Scoring System
Grade Classifications
Your AEO Score (0-100) is interpreted using the same five bands as the analyzer:
Reports show your grade and where the next band starts, for example "Grade (Good starts at 75)". They never set a target score, a gap to close or a deadline: the grade describes where the page stands today.
6 Analysis Dimensions
Your AEO Score is calculated from 6 weighted dimensions, each evaluating different aspects of AI optimization:
What it measures: Infrastructure, security, crawlability, indexation status, robots.txt and sitemap, canonical targets, pages that say "not found" but answer 200, snippet controls (noindex, nosnippet), HTTPS
Why it matters: Without a solid technical foundation, AI systems cannot properly crawl and index your content
What it measures: HTML hierarchy, Schema.org structured data and whether it matches the visible page, semantic HTML5 tags, heading structure, question-based content
Why it matters: Clear structure helps AI systems understand content relationships and context
What it measures: E-E-A-T signals (Experience, Expertise, Authoritativeness, Trust), direct answers, sources, freshness, clarity and depth. It also shows Answer Readiness, AI advice on how quotable the page is, which is unproven and not scored.
Why it matters: Clear, sourced, current content is easier for readers to trust and for AI systems to quote correctly
What it measures: Page speed from PageSpeed lab data (one Google test load, not your real visitors): TTFB, First Contentful Paint, Total Blocking Time, Speed Index, JavaScript execution time and render-blocking resources. Without PageSpeed data it is not measured and the other dimensions are re-weighted.
Why it matters: Fast pages serve visitors well and are fetched reliably; speed is not an AI citation threshold
What it measures: Alt text, image formats, sizes and lazy loading, responsive images, figures, labels on SVG graphics, captions on video and transcripts for audio. A page with no media is not measured here.
Why it matters: Screen readers, search engines and text-based tools get the text that comes with your media, not the picture or the sound, and lighter images keep pages fast
What it measures: Which AI crawlers may open your pages (search, user fetch and training are judged separately), hidden instructions aimed at AI assistants, summaries, clear writing, claim attribution and content dating. Several signals here, such as llms.txt and entity links, are unproven: they are shown but not scored.
Why it matters: AI search and assistants can only use pages their crawlers are allowed to open, and clear, attributed content is easier to quote correctly
Score Calculation
Your AEO Score is a weighted average of all dimension scores:
AEO Score = (Technical × 0.15) + (Structural × 0.15) + (Content × 0.25)
+ (Performance × 0.10) + (Media × 0.15) + (AI-Specific × 0.20)
Each dimension score is the weighted average of its checks' scores. A check counts by its importance within the dimension (the same whether it passes or fails) multiplied by how well the evidence supports it, as explained below. Checks that do not apply to your page, or that we could not test, are left out rather than counted as passes or failures.
Every check gives a score from 0 to 100, and partial scores are common. Each check also has its own pass line: a check below it is marked failed and listed as something to fix, with a severity of critical, high, medium or low. Passed or failed, its score counts in the dimension by the check's weight.
When a dimension has nothing to weigh (no check could be measured, or every measured check is unproven), it is shown as not measured, never as 0, and the other dimension weights are scaled up so that they still add to 100%. Without Performance data, for example, Content counts for about 27.8% instead of 25%.
How Checks Are Weighted
Every check is tied to research on search engines and AI answers. How much that research supports a check decides how much it counts in your score:
| Evidence level | Counts as | What it means |
|---|---|---|
| Official | × 1.0 | Stated by the search or AI engine itself |
| Corroborated | × 1.0 | Confirmed by several independent sources or measurements |
| Practitioner advice | × 0.5 | Common advice from practitioners that the engines have not confirmed |
| Unproven | × 0.0 | Plausible but not shown to matter. Labelled "unproven — not scored" in the report and left out of the score |
The multiplier is applied to the check's importance, which is fixed and only compared with the other checks of the same dimension: robots.txt Validation counts 2.5 in Technical, for example, and every AI-Specific check counts 1.0.
An unproven check is still run and shown, so you can decide whether it suits your site, but it never lowers or raises your score, it has no severity, and it is not ranked in "What to do next". Answer Readiness, the AI advice shown with each scan, is one of them.
AI crawler access is judged by purpose
The AI Bot Permissions check reads your robots.txt the way crawlers do and sorts AI crawlers by what they are for. Search crawlers (such as OAI-SearchBot, PerplexityBot, Claude-SearchBot, Bingbot and Googlebot) and user-fetch crawlers (ChatGPT-User, Perplexity-User, Claude-User) decide whether AI search and assistants can open your page: blocking one on the scanned page fails the check. Training crawlers (such as GPTBot, ClaudeBot, CCBot and Google-Extended) are your policy choice: blocking them never fails the check, and it does not hide your page from AI search. Only search and user-fetch access is scored.
Stages
Each check also carries a stage: the step, from crawler to visitor, where a problem gets in the way. The stage is a label in reports, tickets and the MCP results; it never changes a score or the order of "What to do next".
- Crawler access: can search and AI crawlers reach the page and read its content?
- Indexing: will search engines find the page, keep it in their index and be allowed to show it?
- Answer quality: is the content clear and complete enough to be picked to answer a question?
- Trust and citation: does the page show who stands behind it and back up its claims, so it can be trusted and credited as a source?
- Clicks and visits: when people see the page in results or an AI answer, will they click and find a page that works for them?
Rule-Set Versions
The scoring rules change when the evidence changes. Every analysis records the rule set it was scored with (the current one is version 6). A report scored with an earlier rule set says so, with a one-line summary of the latest change; re-scan the page to see it under the current rules. When only some of a site's pages were scanned with an earlier rule set, the site dashboard's notice says how many.
Scores from different rule sets are never compared: a page's history and score changes only use analyses from the same rule set, and the site trend uses each page's latest score on the newest one. Results judged by a rule that no longer applies are left out of rankings and counts until the page is re-scanned.
Improving Your Score
Your site dashboard ranks the fixes for you under "What to do next": worst severity first, then how much the check counts (its importance times its evidence), then how many pages it fails on. Unproven checks are counted in a note, not ranked. Each fix says who usually does it (you, a developer or a writer) and how big it is (quick, medium effort or a larger project), and you can download the list as developer tickets (Markdown or CSV). Use the Priority Matrix below to plan the rest:
- Let AI search crawlers fetch your pages (robots.txt)
- Add missing meta descriptions
- Remove accidental noindex or nosnippet settings
- Fix missing alt text
- Implement Schema.org markup
- Improve E-E-A-T signals
- Restructure content hierarchy
- Answer your customers' real questions on the page
- Serve images in modern formats with width and height
- Add Open Graph tags
- Implement breadcrumb navigation
- Minor CSS optimizations
- Unproven extras such as llms.txt or speakable markup
- Third-party widget issues