than brand language: 3 from the category phrasing in messaging.category, 3 commercial head queries from clusters, 3 comparison forms against the top 3 named competitors, 3 from the top-impression non-branded GSC queries. With no strategy, take all 12 from icp_hint plus glossary terms.

Run per engine configured via .env (OPENAI_API_KEY, ANTHROPIC_API_KEY, PERPLEXITY_API_KEY) and per market in config.geographies. Any absent key is skipped and named in the confidence section; zero keys skips the whole probe set and this skill still runs its offline half.

Refresh the prompt set whenever say-it-their-way changes buyer_language. Prompt drift silently invalidates a trend line, since a stale prompt set is answering a question buyers stopped asking.

Reading the result

1. Never report an aggregate without its decomposition. A score move localises to one engine, one market or one prompt, and the aggregate hides it. Report per engine and per market always.
2. Record methodology_version on every scan. Never compare across a version boundary; that is the classic false alarm.
3. Mine the citations, not just the mentions. Which domains and URLs do the engines cite for these prompts? Hand that list to get-cited-elsewhere.
4. Distinguish invisible from visible-but-mispositioned. Being cited as the wrong category needs different work from not being cited at all.
5. Every citation claim carries: "Assistant answers are non-deterministic. Treat this as a sample of N probes on , not a ranking." Confidence is capped at 0.5. Never write a verified ledger event from a probe; citation moves append opened and regressed only, and regressed itself requires two consecutive monthly misses at the same methodology version.

Offline half

Schema coverage by tier: which tier 0 and 1 pages lack Product, Offer, FAQPage, Organization, BreadcrumbList. Extractability: tier 1 pages with fewer than 3 standalone claim sentences. llms.txt presence and whether it covers tier 0 and 1 URLs. Crawler access: does robots.txt block GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot, reported as a business consequence rather than a value judgement.

Decision rules

• No probe set below 12 prompts.
• Confidence on any citation finding never exceeds 0.5.
• A single missed scan is a suspicion, not a regression.

Output

Per engine, per market: visibility rate, top cited sources with citation counts, and one concrete content action per gap. The non-determinism sentence appears once per section, not once per row, so it is not lost in repetition but is never omitted either.

Done when

Every citation claim carries its probe date, engine, market and the non-determinism sentence; the offline half ran regardless of API keys; and citation sources are exported in a form get-cited-elsewhere can consume directly.