Methodology v7 · last reviewed May 2026
How I decide whether your page isready to be cited.
Every audit is the output of 14 signals on your page, scored against the way 6 AI engines actually pick what to cite — built on Google's own AEO guide and the latest 54-experiment meta-analysis from May 2026.
What we measure01
Readiness, not citations.
I don't query ChatGPT to ask "is your brand cited?" — that's a different tool. I score your page on the signals each engine uses to decide what's quotable: answer positioning, entity density, freshness, crawler access, format diversity, E-E-A-T. If you score high, you're ready to be cited. If you score low, here's why.
02The pipeline
Six steps. One audit.
01
Discover
Parse robots.txt and sitemap; detect paywalls and JS-heavy rendering. No llms.txt check — Google has confirmed it's not used.
02
Render
Crawl 5 pages via real Chrome (Cloudflare Browser Rendering). Average crawl: 8–14 seconds.
03
Extract
Pull markdown, headings, entities, schema, dates, outbound citations. Classify page intent (pricing / how-to / docs / etc.).
04
Score
14 weighted dimensions, 0–10 each. Hybrid AI-analyst + rule-based scoring. Per-engine re-weighting for ChatGPT, Claude, Gemini, Perplexity.
05
Fixes
8–12 recommendations, sorted by impact ÷ effort. Each fix names which engine it moves.
06
Ship
Persist + render the report. Shareable, exportable, re-runnable with a single click.
14 dimensions03
The fourteen signals I score.
Each dimension scores 0–10, normalized to 0–100 in the overall. Weights are the default; per-engine scores re-weight based on what that specific engine prefers. Re-balanced May 2026 using Google's published guidance + Cyrus Shepard's 54-experiment Zyppy meta-analysis.
01Semantic completeness16%
Does the page cover the full sub-topic graph a user would expect? Pages scoring >8.5/10 here are 4.2× more likely to be cited (Wellows, 2026).
02Search rank + fan-out · NEW · May 202612%
The strongest empirical signal in 2026. Top-10 organic results contain 38% of AI Overview citations; 47% come from below rank #5. AI fans out 5–15 sub-queries per question and pulls from whatever ranks for each.
03Entity knowledge12%
Count and disambiguate named entities. Pages with 15+ connected entities are 4.8× more likely to be selected (Wellows).
04Quotability12%
AI engines extract 134–167-word atomic chunks (Wellows 2026, updated from the 50–150 range used through 2024). I find every passage that could be lifted verbatim and flag the gaps.
05Answer positioning10%
Mirror the buyer's question in your H1, then answer it in 1–2 sentences before expanding. AI extracts from the first ~80 words.
06Multi-modal richness · NEW · May 20268%
Pages combining text + images + video + structured data see +239–317% selection rates vs. text-only (Wellows).
07Outline quality6%
Clean heading hierarchy is the 9th-ranked factor in Shepard's May 2026 meta-analysis. Hierarchy validity, question-form ratio, heading-to-word density.
08Original data points · NEW · May 20266%
Pages with 3+ unique data points (original surveys, internal benchmarks) are 4× more likely to be cited. Google's own guide explicitly contrasts commodity content with non-commodity, expert-driven pages.
09Crawlability6%
Test allow/deny for OAI-Searchbot, GPTBot, ClaudeBot, PerplexityBot, Google-Extended. I deliberately don't check llms.txt — Google says it's not used; Shepard ranks it 2.0/10.
10Schema stack4%
Google: structured data isn't required for AI Overviews. Shepard ranks schema #20/23 with a "typically small" effect. Supplementary, not load-bearing.
11Topical cluster3%
Orphan pages with no internal links score poorly — AI sees no topic territory. I map internal link graph and anchor concentration.
12E-E-A-T2%
Experience, Expertise, Authoritativeness, Trustworthiness. 96% of cited content carries verified E-E-A-T signals (Wellows).
13Performance2%
LCP, JS ratio, render mode. JS-heavy pages get discounted by engines that don't run JS.
14Freshness1%
23% of cited content is <30 days old. AIO heavily discounts pricing pages older than 90 days.
—Page intent · pre-score—
Before scoring, I classify the page: pricing / comparison / how-to / definition / listicle / landing / product / docs / blog / homepage. Every fix from there is intent-tailored.
046 engines
Six engines. Six rule sets.
Only 11% of cited domains overlap between ChatGPT and Perplexity. Google AIO and AI Mode cite the same URLs just 13.7% of the time. Optimizing for one ≠ optimizing for all. Each card shows that engine's top weighted dimensions.
Google AI Overviews
48–60% of searches
- Top weightsSearch rank + fan-out · Semantic completeness · Quotability
- BiasTop-10 organic still drives 38% of citations. But 47% come from below rank #5 — answer-shaped pages outperform rankers.
ChatGPT (Search)
OAI-Searchbot grounding
- Top weightsEntity knowledge · Answer positioning · Freshness
- BiasHeavy weight on entity disambiguation (sameAs, Wikidata). Cites <30-day content at 3.2× the rate of older sources.
Claude
Web search + citations
- Top weightsOutline quality · E-E-A-T · Original data
- BiasHeavy E-E-A-T preference. Author bylines, named customers, real testimonials. Outbound citations to .gov / .edu lift scores meaningfully.
Perplexity
Live RAG
- Top weightsOutbound authority · Multi-modal richness · Search rank
- BiasCites Reddit at +45% of baseline. Off-page social context matters. Pages with comparison tables outperform pure prose.
Gemini
Google AI Mode
- Top weightsMulti-modal richness · Schema · Entity knowledge
- BiasMost multi-modal-aware of the engines. Pages combining text + images + video + schema see +239–317% selection (Wellows).
Bing / Copilot
Bingbot + ChatGPT layer
- Top weightsCrawlability · Semantic completeness · Topical cluster
- BiasStrict on robots.txt. Bingbot misconfig = invisible. Beyond crawl, weights overlap with both Google AIO and ChatGPT.
AEO snake oil05
Myths I'm ignoring (and you should too).
Six tactics that get sold as AEO requirements. I refuse to score for them. Citations below each.
llms.txt files
Google: "you don't need to create new machine readable files, AI text files, markup, or Markdown to appear in generative AI search." Shepard's May 2026 meta-analysis: dead last, 2.0/10, "no credible evidence of impact." Build the page, not the metadata file.
AI-specific content rewriting
Google: "you don't need to write in a specific way just for generative AI search. AI systems understand synonyms and general meaning." Writing for AI alone violates Google's scaled content abuse policy.
Tiny content chunking
Google: "there's no requirement to break your content into tiny pieces for AI to better understand it." Optimize for humans; AI follows.
Inauthentic mentions
Google: "Seeking inauthentic 'mentions' across the web isn't as helpful as it might seem." Real audience signal beats astroturfing.
Special AEO schema
Google: "there's no special schema.org markup you need to add." Schema ranks #20/23 with a typically small effect (Zyppy, 2026). Useful for rich results. Not load-bearing for AEO.
Domain authority hunting
Link-based DA correlation with AI citations: r=0.18 (weak). Spend the budget on original data instead — that has 4× citation lift.
06Questions
Common questions.
Initial weights from public research (Google's AEO guide, Cyrus Shepard's May 2026 Zyppy meta-analysis, Wellows' ranking-factor breakdown). Re-calibrated monthly using delta data from users who re-run audits after shipping fixes — I measure which dimension changes actually moved per-engine scores.