ai-answer-gap

# AI Answer Gap Turn observed AI-answer gaps into a prioritized research and content backlog. Use an existing `ai-visibility-audit` or collect a dated panel first. Read available product context so proposed topics match what the product can credibly help with. ## Use UnifAPI for live evidence Connect with the `unifapi` skill to discover current schemas, prices and authentication before collecting data. ## Workflow 1. **Select relevant prompts.** Group them by use case and search intent. `/geo/keywords/search-volume` supplies optional demand estimates; an absent estimate is unknown, not evidence that a useful niche prompt has no value. 2. **Check current answers.** Use `/geo/answers` for ChatGPT/Gemini, `/geo/serp` for Google AI Mode, and `/seo/serp` with `include_ai_overview` for Google Search AI Overviews. Keep engines, markets and natural/forced-search settings separate. Preserve exact answers and cited source URLs. 3. **Describe the observed gap.** Distinguish no answer, unsuccessful collection, brand absent from this answer, name-only mention, and cited brand. A sample is not proof of universal absence. Inspect references; top-level `is_target` and unused search results do not by themselves establish a citation. 4. **Research sources.** Open the actual cited pages. `/geo/mentions/top-pages`, `/top-domains`, and `/search` add separate indexed-corpus context. They do not supply the winning URL for a particular live prompt. `/cross-aggregated-metrics` gives group counts, not ownership of your prompt-panel gaps. 5. **Propose a response.** Identify the missing information, evidence or third-party representation. Choose an existing-page update, a new useful page, or an external research lead. Each proposed change needs a concrete source and a plausible benefit to readers. Name-only mentions and existing organic rankings are clues, not proof that formatting is the cause. 6. **Prioritize.** Consider product relevance, evidence of demand, gap quality, available expertise, and effort. Show the inputs and uncertainty. Optional scoring is a planning aid, not a predicted citation lift. 7. **Validate later.** Use the same panel and multiple dated runs after changes. Compare paired successful samples, show completion/answer rates, and avoid attributing every change to the edit. ## Output For each priority, give the exact prompt, engine/configuration, observed brand status, cited competitors or sources, suggested page/action, evidence, effort estimate and a validation plan. Preserve request ids and billed credits. An API failure should appear as missing evidence and never as a content gap. See [the audit methodology](../ai-visibility-audit/references/geo-methodology.md) for coverage/share definitions, citation counting and evidence boundaries. Use `llm-mention-tracking` for budgeted snapshots and changes over time. This skill researches and drafts; it does not publish content or send outreach without authorization.