What Is the AEO Score? How Waggle Rates Your Content for AI Search

Waggle’s per-page AEO score (0–100) rates how easily an AI answer engine — ChatGPT, Perplexity, Google AI Overviews, Gemini, Claude — can lift and quote your content. It is a transparent, deterministic rubric: no AI runs to produce it, every point has a reason, and every weight is grounded in published evidence. This page explains each factor, its weight, and why it’s weighted that way, so nothing reads as invented.

The evidence base

The weights are tiered by how strong the evidence for each factor is:

Tier 1 — causal, peer-reviewed. The Princeton/Georgia Tech study “GEO: Generative Engine Optimization” (Aggarwal et al., KDD 2024, arxiv.org/abs/2311.09735) ran controlled experiments over ~10,000 queries and found three page edits that measurably raised how often content was cited — roughly +30–41% each: adding cited sources, statistics, and quotations. These carry the most weight.
Tier 2 — retrievability, well-supported. Where the answer sits and how extractable it is. About 44% of AI citations are lifted from the first ~30% of a page (Ahrefs), and pages whose headings match the query are cited far more often.
Tier 3 — hygiene, modest/correlational. Freshness (AI-cited pages run only ~26% fresher on average — real but small) and author attribution (an E-E-A-T signal).

The 9 components and their weights

Component	Weight	Tier	Why this weight
Early answer	18	2	A direct, self-contained answer in the opening. ~44% of AI citations come from the first ~30% of a page — the opening is what gets lifted.
Cited sources	17	1	Outbound links to authoritative sources. The single biggest page-level lever for content that isn’t already top-ranked (GEO: up to +115% for a mid-ranked page).
Statistics	16	1	Concrete numbers and data. GEO measured ~+30–41% visibility from adding statistics.
Quotations	12	1	Direct quotes from sources/experts. A GEO top-three lever (~+28–40%).
Readability	9	2	Fluent, scannable writing: concise sentences (~18 words is the cited-content average) plus subheadings and lists so passages are self-contained. GEO fluency lift ~+15–30%.
Question headings	8	2	Headings phrased as the questions readers ask. Pages with query-matching headings are cited markedly more (~41% vs 29%).
Freshness	8	3	Recency of the last update. Real but small (~26% fresher), so weighted low and with generous thresholds — evergreen pages aren’t punished.
Attribution	7	3	A visible author bio / expertise signal (E-E-A-T; GEO authoritative-tone lift ~+10–20%).
FAQ coverage	5	3	Visible on-page Q&A pairs — self-contained, extractable answers. Weighted lightly: it overlaps Question headings, and note this scores visible content, not FAQPage schema (see below).

Total: 100. Each component scores full / half / zero against simple, fixed thresholds, and returns a plain-language note plus a fix hint.

What we deliberately do not reward

Because the evidence says these don’t drive AI citations:

Raw word count / long-form for its own sake — near-zero correlation (~0.04 across 174k pages, Ahrefs); about half of AI-cited pages are under 1,000 words.
Keyword stuffing — zero or negative effect in the GEO study.
A table on every page — tables help comparison/commercial queries, not how-to guides, so we reward a table when present (inside Readability) but never penalise a page for not having one.

What the score can’t see (and we say so)

A page-level score can’t measure off-page factors that, for Google AI Overviews and Perplexity, matter most of all:

Organic ranking — ~76–91% of AI Overview / Perplexity citations are pages already ranking in Google’s top 10 (Ahrefs, 1.9M citations).
Off-page brand mentions — the strongest brand-level signal (unlinked brand mentions, including on YouTube) correlates with AI visibility ~3× better than backlinks (Ahrefs, 75k brands).

So a perfect on-page score is necessary but not sufficient: it decides which of the already-eligible pages gets quoted. Waggle surfaces these as guidance, not as scored points it can’t honestly measure.

Honesty notes on adjacent features

llms.txt is discoverability hygiene, not a proven citation driver. There’s no confirmed evidence a major engine consumes it (Google has said its Search systems don’t read it). Waggle serves it and lists it, but never sells it as a ranking win.
Schema / structured data is correlated with citation but not causally proven; controlled tests show no significant citation lift from adding JSON-LD. It’s legibility hygiene (and Bing confirms it aids understanding), not a citation lever.
FAQ / FAQPage schema — Google retired the FAQ rich result (the SERP dropdown) on May 7, 2026, so the JSON-LD earns no Search snippet, and LLMs read rendered text rather than the markup. Waggle’s FAQ score therefore rates visible on-page Q&A content (which is still extractable by AI), not the schema — and weights it lightly because it overlaps question-style headings. The FAQPage markup remains valid and is still read by some non-Google crawlers, but it is never sold as a Google ranking or rich-result feature.