Jonathon's AI Wiki

Zyppy AI Citation Playbook — Fan-out Framework + 7-Step Audit (Cyrus Shepard)

8 min read

Source: raw/newsletter-zyppy-signal-af27fd411a.md (Fan-out Framework), raw/newsletter-zyppy-signal-ee04a24d31.md (7-Step AI Citation Audit Checklist), raw/newsletter-zyppy-signal-13b309c23c.md (Fan-out Query Gap Analyzer Tools) — Cyrus Shepard, Signal by Zyppy. Published Apr-Jun 2026.

The actionable companion to the Zyppy AI Citation Ranking Factors meta-analysis. Where the meta-analysis ranks what correlates with AI citations, this playbook is Shepard’s how: a 5-step Fan-out Framework for finding and covering query expansions, a 7-step pre-publish audit checklist, and two AI query-gap tools he built. Evidence class: practitioner workflow, not a study. It cites the cluster’s empirical findings (Ahrefs, AirOps, Dan Petrovic, Seer) but adds no new data — its value is execution sequencing, not evidence. Rigor is low; actionability is high.

Key Takeaways

  • The #1 predictor of AI-answer inclusion is ranking highly in classical search. Shepard’s supporting stats: Ahrefs — 38% of Google AI Overview citations come from Google’s top 10; AirOps — ChatGPT cited the Google #1 result 43.2% of the time; Semrush — Perplexity answers had ~82% overlap with Google’s top 10. Practical floor: be in the top ~30 for the query, or rank for its fan-outs, to have a shot.
  • The “fan-out myth.” There is no stable, pre-determined list of fan-out queries for a search. Fan-outs are probabilistic and personalized — run the same query 10× on the same engine and get 10 different sets. Don’t chase an exact list; identify the commonalities across many samples and cover those.
  • Grounding queries ≠ fan-out queries (but overlap). Fan-out queries seek new info; grounding queries verify/fact-check existing info. Bing Webmaster Tools exposes grounding queries directly in its AI Performance Report — practically useful despite the technical distinction.
  • You can influence AI answers faster than classical rankings. Citations often move in days, not months — but they’re volatile day-to-day, so track a dataset, not a single check.
  • Don’t build “Godzilla pages” or scaled fan-out pages. Over-covering every fan-out topic backfires now that Google demotes scaled content. Cover the gaps in what you already rank for, with original/first-party material.
  • Extractability beats volume. Per Dan Petrovic’s grounding research (cited by Shepard): a tight ~800-word page can get 50%+ of its content grounded; a 4,000-word page ~13%. Only ~32% of a page’s content gets picked up for Gemini grounding consideration — so front-load and tighten.
  • Being cited pays. Seer Interactive: being cited in Google’s AIO delivers +120% organic clicks per impression (and +41% paid) vs not being cited.

The Fan-out Framework (5 steps)

  1. Start from a keyword topic you already rank for. Best source: Google Search Console. Shepard’s worked example — a title-tag page ranking page-1 for “title tag length” queries but bleeding clicks to AI Overviews.
  2. Gather fan-out queries from multiple sources (there is no single “best” method):
    • QueryFan (queryfan.com) — all-round fan-out generator; works best with a paid OpenAI or Gemini API key (retrieval typically costs “a buck or two”). Optional AlsoAsked API key (Pro) enriches results. Supports “personas” to simulate user profiles.
    • Qforia (iPullRank) — Google-specific; needs a Gemini API key; upload many queries at once; toggle AI Overview vs AI Mode.
    • Non-API options: Dejan’s queryfanout.ai, Otterly’s Query Fan Out Analysis tool.
    • Bing Webmaster Tools → AI Performance Report — real grounding queries for your URLs.
    • Synthetic fan-outs via any LLM — no tool required; use a structured prompt (below).
  3. Determine the most important fan-out topics. A single seed can yield ~400 raw fan-outs — too many. Use an LLM cleanup prompt to consolidate to 20-25 primary keywords + 5-10 secondary. Optionally pull search volume (Ahrefs Keyword Explorer) and run keyword clustering (Shepard uses Keyword Insights) to find intent clusters and coverage gaps vs competitors.
  4. Optimize — update existing pages or create new ones. Prefer updating a page that already ranks and has authority (easier, safer). Align content to the fan-out: put the query (or a close variant) in an H1/H2 and follow it immediately with the answer. Only spin up a new page when the topic can stand on its own and has real search demand.
  5. Measure. Free: Bing Webmaster AI Performance Report (citations per URL + grounding queries) and Google Search Console’s AI features report (impressions in AI Mode/AI Overviews — but it does not reveal the triggering queries). Paid trackers Shepard names without endorsing: Peec, Otterly, Profound, Gumshoe, Ahrefs Brand Radar; SEOTesting estimates LLM click volume.

The 7-Step AI Citation Audit Checklist

  1. Make content accessible to AI engines. Don’t block key crawlers in robots.txt (Googlebot, bingbot, OAI-SearchBot, ChatGPT-User, PerplexityBot, Perplexity-User); check Cloudflare AI Crawl Control. Don’t use display-blocking meta tags or data-nosnippet on important text. Don’t rely on client-side JS for important text (LLMs digest raw HTML; use SSR). Don’t hide text behind tabs/carousels/modals. (Lily Ray’s LLM Content Visibility Scanner tests this.)
  2. Rank for the main query (top ~30 minimum). Pick a query that actually triggers an AI answer with citations.
  3. Find and target fan-out queries via three methods: observed (QueryFan/Qforia/WordLift’s AI Visibility Fan Out), synthetic (Rankability’s tool), and reverse-engineering the AI answer itself. Target the gaps in your coverage, not every fan-out (IntentGaps.com helps).
  4. Choose the right content format for the intent — ranked list for “best places,” comparison page for “X vs Y specs.” Observe what currently ranks.
  5. Match the answer to the query. Strong finding (AirOps/Kevin Indig): pages whose headings closely match the query are cited more. Write titles/headings/answers that mirror the query wording.
  6. Put primary answers at the top. A short, direct summary near the top is more likely to be grounded (only ~32% of a page is retrieved for Gemini grounding).
  7. Make the page easy to cite — five sentence-level moves from the ranking-factors study: Factually Specific, Explicit Phrasing, Cite Sources, Self-Contained Passages, Entity Consistency. Cite sources for facts that need evidence (even if the source is your own study).

Two AI Query-Gap Tools (Pro Templates)

Shepard built two “AI Query Gap Analyzer” tools that automate the manual Fan-out Framework — one in ChatGPT (usable free), one in Claude. Given a primary keyword + URL, they evaluate you and competitors, generate and classify fan-out topics, identify coverage gaps, recommend specific content, and decide new-page-vs-update. Shepard’s self-reported result after running them on a Zyppy page: “overnight we saw an explosion of AI citations in Google,” and Google’s AI answer began recommending Zyppy’s title-tag checker within the answer itself (visible in Bing too). ^[ambiguous] Single-site, self-reported anecdote from the tool’s author — a demo, not evidence.

Implementation

Tool/Service: QueryFan, Qforia, AlsoAsked, Keyword Insights, Bing Webmaster Tools, Google Search Console, Ahrefs. Setup: A Gemini API key (free tier available) or paid OpenAI key unlocks the fan-out generators; billing recommended (queries cost cents). Cost: Fan-out retrieval “a buck or two” per run; clustering/rank tools are paid SaaS; the two Query-Gap tools are gated behind Zyppy Pro Templates. Integration notes: This operationalizes factors 2-6 of the ranking-factors meta-analysis. The synthetic-fan-out and cleanup prompts (below) run in any LLM, so the core loop needs no paid tooling.

Synthetic fan-out prompt (paraphrased from source)

Ask an LLM to act as an SEO/search-intent analyst and generate ~50 fan-out queries for a seed phrase, grouped under: Primary intent, Supporting subtopic, Comparison, Problem/solution, Audience/use-case, and Decision-stage. Constrain to natural search wording, mix short/mid/long-tail, avoid vague/off-topic/redundant queries, one per line, no numbering or commentary. Then run a second “cleanup” prompt that consolidates the raw list into 20-25 primary + 5-10 secondary keywords, stripping topic drift and near-duplicates.

Open Questions

  • How durable are the tool-driven citation gains? Shepard’s “overnight explosion” is a single self-reported case from the tool author; no controlled measurement or decay window.
  • Does the top-30 rank floor hold across engines? The 38% (Ahrefs AIO) / 43.2% (AirOps ChatGPT) / 82% (Semrush Perplexity) stats are from different studies and surfaces; the practical floor is an interpolation.

Try It

  1. Export a GSC query set for one page that ranks page-1 but loses clicks to AI Overviews. That’s your fan-out candidate.
  2. Run the synthetic fan-out prompt in Claude or ChatGPT, then the cleanup prompt, to get 20-25 target queries — no paid tools needed.
  3. Cross-check against Bing Webmaster’s grounding queries for the same URL to ground the synthetic list in observed behavior.
  4. Patch the 2-3 biggest coverage gaps by adding query-matched H2s with the answer immediately below — update the existing page rather than spawning new ones.
  5. Watch citations in Bing + GSC AI reports over 1-2 weeks, not a single check — citations are volatile day-to-day.

Graph View

Backlinks