ChatGPT now serves 800 million weekly active users, Perplexity processed 780 million queries in a single month, and 37% of consumers now start product research with AI rather than Google. Yet most websites have no way to measure their visibility in any of these channels.
AEO scoring changes that. When you run an AEO audit, you get a single number — 0 to 100 — that measures whether AI agents can crawl, understand, and cite your site. This article documents every component: the 16 technical checks, the 6 LLM evaluation dimensions, and the 50/50 blended model that produces the final grade.
What Is an AEO Score?
An AEO score is a 0–100 measure of how well AI agents can crawl, understand, and cite a website. It blends 16 technical checks (50%) with an LLM evaluation across 6 content dimensions (50%) into a single A–F letter grade.
Why AEO Scoring Matters Now
AI search is not a future trend — it is the current state of how information is consumed. Google AI Overviews now appear in roughly 16% of all search queries, and where they appear, organic click-through rates drop by 61%. If your site is not optimized for AI citation, you are losing visibility that traditional SEO cannot recover.
The challenge has always been measurement. Traditional SEO tools track keyword rankings and backlinks. Neither tells you whether an AI agent can read your page, extract a citeable passage, or understand what your site is about. AEO scoring fills that gap with a defined, transparent methodology.
How the Score Is Calculated
The AEO score has two equal components: a deterministic foundational score (50%) and an AI agent evaluation score (50%). Both are normalized to 0–100, averaged, and mapped to a letter grade. The LLM that powers the agent evaluation uses temperature=0 for reproducibility. If the LLM call fails, the system falls back to deterministic-only scoring.
Part 1: The Deterministic Score (50%)
The deterministic half runs 16 binary checks against every crawled page. Each check asks a yes/no question about a technical or structural property. Points are awarded for passing checks, totaled across all crawled pages, and normalized to 0–100.
| Check | What It Tests | Why It Matters for AI |
|---|---|---|
| Title tag | Is a non-empty <title> present? | AI agents extract the title as the primary label for a page |
| Meta description | Is a meta description present? | Summaries help agents pre-classify page intent |
| Open Graph tags | Are og:title and og:description set? | OG metadata is widely consumed by crawlers and feed readers |
| Single H1 | Does the page have exactly one H1? | Multiple H1s signal unclear document structure to parsers |
| Content structure | Is there a logical H2/H3 hierarchy? | Heading hierarchy lets agents decompose a page into named sections |
| Structured data (JSON-LD) | Is any JSON-LD present? | Machine-readable markup is the clearest signal of page type |
| Schema.org types | Are recognized types declared? | Named types (Article, FAQPage, Product) improve classification accuracy |
| Canonical URL | Is a canonical tag present? | Prevents duplicate content from diluting citation signals |
| llms.txt | Is a valid /llms.txt accessible? | The emerging standard for AI-readable site overviews |
| robots.txt accessible | Does robots.txt load without error? | AI crawlers check robots.txt before fetching any page |
| AI-specific meta tags | Are noai / noimageai tags absent? | These tags explicitly block AI ingestion of the page |
| Sitemap available | Is a sitemap.xml accessible? | Sitemaps help crawlers discover all pages efficiently |
| Internal linking depth | Are internal links well-distributed? | Shallow link depth isolates pages from sitewide context |
| Text content length | Does the page have sufficient readable text? | Thin pages provide insufficient content for agents to extract |
| Mobile viewport | Is a viewport meta tag set? | Signals the page is rendered for real users, not scrapers |
| RSS/Atom feed | Is a feed available? | Feeds signal a content-publishing site and aid freshness tracking |
Part 2: The Agent Evaluation Score (50%)
The second half comes from an LLM that reads your top 5 content-richest pages and scores them across 6 dimensions using explicit 0–5 rubrics with behavioral anchors. Each dimension score maps to 0–100 (score × 20). The 6 are averaged to produce the agent evaluation score.
Pages are selected for LLM evaluation by a richness ranking: text length, JSON-LD presence, published/modified dates, heading count, and author attribution. The LLM evaluates the most substantive pages — typically your cornerstone articles, guides, or landing pages.
| Dimension | What the Agent Evaluates | Scoring Anchor (5 = best) |
|---|---|---|
| Answer Readiness | Is there a direct answer in the opening paragraph, before any setup? | 5 = answer appears in sentence 1–2; 1 = answer buried or absent |
| Quotability | Can a clean 40–60 word passage be extracted verbatim as a citation? | 5 = multiple clean, self-contained passages; 1 = all prose requires heavy summarization |
| Evidence Density | Are there statistics, data points, named sources, and specific examples? | 5 = multiple cited statistics with named sources; 1 = all claims are vague or unsupported |
| Content Depth | Is there enough substance to fully answer the topic without other sources? | 5 = comprehensive, authoritative, no gaps; 1 = shallow or partial coverage |
| Freshness | Does the content signal it is current — dates, recent references, updated data? | 5 = explicit recent dates, current references; 1 = undated or stale signals |
| Structural Clarity | Can the page be decomposed into readable, well-labeled sections? | 5 = clear H2/H3 hierarchy, logical flow, no content hidden in JS; 1 = unstructured or layout-dependent |
The Letter Grade Scale
| Grade | Score Range | What It Means |
|---|---|---|
| A+ | 97–100 | Exceptional AI readiness — optimized for citation |
| A | 93–96 | Strong AI readiness — minor gaps only |
| A− | 90–92 | Good — a few structural or metadata issues remain |
| B+ | 87–89 | Above average — notable gaps in content depth or schema |
| B | 83–86 | Average — passes most checks but LLM scores are holding it back |
| B− | 80–82 | Below average — structural issues affecting AI parsing |
| C+ | 77–79 | Weak — AI agents struggle with key sections |
| C | 73–76 | Poor — multiple failing checks and low content quality signals |
| C− | 70–72 | Poor — significant technical and content gaps |
| D | 60–69 | Very poor — AI agents cannot reliably read your site |
| F | <60 | Failing — your site is largely invisible to AI engines |
Who Uses AEO Scoring and How
Yuki, SEO lead at a 25-person B2B SaaS company
Yuki's company has well-ranked blog content but isn't appearing in ChatGPT answers about their category. Running an AEO audit reveals their JSON-LD is missing on all product pages (foundational score: 54) and their content depth dimension scores 2/5 — pages are feature-heavy but don't open with direct answers. She fixes: adding Article schema to all blog posts and rewriting three cornerstone pieces to lead with a direct answer. The next audit brings the foundational score to 82.
Marcus, content manager at a regional news publisher
Marcus's site has hundreds of substantive articles but an AEO score of 48 — well into failing. The reason: an overly broad robots.txt was blocking GPTBot alongside other scrapers. One targeted fix — adding an explicit allow rule for AI crawlers — immediately unlocks AI indexing across 400+ articles. This is one of the most common issues uncovered by AEO audits: 62% of top news sites block at least one AI training bot, often without knowing it.
Dev team at a 60-person e-commerce company
Their product pages score well on foundational signals (74) but land in the C range on agent evaluation because product descriptions lack comparative content worth quoting. The audit's prioritized fix list surfaces 8 category pages as the highest-leverage rewrite targets — pages where adding a structured buying guide above the product grid would lift both the evidence density and quotability scores.
What the Score Doesn't Measure
The AEO score measures technical and content-quality signals within your direct control. It does not measure domain authority, backlink profiles, whether your brand is mentioned elsewhere on the web, or your position in any AI engine's current training data. A high AEO score is a necessary condition for AI citation — it removes the barriers — but it doesn't guarantee specific results in specific queries.
Run a free AEO audit at aeo-check.vercel.app — get your score across all 16 checks and 6 agent dimensions in under 2 minutes.
Is AEO the same as SEO?
No. Traditional SEO is optimized for keyword-based indexing — Google's algorithm scores pages on backlinks, keyword relevance, and PageRank. AEO is optimized for AI comprehension — how well an LLM can extract, understand, and cite your content. Many SEO best practices overlap, but AEO adds a set of signals (structured data depth, content quotability, LLM-parseable structure) that standard SEO tools don't measure.
What is a good AEO score?
Scores of 80+ (B− or above) indicate your site is readable by AI agents with addressable gaps. Scores of 90+ (A range) indicate strong AI readiness. Most sites that haven't optimized for AEO score in the 55–72 range on first audit.
How is an AEO score different from a PageSpeed score?
PageSpeed measures how fast your pages load. AEO scoring measures whether AI agents can read, parse, and cite your content. A fast page with thin content and no structured data will score high on PageSpeed and low on AEO.
How often should I run an AEO audit?
Re-audit after any significant content change, after deploying structured data updates, or after modifying robots.txt. For active publishing sites, a monthly audit cadence catches regressions early. Quarterly audits are a reasonable baseline for most teams.
Can I improve my AEO score without changing my content?
Yes — the 16 deterministic checks are entirely technical. Adding JSON-LD, a meta description, a canonical tag, or correcting your robots.txt can meaningfully improve your foundational score without touching your page copy. The agent evaluation score (50% of total) does require content changes.
Does a high AEO score guarantee AI citations?
No. A high score means AI agents can read and parse your site correctly. Whether you appear in a given AI response also depends on whether your content is the best available answer to the specific query and the AI model's training and retrieval logic — factors outside this audit's scope.
Sources
- ChatGPT reaches 800M weekly active users — TechCrunch, Oct 2025
- Perplexity: 780M queries in May 2025 — TechCrunch, Jun 2025
- 37% of consumers start searches with AI — Search Engine Land, Jan 2026
- AI Overviews in ~16% of queries — Semrush, Dec 2025
- Organic CTR drops 61% where AI Overviews appear — Seer Interactive, Sep 2025
- 62% of top news sites block at least one AI training bot — BuzzStream, Dec 2025
- 41% of pages use JSON-LD — HTTP Archive Web Almanac 2024