AEO Score: How We Calculate Your 0-100 Rating

Two sites. One scores 34. The other scores 88. The gap is not taste, not opinion, not some AI in a different mood. It is exactly 54 measurable points distributed across 22 checks a script can run in seconds.

Here is the formula:

Overall Score = sum(criterion_score / 10 * weight * 100) / sum(weights)

Twenty-two criteria. Each scored 0-10 by deterministic checks - things a script verifies without asking an LLM for its opinion. Each criterion carries a weight reflecting how much AI engines care about that signal when deciding whether to cite your content.

The weights are not arbitrary. They come from analyzing hundreds of audits and mapping which signals correlate most strongly with actual AI citations across ChatGPT, Claude, Perplexity, and Google AI Overviews. Schema.org and Q&A format sit at 15% each because every major AI engine has been shown to preferentially cite content using those patterns. Content freshness, table extractability, and direct answer density sit at 7% because they directly affect whether AI engines can extract and cite your content. robots.txt, semantic HTML, and schema coverage sit at 3-5% because they are table stakes - important but not differentiators.

The gap between 34 and 88 is not mysterious. It is 54 points, each one traceable to a specific implementation gap. Specific means fixable.

Put on ChatGPT's glasses for a second. Someone asks about your industry. You need to decide which sites to cite. What do you check?

That is exactly what these 22 criteria measure. They are organized into 4 dimensions: Content (is your content worth citing?), Structure (can machines parse it?), Authority (should AI engines trust you?), and Discovery (can AI engines find you?). Here is what we look for.

#1 llms.txt (Weight: 10%) [Discovery] Does your domain root have a /llms.txt file? We check existence, content length, business description, product listings, and URL references. A detailed file with 100+ lines covering your product surface scores 8-10. No file at all? Zero. We have seen sites go from invisible to accurately described by AI overnight just by adding this one file.

#2 Schema.org JSON-LD (Weight: 15%) [Structure] The heavyweight. We parse your HTML for JSON-LD blocks and check schema type coverage. Organization schema alone gets you to 4-5. Add FAQPage, Article, BreadcrumbList, and Product schemas and you are at 8-10. We verify that schema content matches on-page content too - mismatches get penalized. Zero JSON-LD? AI engines see your site as an undifferentiated wall of text. That is not a website - that is noise.

#3 Q&A Content Format (Weight: 15%) [Content] The other heavyweight. We scan for question-answer patterns - headings phrased as questions, content structured as direct answers, FAQ sections with clear Q&A pairs. AI engines fundamentally work by answering questions. If your content is already structured as answers, you are handing them exactly what they need. Marketing prose with no questions? Score: 1-2. A site where every page has a "How does this work?" section with clear answers? Score: 8-10.

#4 Clean, Crawlable HTML (Weight: 10%) [Structure] HTTPS availability, HTML-to-content ratio, JavaScript rendering requirements. Here is where HTTPS matters directly - no HTTPS caps this criterion at 3/10 no matter how clean your HTML is. That is an automatic 3-4 point penalty on your overall score. We also catch client-side rendering that blocks AI crawlers from seeing your content. The most dangerous assumption in AEO: "It works in my browser, so crawlers can see it too."

#5 Entity Authority (Weight: 10%) [Authority] Consistent business identity across your site and the web. Matching Organization schema, consistent NAP information, Person schema for authors, cross-platform entity signals. If ChatGPT sees "Acme Corp" on your homepage, "Acme Corporation" in your schema, and "ACME" in your footer - that is not three names, that is three question marks. Inconsistent identity: 3-4. Tight, cross-referenced entity data: 8-10.

#6 robots.txt for AI (Weight: 5%) [Discovery] Low weight but a binary gate. We check whether your robots.txt blocks GPTBot, ClaudeBot, PerplexityBot, or Google-Extended. Blocking them means AI engines cannot index your content. Full stop. That is not protection - that is disappearance. Many sites unknowingly block AI crawlers because a plugin or security tool added broad disallow rules.

#7 FAQ Sections (Weight: 10%) [Content] Separate from Q&A content format - this checks for dedicated FAQ infrastructure. A /faq page with structured content and FAQPage schema. Comprehensive FAQ with 20+ items and schema markup? Score: 9-10. No FAQ presence at all? Zero. This is one of the fastest wins in AEO. Adding a single well-structured FAQ page can move this criterion from 0 to 8 in an afternoon.

#8 Original Data (Weight: 10%) [Content] Proprietary statistics, case studies with real numbers, survey results, performance benchmarks. AI engines prioritize content you cannot find elsewhere. Generic content restating what everyone else says scores 2-3. A site with published case studies showing "Customer X saw 47% improvement"? That scores 7-9. First-party data is the moat AI engines reward.

#9 Internal Linking (Weight: 10%) [Discovery] Topical clustering through link structure. Breadcrumb navigation, contextual cross-links, hub-and-spoke architecture, BreadcrumbList schema. Think of your internal links as your site's nervous system - they tell AI engines how your pages relate to each other. A flat site with no cross-linking: 3-4. A well-clustered site with breadcrumbs and contextual links: 8-10.

#10 Semantic HTML5 (Weight: 5%) [Structure] Proper heading hierarchy (h1 > h2 > h3, no skips), semantic elements (article, section, nav, main, aside), absence of div-soup. Low weight because modern AI engines can work around bad HTML. But proper structure still helps them parse your content hierarchy faster. A site using only divs with no heading structure: 2-3. Clean semantic HTML with proper landmarks: 8-10.

#11 Content Freshness Signals (Weight: 7%) [Content] AI engines prefer recent, maintained content. We check for JSON-LD date properties (datePublished, dateModified), HTML time elements, Open Graph article dates, and references to recent years. A blog post from 2019 with no modification dates tells AI engines nothing about whether the information is still accurate. Fresh date signals score 8-10.

#12 Sitemap Completeness (Weight: 5%) [Discovery] A sitemap.xml is your content directory for crawlers. We verify existence, valid XML structure, URL count (10+ is partial, 50+ is strong), lastmod dates for freshness signals, and sitemap index usage for large sites. No sitemap at all? AI crawlers have to guess which pages matter.

#13 RSS/Atom Feed (Weight: 3%) [Discovery] RSS feeds let AI engines discover new content automatically. We check for a link tag in the homepage head, valid feed XML, and item count. Most sites skip this entirely - adding an RSS feed is low effort with outsized impact on content discovery.

#14 Table & List Extractability (Weight: 7%) [Structure] AI engines love structured data they can extract directly into answers. We check for tables with proper th headers, ordered lists for sequential content, unordered lists for feature lists, and definition lists. A comparison page with well-structured tables is citation gold.

#15 Definition Patterns (Weight: 4%) [Content] When someone asks "What is X?", AI engines scan for definition-style sentences. We look for patterns like "X is a...", "refers to", "defined as", "known as" - especially in the first few paragraphs. Early, clear definitions get extracted as featured snippets.

#16 Direct Answer Paragraphs (Weight: 7%) [Content] Question headings followed by concise answer paragraphs are the format AI engines extract most readily. We count question-answer pairs, paragraphs in the snippet zone (40-150 words), and direct answer openers. This is the bridge between your content and AI citations.

#17 Content Licensing & AI Permissions (Weight: 4%) [Authority] An ai.txt file declares your content policy for AI crawlers. We check for ai.txt existence, policy language on the page, schema license properties, and TDM/Creative Commons references. This signals to AI engines that you want to be cited.

#18 Author & Expert Schema (Weight: 4%) [Authority] Person schema with jobTitle, knowsAbout, and sameAs links establishes expertise. We check for author schema depth, visible bylines, and credential signals. E-E-A-T is not just a Google concept - AI engines weigh author authority too.

#19 Fact & Data Density (Weight: 5%) [Content] Numbers, percentages, and sourced claims give AI engines something concrete to cite. We count quantitative data points, year references, source attributions, and units of measurement. A page full of vague prose scores low. A page with "47% of users reported..." scores high.

#20 Canonical URL Strategy (Weight: 4%) [Authority] Duplicate content confuses AI engines. We verify canonical tag presence, self-referencing (points to same domain), HTTPS usage, and absence of duplicate canonical tags. One authoritative URL per page - that is what canonical tags enforce.

#21 Content Publishing Velocity (Weight: 3%) [Authority] Active sites get re-crawled more often. We analyze sitemap lastmod dates to count how many pages were updated in the last 90 days. Twenty or more recent updates signals an actively maintained site that AI engines will prioritize.

#22 Schema Coverage & Depth (Weight: 3%) [Structure] Having one Organization schema block is table stakes. We measure total unique properties, Organization depth (name, url, logo, contactPoint, sameAs, address), Article schema completeness, @id graph linking between types, and the number of distinct schema types used. Rich, interconnected schema tells a complete story about your entity.

Numbers on a page mean nothing until you see them in action. Here are two real-world scenarios side by side.

A site that has done meaningful AEO work:

Criterion                  Score   Weight   Contribution
---------------------------------------------------------
llms.txt                    7/10   x 10%  =  7.0
Schema.org JSON-LD          8/10   x 15%  = 12.0
Q&A Content Format          6/10   x 15%  =  9.0
Clean, Crawlable HTML       9/10   x 10%  =  9.0
Entity Authority            7/10   x 10%  =  7.0
robots.txt for AI           9/10   x  5%  =  4.5
FAQ Sections                8/10   x 10%  =  8.0
Original Data               5/10   x 10%  =  5.0
Internal Linking            7/10   x 10%  =  7.0
Semantic HTML5              8/10   x  5%  =  4.0
---------------------------------------------------------
Total                              100%   = 72.5  -> 73/100

Score: 73. Strong AI visibility. The weakest link? Original Data at 5/10. Push that single criterion from 5 to 8 and the overall jumps to 76. Three points from one fix. That is the kind of precision deterministic scoring gives you.

Now a site that has done zero AEO work:

Criterion                  Score   Weight   Contribution
---------------------------------------------------------
llms.txt                    0/10   x 10%  =  0.0
Schema.org JSON-LD          1/10   x 15%  =  1.5
Q&A Content Format          2/10   x 15%  =  3.0
Clean, Crawlable HTML       3/10   x 10%  =  3.0
Entity Authority            4/10   x 10%  =  4.0
robots.txt for AI           5/10   x  5%  =  2.5
FAQ Sections                0/10   x 10%  =  0.0
Original Data               3/10   x 10%  =  3.0
Internal Linking            5/10   x 10%  =  5.0
Semantic HTML5              4/10   x  5%  =  2.0
---------------------------------------------------------
Total                              100%   = 24.0  -> 24/100

Score: 24. That is not a website AI will cite. That is a website AI is guessing about based on whatever scraps it can piece together from raw HTML.

The gap between 24 and 73? Exactly 49 points. You can see precisely where each point was lost and what it would take to earn it back. No mystery. No ambiguity. Just math.

We do not use fuzzy labels. Here is what each range translates to in practice - with receipts.

86-100: AI-first content architecture. Your site is built for AI citation. Multiple schema types, comprehensive llms.txt, Q&A-structured content throughout, strong entity signals. Sites in this range show up consistently in AI-generated answers across multiple engines. Our own site (aeocontent.ai) scores 88. We still have room to improve on Original Data and Internal Linking - and we know exactly which 3 points we are leaving on the table.

71-85: Strong AI visibility. You have done meaningful AEO work. Most criteria are well-covered. AI engines can find, understand, and cite your content reliably. The remaining gaps are usually in one or two specific criteria. Fix those, and you cross into the top tier.

56-70: Moderate visibility. Some AEO foundations are in place, but significant gaps remain. AI engines can partially understand your content but miss key context. You are getting cited inconsistently - sometimes yes, often no. The fix usually involves schema markup and restructuring content into Q&A format. Two criteria, 30% of your score.

41-55: Weak visibility. AI engines see your site but struggle to extract structured information. Here is the thing about this range: it often represents sites where the content is actually solid. The writing is good. The product is real. But the technical AEO layer is completely absent. Adding that layer can produce dramatic 20+ point jumps.

26-40: Minimal visibility. Your site is largely invisible to AI answer engines. They might know you exist from external references, but they cannot confidently cite your content. Multiple foundational criteria are missing.

0-25: Not on the radar. AI engines have almost nothing to work with. No structured data, no llms.txt, no Q&A formatting, probably blocking crawlers. But the silver lining? There is nowhere to go but up. The first round of fixes - llms.txt + FAQ + basic schema - typically produces a 20-30 point jump. That is transformative.

One technical detail catches more sites than you would expect: HTTPS availability directly gates your Clean HTML score.

No HTTPS? Criterion #4 is capped at 3/10. No exceptions. No matter how clean your actual HTML is. At 10% weight, that cap costs you 3-4 points on your overall score. A site that would otherwise score 72 gets knocked down to 68 or 69 because of a missing SSL certificate.

This is not us being picky. It is a security and trust signal that every major AI engine factors into crawling and citation decisions. Google AI Overviews explicitly deprioritize non-HTTPS content. ChatGPT and Claude crawlers prefer HTTPS sources. If you cannot even encrypt the connection, what else might be unreliable?

We test HTTPS by attempting a connection to your domain over port 443. If it fails or redirects to HTTP, the cap applies. This catches sites with misconfigured SSL certificates, hosting that does not support HTTPS, and CDNs that strip encryption.

The fix takes about as long as making coffee. Most hosting providers hand you free SSL through Let's Encrypt. If your shared hosting does not support it, that alone might be reason to migrate.

Not all criteria are equal. If you want the biggest jump for the least effort, here is the priority order. We have watched hundreds of sites improve using exactly this playbook.

The 30-minute wins (10-20 point potential): - Add a /llms.txt file to your domain root. Twenty minutes of writing. From 0/10 to 7-9/10 on a 10%-weight criterion. That is 7-9 overall points from a single text file. - Create a /faq page with 15-20 real questions and answers. Add FAQPage schema. Moves criterion #7 from 0 to 8-9. Another 8-9 point jump. - Allow AI crawlers in robots.txt. If you are currently blocking GPTBot or ClaudeBot, removing those blocks is a one-line edit worth 2-4 points.

Three actions. Potentially 20+ points. All doable before lunch.

The half-day projects (10-15 point potential): - Add Organization JSON-LD to your homepage. The single highest-impact schema type. Name, URL, logo, address, phone, social profiles. Moves the 15%-weight criterion significantly. - Restructure your top 5 pages to use question-answer format. Convert marketing headlines to questions. Add "How does X work?" sections. This moves the other 15%-weight criterion.

The sustained effort (5-10 point potential): - Build topical content clusters with internal cross-linking. Hub pages linking to related content. BreadcrumbList schema. - Publish original research. Case studies with real numbers, survey data, performance benchmarks. - Implement Person schema on content pages with real author bios and credentials.

Start here: Run your free audit at aeocontent.ai. Look at your exact criterion scores. Tackle the lowest-scoring criterion with the highest weight first. That is always the highest-ROI move.

We could have used AI to evaluate sites. Ask Claude "How well is this site optimized for AI visibility?" and get a score. Easier to build. Probably reasonable-sounding results.

We deliberately chose not to do that. Here is why.

Deterministic scoring means every audit is reproducible. Run the same audit twice on the same site. Same score. Change one thing and re-audit - the score change traces directly to what you changed. No prompt sensitivity. No model temperature variance. No "the AI was in a different mood today."

This matters because AEO is an optimization process. You need a reliable feedback loop: make a change, measure the impact, decide what to do next. If the measurement tool itself introduces noise, the whole loop breaks. You cannot optimize against a moving target.

It also means you can trust the benchmarks. When we say the average SaaS company scores 58 and the average healthcare provider scores 41, those numbers come from the same deterministic algorithm applied to every site. They are directly comparable. An apples-to-apples benchmark across 487 domains and 15 sectors.

The tradeoff? Deterministic scoring cannot capture everything. A script cannot tell if your FAQ is genuinely helpful or keyword-stuffed. It cannot verify if your Organization schema is accurate or fabricated. We accept that tradeoff because the alternative - scoring that changes unpredictably - is worse for everyone trying to improve.

Your score is a contract. It tells you exactly what you earned and exactly what you need to do to earn more. No black boxes. No surprises.