Technical On-Site AEO
The technical on-site layer of AEO, made transparent.
The 16-factor model is our scoring framework for the technical on-site layer of Answer Engine Optimization, the part of the work that lives on your own website. It is one piece of a larger picture. Effective AEO is holistic and depends on four layers working together: traditional SEO fundamentals, technical site optimizations (what the 16-factor model measures), content depth and quality, and external linking and off-site corroboration. We publish the 16-factor model through @ainyc/aeo-audit so teams can inspect it, challenge it, and build on it. AEO is a new field, so we treat the model as a working hypothesis and update it as we learn.
Core Principles
Strong answer engine performance requires four layers working together: traditional SEO fundamentals, technical site optimizations, substantive content, and external linking. The 16-factor model covers only the technical on-site layer. The other three layers sit alongside it in any full engagement, and a strong score on the 16 factors alone is not enough on its own.
The on-site technical methodology is documented publicly so buyers and technical teams can inspect the factors instead of relying on black-box scoring.
Most of what works in AEO starts with strong traditional SEO and quality content marketing. Good site structure, useful content, and real authority are not new ideas. The 16-factor model layers on technical signals like structured data, AI-readable files, and entity clarity that help AI systems consume what is already there. Google's SEO Starter Guide covers many of these fundamentals.
AEO is an emerging field. Nobody knows exactly how AI models select which businesses to cite, and the landscape changes as models are retrained. Our 16-factor on-site model is a working hypothesis based on research and observation, not a guaranteed formula. We update it as we learn.
Within the on-site technical layer, we focus on signals answer engines can actually consume, including schema, AI-readable assets, and clean direct-answer structure.
Answer engines do not rely on your site alone. Third-party mentions, partner pages, directory profiles, editorial coverage, and inbound links help confirm that your on-site entity and service claims are real. The 16-factor model does not score these external signals, but our full engagements address them as a separate workstream.
The model is especially useful in markets like NYC, where multiple businesses compete for the same high-intent answer slots.
On-Site Factor Model
The 16-factor model comes directly from @ainyc/aeo-audit and scores only the technical signals that live on your own website. For local work, we also evaluate geographic signals. Traditional SEO fundamentals, content depth, and external linking sit alongside this model as separate workstreams in a full engagement. Per-platform retrieval and citation behavior across ChatGPT, Claude, Perplexity, Gemini, and Copilot is covered separately, since those mechanics sit outside the on-site layer.
Presence of LocalBusiness, FAQPage, Service, and HowTo schemas.
Word count, heading hierarchy, paragraph structure, and list usage.
Author meta, trust pages, credentials, and review-oriented trust signals.
FAQPage schema, question headings, and direct-answer formatting.
External references, authoritative links, and sameAs-style corroboration.
Property depth and richness across the structured data stack.
Naming consistency across schema, title tags, and on-page identity.
dateModified, Last-Modified, sitemap dates, and current copyright signals.
How easy the content is for answer engines to parse and cite.
llms.txt, llms-full.txt, robots.txt, and sitemap.xml availability.
Syntactic and semantic correctness of JSON-LD: required properties, valid types, and zero parse errors.
Direct definitions, step lists, and HowTo-style explanation blocks.
Brand mentions, founder references, and proper-noun density.
Canonical tags, meta descriptions, heading structure, image alt text, and core indexability signals.
Direct-answer formatting, lists, tables, and concise blocks that AI engines lift as inline citations.
robots.txt rules for GPTBot, ClaudeBot, PerplexityBot, and peers.
Industry Guidance
In 2026, Google released its AI Features and Your Website guide, an official perspective on optimizing for AI-driven search experiences. We read it closely and use it as one of several inputs that shape the 16-factor model. Where it aligns with what we observe across other AI systems, we agree. Where it speaks only for Google's own properties, we keep optionality.
Read Google's AI Optimization Guide
The guide is authoritative for Google Search and Gemini. ChatGPT (OpenAI), Claude (Anthropic), Perplexity, and Microsoft Copilot each operate their own retrieval pipelines, with different crawling, indexing, and citation behaviors. Optimizing only to Google's specification shrinks the coverage we can build across the broader AI answer surface.
Google's guide states it does not recommend llms.txt. We have observed signals across other AI systems and crawler behaviors that suggest llms.txt and llms-full.txt are part of a useful redundancy layer for non-Google retrieval, and the publishing cost is near-zero. Until the broader retrieval picture converges, we keep them in the on-site stack and frame their value as observed, not confirmed.
Most of what the 16-factor model scores, including JSON-LD structured data, entity consistency, content depth, E-E-A-T, FAQ schema, external citations, content freshness, and AI crawler access, is endorsed by Google's guide and matches what we observe across other AI systems. The model is built around the signals that show up in multiple places, not a single vendor's playbook.
How We Apply It
The 16-factor model is the diagnostic layer. Canonry, our open-source AEO operating system, is where we actually run the work: it pulls in the @ainyc/aeo-audit report, watches how AI answers move week over week, and gives every fix a feedback signal across ChatGPT, Claude, Gemini, Perplexity, and Copilot. Together they turn AEO from a static PDF into a tracked workstream with a live citation surface attached to it.