Open-source agentic AI Search Execution Platform. Track AI visibility, find opportunities, and let agents turn signals into prioritized actions and drafted fixes across ChatGPT, Claude, Gemini, Google AI Overviews, Google AI Mode, Perplexity, Grok, and Microsoft Copilot.
TypeScript aeoai-agentai-searchanswer-engine-optimizationbrand-monitoringchatgptclaudecopilotdockergeminigeogoogle-ai-overviewsgrokmcpnextjsopen-sourceperplexityseosupabasetypescript
{
"version": "v1.1",
"description": "48-signal AEO/GEO scoring rubric for measuring whether a page will get cited by AI answer engines.",
"totalSignals": 48,
"categories": [
{
"key": "structure",
"label": "Structure",
"weight": 0.25,
"description": "How parseable your page is to an AI engine. Schema, headings, lists, tables, alt-text, OG meta. The substrate that makes everything else legible.",
"signalCount": 13
},
{
"key": "authority",
"label": "Authority",
"weight": 0.2,
"description": "Whether your page reads as authored by an expert, citing real sources. Bylines, citations, statistics, quotations, freshness, technical terms.",
"signalCount": 8
},
{
"key": "content",
"label": "Content",
"weight": 0.25,
"description": "Whether the page actually answers the buyer query — directly, completely, in the right register. BLUF, query coverage, sub-question coverage, readability, and on product pages the description's opening.",
"signalCount": 12
},
{
"key": "trust",
"label": "Trust",
"weight": 0.1,
"description": "Crawl-time + share-time trust signals: HTTPS, canonical, viewport, AI-bot access via robots/llms.txt, Twitter card, meta description.",
"signalCount": 7
},
{
"key": "eeat",
"label": "E-E-A-T",
"weight": 0.2,
"description": "Google's framework for Experience, Expertise, Authoritativeness, Trustworthiness. First-person evidence, case studies, author credentials, brand entity in Wikidata, press mentions, privacy policy, contact info.",
"signalCount": 8
}
],
"signals": [
{
"key": "structural-depth",
"category": "structure",
"label": "Structural depth (GEO-SFE)",
"what": "Heading hierarchy + nested structure depth — H1→H2→H3 nesting that mirrors a real outline.",
"why": "arXiv:2603.29979 shows a +17.3% citation lift for pages with measurable structural depth vs flat documents.",
"howToFix": "Group your content under named H2 subsections. Use H3 for sub-points. Don't skip levels.",
"source": "GEO-SFE (arXiv:2603.29979) + Aggarwal et al. NeurIPS 2024",
"impactTier": "high"
},
{
"key": "internal-linking",
"category": "structure",
"label": "Internal linking",
"what": "Links from this page to other pages on your site — signal of a navigable graph.",
"why": "AI crawlers expand context via internal links the same way they crawl. Isolated pages get under-indexed.",
"howToFix": "Add 5–10 internal links per page to related concepts, glossary terms, or supporting pages.",
"source": "Anthropic web-search retrieval docs",
"impactTier": "medium"
},
{
"key": "page-weight",
"category": "structure",
"label": "Page weight",
"what": "Total HTML size, lower is better. We flag pages over 500 KB.",
"why": "AI render bots have shorter timeouts than humans. Heavy pages get truncated or skipped.",
"howToFix": "Strip unused JS, defer below-the-fold images, kill render-blocking third-party scripts.",
"source": "Google SGE crawler docs",
"impactTier": "standard"
},
{
"key": "json-ld-presence",
"category": "structure",
"label": "JSON-LD presence",
"what": "Whether the page ships at least one schema.org JSON-LD block.",
"why": "Schema.org is the single highest-leverage signal for AI Overview eligibility — engines parse it before text.",
"howToFix": "Add at minimum an Organization or Article block. Our /tools/schema-generator outputs valid JSON-LD.",
"source": "Google SGE evaluation guide",
"impactTier": "high"
},
{
"key": "json-ld-validity",
"category": "structure",
"label": "JSON-LD validity",
"what": "Whether the JSON-LD parses, has @context = schema.org, and a valid @type.",
"why": "Malformed JSON-LD is worse than absent — engines log a parse error and downrank trust.",
"howToFix": "Run your JSON-LD through Google's Rich Results Test before shipping.",
"source": "Google SGE evaluation guide",
"impactTier": "medium"
},
{
"key": "json-ld-relevance",
"category": "structure",
"label": "JSON-LD type relevance",
"what": "Whether the @type matches the page intent (Article on a blog, Product on a PDP, FAQPage on a FAQ).",
"why": "Wrong-type schema gets ignored — Article markup on a homepage is parsed as noise.",
"howToFix": "Pick the closest schema.org type for the page's actual content, not a generic catch-all.",
"source": "schema.org documentation",
"impactTier": "medium"
},
{
"key": "faq-schema",
"category": "structure",
"label": "FAQ schema",
"what": "FAQPage JSON-LD with at least 2 Q&A pairs.",
"why": "FAQPage is the highest single-type lift for AI Overview citations per Google's own SGE guidance.",
"howToFix": "Add FAQPage schema with 3–5 question/answer pairs that mirror real buyer queries.",
"source": "Google SGE evaluation guide",
"impactTier": "medium"
},
{
"key": "h1-quality",
"category": "structure",
"label": "H1 quality",
"what": "Exactly one H1, length 20–70 chars, contains a primary query token.",
"why": "AI engines treat H1 as the page's claim — multiple H1s confuse the topical model.",
"howToFix": "One H1 per page, descriptive, 30–60 chars, includes the page's primary topic.",
"source": "Google SGE crawler docs",
"impactTier": "medium"
},
{
"key": "h2-coverage",
"category": "structure",
"label": "H2 coverage",
"what": "≥3 H2 subsections that segment the content for skim-readability.",
"why": "AI engines extract H2s as section anchors — pages without them are harder to summarize.",
"howToFix": "Break content into 3–6 H2 subsections, each addressing a sub-question.",
"source": "Aggarwal et al. NeurIPS 2024",
"impactTier": "medium"
},
{
"key": "lists",
"category": "structure",
"label": "Lists",
"what": "≥1 <ul> or <ol> with ≥3 items.",
"why": "Lists are the unit AI engines preferentially extract for 'how to X' and 'best Y' queries.",
"howToFix": "Convert at least one paragraph of enumerated reasons or steps into a real <ul>/<ol>.",
"source": "Aggarwal et al. NeurIPS 2024",
"impactTier": "standard"
},
{
"key": "tables",
"category": "structure",
"label": "Tables",
"what": "≥1 <table> with proper <thead>/<tbody> for structured data.",
"why": "Tables are the highest-extraction-rate format for comparison and spec-sheet queries.",
"howToFix": "If you have specs, prices, or comparisons, render them as real <table>s, not images.",
"source": "Aggarwal et al. NeurIPS 2024",
"impactTier": "standard"
},
{
"key": "alt-text",
"category": "structure",
"label": "Alt-text",
"what": "≥80% of <img> tags have non-empty alt attributes.",
"why": "Multi-modal engines (Gemini, Claude, GPT-4o) read alt-text as page semantic context.",
"howToFix": "Add descriptive alt to every meaningful image. Decorative images get alt=\"\".",
"source": "Google SGE evaluation guide",
"impactTier": "standard"
},
{
"key": "open-graph",
"category": "structure",
"label": "Open Graph metadata",
"what": "og:title, og:description, og:image, og:type all present.",
"why": "AI engine link previews + social shares use OG meta as the citation card.",
"howToFix": "Add the 4 OG tags. Use 1200×630 og:image. Our /tools/og-checker validates them.",
"source": "Open Graph Protocol",
"impactTier": "medium"
},
{
"key": "outbound-authority-links",
"category": "authority",
"label": "Outbound authority links",
"what": "Links to .edu, .gov, Wikipedia, peer-reviewed journals, or top-tier news sources.",
"why": "Linking out to trusted sources transfers authority backwards (citation graph effect).",
"howToFix": "Cite primary sources with real <a> tags, not bare-text references. Aim for 2–3 per long page.",
"source": "GEO paper §4.2 + Anthropic retrieval docs",
"impactTier": "medium"
},
{
"key": "citation-density",
"category": "authority",
"label": "Citation density",
"what": "Inline citations per 1000 words (target ≥2).",
"why": "Pages that cite sources get cited as sources. Citation density is the strongest non-link authority signal.",
"howToFix": "Replace 'studies show' with 'a 2024 Stanford study (link)'. Specificity wins.",
"source": "GEO paper §4.2",
"impactTier": "medium"
},
{
"key": "statistic-density",
"category": "authority",
"label": "Statistic density",
"what": "Quantitative claims with numbers per 1000 words (target ≥3).",
"why": "AI engines preferentially cite content with measurable claims over hand-wavy text.",
"howToFix": "Add real numbers — percentages, dollar amounts, time durations — to your claims.",
"source": "Aggarwal et al. NeurIPS 2024",
"impactTier": "medium"
},
{
"key": "quotations",
"category": "authority",
"label": "Direct quotations",
"what": "Inline <blockquote> or quoted attributions.",
"why": "Quotations imply primary research. They lift citation rates ~12% in the GEO paper test set.",
"howToFix": "Quote one customer, one expert, or one primary source per long page.",
"source": "GEO paper",
"impactTier": "standard"
},
{
"key": "author-byline",
"category": "authority",
"label": "Author byline",
"what": "A named author with at least a real name in the byline area.",
"why": "Anonymous content is treated as low-E-E-A-T. AI engines surface authored content.",
"howToFix": "Add 'By [Name]' at the top. Link to a real author bio page. Use Person schema.",
"source": "Google SGE + E-E-A-T docs",
"impactTier": "medium"
},
{
"key": "date-markup",
"category": "authority",
"label": "Date markup",
"what": "datePublished + dateModified in JSON-LD or <time datetime>.",
"why": "Engines penalize undated content as potentially stale. Dated content gets freshness scoring.",
"howToFix": "Add JSON-LD datePublished + dateModified, or a visible <time datetime> tag.",
"source": "Google SGE evaluation guide",
"impactTier": "standard"
},
{
"key": "freshness",
"category": "authority",
"label": "Freshness",
"what": "Date modified within the last 365 days.",
"why": "Stale content gets downranked for time-sensitive queries. Re-publishing dates resets the clock.",
"howToFix": "Refresh top pages at least annually. Bump dateModified when you do.",
"source": "Google freshness algorithm docs",
"impactTier": "standard"
},
{
"key": "technical-terms",
"category": "authority",
"label": "Technical term density",
"what": "Domain-specific terminology indicating real expertise vs surface-level writing.",
"why": "Pages that use precise technical vocabulary signal genuine expertise.",
"howToFix": "Use the field's actual technical language. Glossary-link or define the first occurrence.",
"source": "GEO paper §4.3",
"impactTier": "standard"
},
{
"key": "direct-answer",
"category": "content",
"label": "Direct answer to query",
"what": "Whether the buyer query appears answered in the first 200 chars.",
"why": "AI engines preferentially extract direct answers. Burying the answer = no citation.",
"howToFix": "Lead the page with a 1–2 sentence direct answer to the primary query. Then expand.",
"source": "GEO paper §3.2",
"impactTier": "high"
},
{
"key": "bluf-answer",
"category": "content",
"label": "BLUF (Bottom Line Up Front)",
"what": "A clear thesis statement in the first paragraph stating the page's main claim.",
"why": "Military-doctrine writing style — BLUF dramatically improves AI extractability.",
"howToFix": "Open with 'The answer is X. Here's why.' Don't bury the lede.",
"source": "U.S. military writing doctrine + adopted by Anthropic prompt eng team",
"impactTier": "medium"
},
{
"key": "query-coverage",
"category": "content",
"label": "Query token coverage",
"what": "Fraction of query tokens that appear in the page body (target ≥80%).",
"why": "Coverage is a baseline relevance signal — missing tokens means the page misses the query.",
"howToFix": "Make sure every meaningful query token appears at least once in body content.",
"source": "BM25 + cosine retrieval fundamentals",
"impactTier": "medium"
},
{
"key": "entity-coverage",
"category": "content",
"label": "Entity coverage",
"what": "Named entities mentioned that match the query's entity space.",
"why": "AI engines build an entity graph per page — missing related entities reduces topical authority.",
"howToFix": "If the page is about X, mention X's known related entities (people, places, products).",
"source": "Knowledge graph extraction literature",
"impactTier": "medium"
},
{
"key": "readability",
"category": "content",
"label": "Readability (Flesch)",
"what": "Flesch Reading Ease score (target 50–70 for general audiences).",
"why": "Engines extract from content humans can read. Sub-30 (academic) and 80+ (childlike) both underperform.",
"howToFix": "Aim for 50–70. Under 50: shorten sentences and cut adverbs, about 16 words a sentence. Over 70: write fuller sentences with the detail readers expect.",
"source": "Flesch (1948) + adapted by SGE",
"impactTier": "standard"
},
{
"key": "length",
"category": "content",
"label": "Page length",
"what": "Word count in the main content (target 800–3000).",
"why": "Too short = thin; too long = engines truncate. Sweet spot is medium-long form.",
"howToFix": "If under 600 words, expand with examples + sub-sections. If over 4000, split into hub + spokes.",
"source": "Empirical GEO paper data",
"impactTier": "standard"
},
{
"key": "info-density",
"category": "content",
"label": "Information density",
"what": "Ratio of substantive content to filler.",
"why": "Dense content gets preferentially cited. Engines model 'value per token' implicitly.",
"howToFix": "Cut throat-clearing sentences. Replace 'It is important to note that' with the actual point.",
"source": "Empirical GEO paper data",
"impactTier": "standard"
},
{
"key": "mega-page-coverage",
"category": "content",
"label": "Mega-page coverage",
"what": "Whether the page covers multiple closely-related sub-topics under one URL.",
"why": "Mega-pages outrank thin pages on AI engines because they cover entity neighborhoods.",
"howToFix": "Consolidate 3 thin pages into one comprehensive page with H2-segmented sub-topics.",
"source": "Ahrefs research + Cleartopic.io",
"impactTier": "medium"
},
{
"key": "youtube-embed",
"category": "content",
"label": "YouTube embed",
"what": "Embedded YouTube video alongside the article.",
"why": "Multi-modal engines weight pages with on-topic video. 0.737 correlation with citation rate.",
"howToFix": "Create or embed a 2–4 minute video on the page topic. Mark up with VideoObject schema.",
"source": "Ahrefs 2025 study",
"impactTier": "standard"
},
{
"key": "product-description-opening",
"category": "content",
"label": "Product description opening",
"what": "On a product page, the description's first 40 words (or 250 characters) name the product type and at least two key attributes: a measure, a material, a fit or a use.",
"why": "ChatGPT appears to read only about the first 50 tokens of a product description when it matches products to a shopping question. A description that opens with the brand's story may never be matched.",
"howToFix": "Put the product type and key attributes in the first 40 words.",
"source": "brightonSEO 2026 research on ChatGPT Shopping (to be confirmed by our own tests)",
"impactTier": "high"
},
{
"key": "sub-query-coverage",
"category": "content",
"label": "Sub-query coverage",
"what": "Fraction of decomposed sub-questions answered in body content.",
"why": "Engines decompose seed queries into 4–6 sub-queries (Profound 'fanouts') — missing answers = missing citations.",
"howToFix": "Add a Q&A section that explicitly addresses the top 5 sub-questions for your topic.",
"source": "Profound Query Fanouts methodology",
"impactTier": "medium"
},
{
"key": "definitions",
"category": "content",
"label": "In-line definitions",
"what": "First-mention terms defined inline ('X (definition…)').",
"why": "Defined-in-place pages get cited as glossary sources. AI engines preferentially link to them.",
"howToFix": "Define the key 3–5 terms on first use, in-line.",
"source": "Stanford NLP citation extraction work",
"impactTier": "standard"
},
{
"key": "ai-bot-access",
"category": "trust",
"label": "AI bot access (robots.txt)",
"what": "Whether robots.txt explicitly addresses GPTBot/ClaudeBot/PerplexityBot.",
"why": "Default-allow gets you crawled, but explicit allow signals intent + unlocks niche AI bots.",
"howToFix": "Add User-agent: GPTBot/ClaudeBot/OAI-SearchBot/PerplexityBot rules to robots.txt.",
"source": "OpenAI / Anthropic / Perplexity bot docs",
"impactTier": "medium"
},
{
"key": "llms-txt-presence",
"category": "trust",
"label": "llms.txt presence",
"what": "Whether /llms.txt exists per llmstxt.org spec.",
"why": "Explicit channel for telling AI engines exactly which content to ingest. 2026 frontier signal.",
"howToFix": "Generate llms.txt via our /audit/llms-txt tool. Add to site root.",
"source": "llmstxt.org spec",
"impactTier": "medium"
},
{
"key": "https",
"category": "trust",
"label": "HTTPS",
"what": "Site responds correctly over TLS without mixed content.",
"why": "AI crawlers refuse to crawl mixed-content pages. HTTPS is table stakes.",
"howToFix": "Migrate to HTTPS. Force redirect with HSTS. Fix mixed-content warnings.",
"source": "Web standards + Google ranking docs",
"impactTier": "standard"
},
{
"key": "canonical",
"category": "trust",
"label": "Canonical URL",
"what": "<link rel=canonical> pointing to the page's preferred URL.",
"why": "Without canonical, AI engines may split citations across duplicate URLs and downrank each.",
"howToFix": "Add <link rel=canonical href=…> to every page. Self-canonical on canonical pages.",
"source": "Google canonical docs",
"impactTier": "standard"
},
{
"key": "viewport",
"category": "trust",
"label": "Viewport meta",
"what": "<meta name=viewport content=…> present.",
"why": "Mobile rendering signal — AI crawlers favor mobile-first content.",
"howToFix": "Add <meta name=viewport content=\"width=device-width, initial-scale=1\">.",
"source": "Google mobile-first indexing docs",
"impactTier": "standard"
},
{
"key": "twitter-card",
"category": "trust",
"label": "Twitter card",
"what": "<meta name=twitter:card> + twitter:image present.",
"why": "AI engines fall back to Twitter card on platforms without OG support.",
"howToFix": "Add twitter:card=summary_large_image + twitter:image (can reuse og:image).",
"source": "X (Twitter) card docs",
"impactTier": "standard"
},
{
"key": "meta-description",
"category": "trust",
"label": "Meta description",
"what": "<meta name=description> with 80–160 chars of summary text.",
"why": "AI engines use meta description as a primary summary candidate when generating cites.",
"howToFix": "Write a 130-char meta description for every page. Include the primary query.",
"source": "Google docs + SGE evaluation guide",
"impactTier": "standard"
},
{
"key": "first-person",
"category": "eeat",
"label": "First-person evidence",
"what": "First-person language indicating real experience ('we tested', 'I built').",
"why": "Google's E-E-A-T update prioritizes lived experience. First-person signals it.",
"howToFix": "Use first-person sparingly but precisely — 'we tested', 'in our experience', 'I found that…'.",
"source": "Google E-E-A-T quality rater guidelines",
"impactTier": "medium"
},
{
"key": "case-study-evidence",
"category": "eeat",
"label": "Case study evidence",
"what": "Named customer outcomes, numbers, or before/after evidence in-line.",
"why": "Case studies are the highest-conversion E-E-A-T signal — prove it, don't just claim it.",
"howToFix": "Name one customer. Quote one outcome with numbers. 'We helped X go from Y → Z.'",
"source": "Google E-E-A-T guidelines",
"impactTier": "medium"
},
{
"key": "author-credentials",
"category": "eeat",
"label": "Author credentials",
"what": "Author has visible credentials — years of experience, prior role, education, certifications.",
"why": "Anonymous bylines fail E-E-A-T. Credentialed authors get treated as expert sources.",
"howToFix": "Each author page lists credentials, prior roles, years in field, links to LinkedIn/Twitter.",
"source": "Google E-E-A-T guidelines (Quality Rater Guidelines §3.2)",
"impactTier": "medium"
},
{
"key": "byline-depth",
"category": "eeat",
"label": "Byline depth",
"what": "Byline links to a real author bio page with sufficient detail.",
"why": "Byline → bio page → credentialed author chain is the E-E-A-T graph engines trace.",
"howToFix": "Make every byline link to /authors/[name] with bio, photo, credentials, prior work.",
"source": "Google E-E-A-T guidelines",
"impactTier": "standard"
},
{
"key": "brand-entity",
"category": "eeat",
"label": "Brand entity in Wikidata",
"what": "Brand has a Wikidata entity (Q-number) that AI engines can dereference.",
"why": "Wikidata is the canonical entity database — engines use it to validate brand identity. No entity = no trust anchor.",
"howToFix": "Get a Wikidata entry: requires 2–3 press mentions as citations. Submit at wikidata.org.",
"source": "Wikidata + knowledge graph literature",
"impactTier": "medium"
},
{
"key": "press-mentions",
"category": "eeat",
"label": "Press mentions",
"what": "References to or links from top-tier press (NYT, WaPo, Bloomberg, FT, etc.).",
"why": "Press mentions transfer authority across the citation graph and validate brand entity claims.",
"howToFix": "Earn press. Display logos. Link back to the original article (not your case study).",
"source": "Empirical brand-citation research",
"impactTier": "standard"
},
{
"key": "privacy-terms",
"category": "eeat",
"label": "Privacy + terms",
"what": "Linked privacy policy + terms of service from every page.",
"why": "YMYL (Your-Money-Your-Life) categories require these — and engines treat them as legitimacy signals everywhere.",
"howToFix": "Footer-link privacy and terms on every page. Real pages, not coming-soon stubs.",
"source": "Google YMYL + E-E-A-T guidelines",
"impactTier": "standard"
},
{
"key": "contact-info",
"category": "eeat",
"label": "Contact info",
"what": "Real contact info — email, phone, or address visible on the page or linked /contact.",
"why": "Anonymous brands fail E-E-A-T. Real contact info signals real entity behind the content.",
"howToFix": "Add an email + physical address (real or registered office) to the footer.",
"source": "Google E-E-A-T guidelines",
"impactTier": "standard"
}
]
}