ai-search
Checks whether ChatGPT, Perplexity, Gemini and Google's AI answers name and cite a business for the questions its buyers ask, scores that over the answers actually observed, finds why the winners win, fixes what's fixable on existing pages and listings, and monitors the result. Works for online businesses and, with city-specific question sets that include Google Maps' Ask feature, for local ones. Use when someone asks whether AI assistants recommend them, why a competitor gets recommended instead, or how to get cited.
boringmarketer · Static checks on file
Installation
Read the source and dependencies before running a command. Copying does not install or execute anything.
npx skills add boringmarketer/boring-marketing-skills --skill ai-searchAbout this skill
name: ai-search description: Checks whether ChatGPT, Perplexity, Gemini and Google's AI answers name and cite a business for the questions its buyers ask, scores that over the answers actually observed, finds why the winners win, fixes what's fixable on existing pages and listings, and monitors the result. Works for online businesses and, with city-specific question sets that include Google Maps' Ask feature, for local ones. Use when someone asks whether AI assistants recommend them, why a competitor gets recommended instead, or how to get cited. when_to_use: ai search audit, am I cited by ChatGPT, does ChatGPT recommend my business, AI Overviews visibility, share of model, AEO, GEO, generative engine optimization, why does Perplexity recommend my competitor, AI recommends my competitor instead of me, Ask Maps, llms.txt, get cited by AI, AI citation monitoring, answer engine optimization argument-hint: "[brand/domain or nothing] | optimize | monitor" metadata: boring: version: 4.1.1 layer: growth reads: [positioning, keyword-plan, competitors#competitor-profiles, voc, brand-manifest] writes: [] optional-keys: [FIRECRAWL_API_KEY, EXA_API_KEY] enhances_with: - toolset: web-scrape gains: "fetches the winning pages in full to read their answer blocks, dates and schema"
AI Search (AEO/GEO)
A growing share of buyers never see a homepage. They see a paragraph an AI assistant wrote about the category, with a few sources cited, or a short list of three to five local businesses with a reason for each. This skill measures whether the business is in that paragraph, diagnoses why the winners are, fixes what's fixable on pages and listings the business already has, and tracks the result. Done means an audit another person could re-run with the same questions, a fix plan ordered by citation leverage, and an honest account of what was observed and what was a proxy. The method is a floor. A better path is welcome with a one-line reason. MONITOR's zero-question contract and the measurement rules below don't flex.
Context
Only if ./brand/, ./brands/ or MARKETING.md exists (nothing to load otherwise), run:
python3 ${CLAUDE_SKILL_DIR}/../start-here/scripts/brand_context.py --for ai-search
(${CLAUDE_SKILL_DIR} is this skill's folder; outside Claude Code use the folder this file was read from, e.g. .agents/skills/ai-search. If the script or python3 is unavailable, read the brand folder directly: ./brands/<slug>/ when ./brands/active names one, else ./brand/; then MARKETING.md, preferences.md and the files in reads:. If none exist, proceed standalone and ask only for the domain and one sentence on what it sells.)
Load references/aeo-facts.md before any mode (check its header; refresh if stale). Load references/aeo-judgment.md before diagnosis and before any OPTIMIZE recommendation. For a local business, also load references/local-ai-facts.md (check its header) and references/local-prompt-sets.md.
Is this a local business? If it serves customers in places (a trade, a clinic, a shop, a restaurant), run the Local branch wherever it appears below. Otherwise skip it.
Modes
| Trigger | Mode |
|---|---|
| Default: a domain, a brand name, or nothing | AUDIT |
| "optimize", "fix", "llms.txt", "make us citable" | OPTIMIZE |
| "monitor", "what changed", or any scheduled or non-interactive run | MONITOR |
MONITOR asks zero questions. OPTIMIZE without a prior audit runs a compressed AUDIT first. Quick Mode: a narrow ask ("are we cited for this one question?") gets a single capture and a one-paragraph answer, no files unless asked.
Measurement rules (all modes)
These are what make an audit worth re-running:
- Sample, don't sweep. 10 to 15 questions is the useful range. Freeze them verbatim in the audit; MONITOR repeats exactly that list.
- Count over what you observed. Report "named in 7 of 12 observed answers", never "58% visible" over questions you attempted. A question whose answer you couldn't capture is NOT OBSERVED, never ABSENT.
- Four states per answer and entity: CITED (its own site is a source), MENTIONED (named, but its site isn't a source, so others tell its story), ABSENT (the answer was observed and it isn't there), NOT OBSERVED.
- Answers vary run to run. Repeated runs of the same prompt share only a minority of their sources (figures in
references/aeo-facts.md). Run the top questions twice or across engines, and treat one run as a sample. - Claims are about the answers collected, not about "what ChatGPT thinks". Say which engine, which date, which location.
- A failed capture is never a finding. A timeout, an out-of-credit error or a page that wouldn't load is NOT OBSERVED. One audit tool once counted errors as gaps and told well-known brands they were missing from answers they were in. Never write "you're invisible" over a run with failed or unobserved captures, and never claim a site is blocked or unreadable because your own fetch failed.
Mode 1: AUDIT
Phase 1. Build the question set
If keyword-plan.md exists, start from its commercial and comparison families. Rewrite each keyword as the question a buyer would type into an assistant, because the phrasing changes which sources get retrieved. "crm for freelancers" and "what's the best CRM for a solo freelancer?" pull different citation sets; keep both when they plausibly diverge.
Keep the set on one topic when you can. Fifteen questions on the topic the business is committing to beat fifteen spread across everything it sells, because the result has to pick a fix. Include the AI-native forms: the long, conversational questions with near-zero Google volume (from keyword-plan.md, voc.md, call notes, reviews and forums). They often decide who gets recommended.
If not, derive from positioning.md across four kinds, skewed toward where the money is:
- Category: "best for " (