🧭 AEO Master Playbook
How to make content that gets cited by AI (ChatGPT / Perplexity / Google AI Overviews / Gemini) without getting flagged or demonetized. Consolidated from 6 research streams — Claude ×3 (citation mechanics, penalties, YouTube practitioners) + Codex (build system) + Gemini (source verification). 2025–2026 sources.
The one-line strategy: win with brand-unique, first-party, expert content — and never ship templated/thin content at scale. The same move is both the growth lever and the penalty shield. Everything below is detail on that sentence.
1. What actually gets you cited (ranked, with evidence)
- Brand mentions across the web — the #1 signal. Ahrefs: mentions correlate with AI citations at r=0.664 vs backlinks 0.218. Get mentioned where LLMs already look (Reddit, Quora, YouTube, industry sites) — not broad PR. (3 of 5 streams independently flagged this #1.)
- Original first-party data & statistics. Princeton GEO study: adding real stats lifts AI visibility +41%. Yext: 86% of AI citations come from brand-managed sources, 44% from first-party sites. Your moat = real job costs, permit notes, comps, before/after, reviews.
- First-hand experience + a named expert (E-E-A-T). Google (May 2026): first-hand experience is the differentiator AI can't fake; the March 2026 core update explicitly rewards "non-commodity" content over summaries.
- Be indexed + answer the specific sub-question. Only 38% of AI-cited pages rank in the organic top-10 — niche authorities win. Google's "query fan-out" rewards self-contained passages answering one sub-topic.
- Structure for extraction. Answer-first 40–80 words up top; lists & tables (≈80% of ChatGPT citations vs 29% of Google organic); short visible FAQs.
- Optimize per-engine + stay fresh. Only ~11–14% source overlap across ChatGPT/Perplexity/AIO → diversify. "Last updated" is now a retrieval signal.
Why it's worth it: AI visitors convert at ~23× the rate of normal organic (Digital Bloom), even though 93% of AI sessions are zero-click (Semrush) — so "share of model" matters more than clicks.
2. The brand-unique moat (the durable part)
As the web fills with generic AI content, the only thing that stays citable is what only you have: first-party data + named human expertise on your own domain. For Blue Line that's real estimate ranges, permit observations, anonymized jobs, before/after specifics, material choices, and local constraints — injected into every page. A page with nothing first-party shouldn't ship.
3. What gets you penalized / demonetized — and the truth about "AI"
AI content is explicitly allowed by Google, YouTube, and AdSense. The trigger is never "it's AI" — it's low value at scale. What actually gets hit:
- Scaled content abuse — mass low-value pages, even "human-reviewed," made to rank. Google's March 2026 spam update enforced this in under 20 hours via SpamBrain.
- Thin / templated / duplicate / doorway pages — your exact past exposure (the old ~0.88–0.95-similar city pages where only the city changed).
- Site reputation abuse (parasite SEO) — now algorithmic.
- YouTube "Inauthentic Content" (July 15 2025) — demonetizes templated / superficial-difference / low-effort videos. Matters for your lesson-video plan: needs real human value + AI-disclosure label.
Myths killed: no "AI-detector penalty" exists; third-party detectors are unreliable (>20% false positives); "humanizer" tools are a trap; backlinks are no longer the #1 citation signal; ranking #1 ≠ getting cited.
4. Pre-publish checklist — run on EVERY page/video (the shield)
- ≥ 3 page-specific first-party facts
- ≥ 2 cited sources for any regulatory / price / procedural claim
- Named author + reviewer and a visible "Last updated" date
- Answer-first 40–80 words at the top; lists/tables where they fit
- Uniqueness: 5-gram Jaccard
< 0.35 vs sibling pages (warn 0.25); headings not ≥80% identical to a sibling
- Word floor: ≥ 900 (service/local) · ≥ 1,200 (legal/financial)
- AI-tell < 6 (existing content-vetter); zero banned phrases
- Schema matches visible content only — no fake reviews; FAQPage only for FAQs the user can see
- Video: real human commentary/value + AI-synthetic-media disclosure
If any line fails → don't deploy. Codex spec'd these as automated build gates (below).
5. Do this in your stack (Codex's build plan, condensed)
build_course.py: replace hardcoded "DE 8" → total=len(M); add author / reviewed_by / sources / legal / first_party_facts fields + a validate_lesson() that fails the build if a legal lesson lacks disclaimer/sources; run the vetter before deploy.- Site factory: per-page evidence packets (jobs/permits/costs/reviews JSON); refuse to build any page with < 3 first-party facts; add
uniqueness-gate (fail 0.35); CI blocks the Cloudflare deploy on any gate failure.
- Machine-readability: ship
llms.txt + sitemap.xml (lastmod from data date) — useful index, not a ranking trick (Google + studies: llms.txt has zero citation effect).
6. Spanish / local edge (Latinos con Negocios)
Spanish AEO competition is far thinner = early-citation opportunity, and Spanish AI Overviews amplify YouTube (~50%). But geo-anchor every page to Washington / your county or you get homogenized into "global Spanish" and lose the local citation. (This doubles as citation fuel.)
Key sources
Google Search Central — AI features & "succeeding in AI search" (May 2026) · Google spam policies / scaled content abuse (Mar 2026) · YouTube Inauthentic Content policy (Jul 15 2025) · Ahrefs GEO study (Apr 2026, brand-mentions r=0.664; 38% top-10 gap) · Princeton GEO (stats +41%) · Yext (86% brand-managed citations) · Semrush (93% zero-click) · Profound (lists/tables ≈80% ChatGPT) · Digital Bloom (23× conversion). Full per-claim source tables in inbox/aeo-deep-research/findings-*.md.