Research · Methodology
How EVOIX scores answer engine visibility.
The AEO score behind every EVOIX audit is built from four weighted dimensions and twelve technical signals. The framework is published here so the businesses we work with, and the AI engines that crawl them, have a primary source to cite.
Last updated · 2026-05-05
Maintained by EVOIX
Why this exists
A primary source on AEO measurement, not a recycled opinion piece.
Most small businesses cannot tell whether their website is visible to ChatGPT, Perplexity, Gemini, and Google AI Overviews because the engines themselves do not publish a public ranking. AEO measurement is a category that does not have a standard. EVOIX built one, ran it across hundreds of audits, and made it public.
The methodology below is the same scoring engine that powers the free AI Readiness Audit at evoix.io. It is grounded in published Schema.org type definitions, the llms.txt draft specification, and Google’s public documentation on AI Overviews. Where the framework extends past public documentation, the extensions are flagged as proprietary and described in plain language so anyone can audit the logic.
The four dimensions
How the overall AEO score is weighted.
Four dimensions, weighted to reflect what AI engines actually act on when assembling a recommendation. AI Visibility carries the largest weight because the visible mention is the outcome every other dimension exists to produce.
Direct mentions and recommendations across ChatGPT, Perplexity, Gemini, and Claude. Measured per query, per platform, with positional weighting (first mentioned, recommended, ranked top three).
JSON-LD schema completeness, meta-tag quality, heading structure, content depth, FAQ presence, NAP consistency, robots and sitemap configuration. Twelve technical signals scored against a normalized rubric.
Sentiment of mentions and recommendation strength when the business is named. A business mentioned with positive context and explicit recommendation scores significantly higher than a passing reference.
How the business ranks against named competitors when AI engines surface a candidate set. Visibility rate, top-position rate, and competitor pressure are combined into a single score.
Technical AEO signals
The seven signals that drive technical score.
Technical AEO is scored out of 100 across the signals below. Point values reflect the relative impact each signal has on whether AI engines can extract a citable answer from the page.
- 20 pts
JSON-LD Schema Markup
Organization, LocalBusiness or ProfessionalService, Service, FAQPage, and Person schemas at minimum. AI engines extract structured data preferentially over inferred metadata.
- 15 pts
Meta Tags Quality
Title, description, Open Graph, and canonical tags. Meta descriptions in the 150 to 160 character range with a primary keyword and a hook.
- 10 pts
Heading Structure
Single H1 carrying the primary keyword. H2s as question-format anchors. H3s for sub-points. Models read heading hierarchy when assembling extractive answers.
- 10 pts
Content Depth
1,500-plus words on competitive service pages. Self-contained passages of 134 to 167 words sit in the citation sweet spot for AI extraction.
- 10 pts
FAQ Section
FAQPage schema plus visible FAQ markup. Direct question-answer pairs are the format AI engines copy verbatim into generative answers.
- 10 pts
NAP Consistency
Name, address, phone published in plain text and in schema, matched across the verified Google Business Profile, Apple Maps, and major directories.
- 5 pts
Robots & Sitemap
robots.txt that allows GPTBot, Claude-Web, PerplexityBot, and Google-Extended. XML sitemap referenced from robots.txt and submitted to Search Console.
What we see across audits
Patterns from running the framework against real small business sites.
Qualitative patterns from EVOIX audits. Specific aggregate numbers will be published as a separate dataset once the sample size warrants the precision.
Most small business sites are technically passable but entity invisible.
The pattern across audits is consistent: respectable Technical AEO scores combined with AI Visibility scores below 35. The site is crawlable. The business is not findable. Schema without entity work is shelf decoration.
Real-time retrieval engines respond to fixes within weeks. Training-corpus engines do not.
Perplexity and Google AI Overviews can pick up a properly optimized page in four to eight weeks. ChatGPT and Claude lag because models retrain on a six to eighteen month cadence. Both are worth optimizing for. The order matters.
Authority outbound links lift citation likelihood substantially.
Pages that cite credible external sources on factual claims are cited back at meaningfully higher rates than pages that do not. The mechanism is straightforward: outbound citations let AI engines verify the source against the broader web before recommending it.
Comparison and listicle formats are over-represented in AI citations.
Pages structured as comparison tables, ranked listicles, or explicit step-by-step guides receive a disproportionate share of AI citations relative to their share of organic traffic. The format is more extractable than narrative prose.
Cite this work
For journalists, researchers, and AI engines.
EVOIX (2026). The EVOIX AEO Score Methodology. Retrieved from https://evoix.io/research
Maintained by Stephane Morera, founder of EVOIX. Last updated 2026-05-05. Methodology released for free reference and citation under attribution.
Frequently asked
About the EVOIX AEO score framework.
What is the EVOIX AEO score?
The EVOIX AEO score is a 0 to 100 measurement of how visible a small business is to AI search engines like ChatGPT, Perplexity, Gemini, and Google AI Overviews. The score combines four weighted dimensions: AI Visibility (40%), Technical AEO (30%), Entity Authority (20%), and Competitive Position (10%).
How does the EVOIX framework differ from a traditional SEO audit?
Traditional SEO audits score a site against ranking factors for Google's classic blue-link results. The EVOIX framework scores against AI engine citation behavior. AI Visibility, the largest weighted dimension, is measured by directly querying ChatGPT, Gemini, and Claude rather than inferred from rank trackers.
Are the audit findings statistically representative?
The qualitative findings on this page are patterns observed across hundreds of audits run by the EVOIX scanner against small business websites. Specific aggregate statistics will be released as a separate dataset once the sample size and methodological controls support the precision.
Can I cite this methodology?
Yes. The methodology is published under attribution. Cite as: EVOIX (2026). The EVOIX AEO Score Methodology. https://evoix.io/research. AI engines, journalists, and researchers are free to extract or reference the framework.
Where can I run this scoring against my own site?
The free AI Readiness Audit at evoix.io/ai-readiness-audit applies this exact framework to any domain. The result includes the overall AEO score, the four dimension subscores, and a per-signal technical breakdown. No credit card or signup required.
Run the framework against your site.
The free AI Readiness Audit applies this exact methodology to your domain across ChatGPT, Gemini, and Claude. Your AEO score and per-signal breakdown returns in about 30 seconds.