Structure: A shared core of factors that apply to any site, then three divergence tracks (ecommerce, lead gen, and nonprofit / purpose-driven) covering the layers that genuinely differ by business model: schema, cited content formats, query intent, off-site surfaces, and measurement.

Download this taxonomy as a PDF.

This is not a “Score”

No standardized GEO/AEO score exists. The academic base (Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024) measured citation visibility inside generated answers, not a page-level grade. Every vendor score (Geoptie, Profound, Peec, Semrush AI Toolkit, Otterly) is a private composite with undisclosed weightings.

Readiness is per-engine. Google AI Overviews leans on its own index and top organic results. ChatGPT Search runs on OAI-SearchBot with Bing index heritage. Perplexity does live retrieval and favors fresh, structured, citation-forward pages. Gemini rides the Google index. A single composite averages these away and hides where you are actually weak. Score each engine you care about separately.
Most of this is downstream of traditional SEO. Crawl access, clean HTML, authority, and structured content are prerequisites. GEO adds a citation-worthiness layer and an off-site consensus layer on top. The off-site layer is where you have influence without full control, and it is where most tools score poorly.

Confidence legend: [High] strong evidence it moves citation. [Moderate] supported but correlational or contested magnitude. [Low] plausible, thin or disputed evidence. [Unknown] opaque to outside measurement.

PART 1: SHARED CORE
(applies to ecommerce and lead gen)

Layer 1: Crawl & Access
Can the AI systems reach and fetch the content.

Retrieval-crawler access in robots.txt [High]: allow the search/retrieval agents that power live citation: OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-SearchBot, Claude-User, PerplexityBot, plus standard Googlebot and Bingbot.
Training vs retrieval distinction handled deliberately [High]: you can block training crawlers (GPTBot, Google-Extended, CCBot, anthropic-ai, Applebot-Extended, Bytespider) while keeping retrieval agents open. Blocking retrieval agents removes you from live answers.

Server / WAF / CDN rules not silently blocking AI bots [High]: Cloudflare bot-fight mode, managed AI-bot blocking, and aggressive rate limiting frequently block retrieval crawlers regardless of what robots.txt says.
HTTP status hygiene [High]: clean 200s, correct 301/404/410, no soft 404s, no redirect chains on key pages.

XML sitemap present, accurate, fresh [Moderate]: aids discovery and recrawl prioritization.

Server responsiveness under bot load [Moderate]: timeouts and 5xx during crawl drop pages from eligibility.

No unintended auth / paywall / geo-block on citable content [High]: content behind login or IP restriction cannot be cited.

Reverse-DNS-verifiable bot handling [Low]: spoofed user agents are common; verification affects your own log analysis more than citation directly.

llms.txt file [Low]: proposed convention to guide LLM access. Adoption by major engines is unconfirmed. Treat as cheap-to-ship experiment.

Layer 2: Rendering & Parsing
Can they extract clean text once they fetch.

Content in the initial server-rendered HTML [High]: most retrieval crawlers execute JavaScript inconsistently or not at all. Content injected client-side after load is frequently invisible to them. Server-side rendering or static generation for anything you want cited.

Semantic HTML with correct heading hierarchy [High]: single H1, logical H2/H3 nesting, semantic elements (article, section, nav, table).

Key content not hidden behind interaction [Moderate]: accordions, tabs, and "read more" toggles that keep text out of the raw DOM reduce extractability.

Text not locked inside images, canvas, or unlabeled iframes [High]: no OCR dependency for facts you want cited.

Clean DOM, reasonable text-to-code ratio [Moderate]: heavy script/markup noise degrades parsing.

Language and encoding declared [Moderate]: lang attribute, correct charset, hreflang for multilingual.

No cloaking / bot-user content parity [High]: divergence between what bots and users see risks suppression and is a policy violation.

Mobile parity [Moderate]: the mobile render is often the canonical one.

Layer 3: Content Structure & Extractability
Retrieval-Augmented Generation systems retrieve and cite passages, not whole pages. Structure for chunk-level extraction.

Answer-first structure / answer capsule near the top [High]: a direct, self-contained answer under the H1. Retrieval-based engines weight opening content heavily.

Front-loaded key information [High]: a large share of citations pull from the introduction and first portion of the text (Growth Memo / ConvertMate 2026 benchmarking). Don’t bury the answer because you won’t get the citation.

Question-based headings [High]: H2/H3 phrased as the actual queries people ask.

Chunkable, self-contained passages [High]: each section makes sense pulled out of context, with its own subject stated rather than relying on a pronoun from three paragraphs up.

Lists and structured elements [High]: ordered and unordered lists are disproportionately cited (Profound 2026: majority of cited pages use lists/structured elements).

Comparison and data tables in semantic HTML [High]: tables rendered as image files are not extractable.

Short paragraphs, scannable formatting [Moderate].

Definition blocks, TL;DR, summary sections [Moderate].

FAQ sections [Moderate]: maps directly to conversational query patterns.

Consistent terminology for entities and concepts [Moderate]: synonym-swapping fragments the entity signal.

One idea per section (topical focus) [Moderate].

Layer 4: Content Substance & Citation-Worthiness

Original research, proprietary data, statistics [High]: "Statistics Addition" was among the strongest visibility levers in the KDD 2024 study. Specific, verifiable numbers outperform vague claims.

Direct quotations from authorities [High]: "Quotation Addition" produced the single largest measured lift in the same study.

Inline citations to credible sources [High]: "Cite Sources" was a top-three lever. Sourced content is cited more.

Factual specificity [High]: concrete dates, named entities, exact figures over generalities.

Comprehensiveness / topical coverage [Moderate]: covering the subtopics implied by the query.

Factual accuracy [High]: content contradicting established facts is suppressed.

Content freshness of the facts stated [High]: see Layer 8.

Named author with demonstrable credentials (Expertise, Authoritativeness) [Moderate].

First-hand experience signals (Experience) [Moderate]: heavier for YMYL topics.

Non-promotional tone [Moderate]: overtly salesy copy is cited less than neutral, informational framing.

Uniqueness / non-templated content [Moderate]: boilerplate repeated across pages dilutes.

Content-type match to query intent [High]: listicles, how-tos, and comparisons over-index in citations (Lantern 2026 content-type analysis).

Layer 5: Entity & Semantic
Does the model understand who and what you are as an entity.

Unambiguous entity identification [High]: Clear, consistent naming and description of the brand/org/person.

Entity presence in knowledge graphs [Moderate]: Google Knowledge Graph, Wikidata, Wikipedia. Notability drives inclusion.

Wikipedia / Wikidata entry [Moderate]: Strong corroborating signal where notability warrants it. Do not fabricate notability.

Consistent third-party descriptors [Moderate]: How external sources describe you shapes how models describe you.

Disambiguation from same-named entities [Moderate].

sameAs linkage to authoritative profiles [Moderate]: Connects your entity to known nodes.

Author entities recognized independently [Low to Moderate].

Topical authority (entity-topic association) [Moderate]: The corpus associating your brand with your category.

Layer 6: Structured Data / Schema (generic core)

Valid JSON-LD schema [Moderate]: Aids machine understanding and feeds traditional rich results that in turn feed AI Overviews. Direct causal lift on LLM citation is contested; treat as understanding-and-eligibility infrastructure.

Core types deployed [Moderate]: Organization, WebSite, Article, BreadcrumbList, Person, FAQPage, HowTo where applicable.

Connected schema graph via @id [Low to Moderate].

Schema-to-content match [High]: Markup must reflect visible content. Schema spam risks manual action and erodes trust.

Author markup [Low to Moderate].

Speakable schema [Low]: Limited current relevance.

Layer 7: Authority, Trust & Off-Site (generic core)
You influence this without controlling it. Most scoring tools underweight it because it is hard to measure.

Third-party citations and earned media [High]: External mentions carry heavy weight (Muck Rack + Seer 2026 found earned-media trust signals among the strongest citation correlates).

Presence in "best of" and roundup listicles [High]: AI answers lean heavily on curated lists. Getting placed in the lists others publish is a primary lever.

Brand mentions across the web, linked and unlinked [High].

Referring domain count and diversity [Moderate]: SE Ranking's 400,000-URL analysis found referring-domain correlation with citation probability.

Backlinks from authoritative domains [Moderate]: Still correlated, weaker than in classic ranking.

Reddit / forum presence and sentiment [Moderate to High]: Several engines weight community discussion heavily. Sentiment matters, not just presence.

Domain-level trust [Moderate].

Sentiment of third-party coverage [Moderate].

Consistency of brand facts across the web [Moderate]: Contradictory external data suppresses confident citation.

Layer 8: Freshness & Maintenance

Visible publish and last-updated dates [High].

Refresh cadence against the citation-decay curve [High] — a documented citation cliff appears around the 3-month mark for stale content. Publish-and-forget loses visibility.

Dated statistics kept current [High].

Content pruning and updating of legacy pages [Moderate].

Layer 9: Brand Presence & Corpus Consensus (model-level)
Off your own site, and largely opaque. This is where you least control and least measure.

Frequency of brand appearance in retrieval index and training corpora [Unknown].

Cross-source consensus about facts related to your brand [Moderate]: Agreement across sources raises citation confidence.

Share of voice in answers for your category [Moderate].

Co-citation patterns with competitors [Unknown].

Prominence / position within an answer when cited [Moderate].

Sentiment in AI-generated brand descriptions [Moderate].

Presence in the specific sources each engine favors [Moderate]: Engine-dependent source preferences.

Layer 10: Technical Performance (indirect)
Matters mainly through the traditional ranking that feeds AI Overviews.

Core Web Vitals / page speed [Low to Moderate].

HTTPS / security [Moderate].

Site architecture, internal linking, crawl depth [Moderate].

URL clarity and canonicalization hygiene [Moderate].

Duplicate content control [Moderate].

Layer 11: Platform Divergence (scored separately)
Google AI Overviews: index-and-top-organic dependent. Classic SEO strongly feeds it. Overlap between top-10 organic and AI citations has been dropping through 2026, so do not assume ranking guarantees citation.

ChatGPT Search: OAI-SearchBot plus Bing index heritage; favors certain trusted domains.

Perplexity: live retrieval, citation-forward, rewards fresh and structured content.

Gemini: Google index plus Google-Extended.

Copilot: Bing index.

Claude: web search plus ClaudeBot.

PART 2: ECOMMERCE DIVERGENCE TRACK

Schema (ecommerce)

Product schema [Moderate]: with GTIN/MPN/brand/model for product-entity clarity.

Offer schema [Moderate]: price, priceCurrency, priceValidUntil, availability, shippingDetails, hasMerchantReturnPolicy.

AggregateRating and Review schema [Moderate].

Product feed / Merchant Center accuracy [Moderate]: feeds product surfaces and shopping-integrated answers.

Cited content formats (ecommerce)

Product comparison pages with spec-and-price tables [High].
"Best [product] for [use case]" listicles [High]: the dominant format AI cites for commercial-investigation queries.

Buying guides and category explainers [High].

Spec / feature reference tables [Moderate].

Query intent (ecommerce)

Commercial investigation and transactional queries ("best," "vs," "cheapest," "is X worth it," "alternatives to X"). Structure content to answer these directly.

Off-site surfaces (ecommerce)

Marketplace presence [Moderate to High]: Amazon and category marketplaces.

Retail review platforms [High]: review volume, recency, and average rating heavily influence whether you are cited as a recommendation.

Reddit product threads and YouTube reviews [Moderate to High].

Price-comparison and aggregator presence [Moderate].

Product-entity consistency (GTIN/brand/model) across the web [Moderate].

Measurement / KPIs (ecommerce, revenue-based)

AI-source-attributed revenue: revenue from sessions referred by AI engines (GA4 channel/referrer segmentation, supplemented by server-log crawler evidence since some AI referrals arrive without clean referrer data).

Conversion rate by AI source: segment ChatGPT, Perplexity, Gemini, Claude referrals. Independent 2025 to 2026 analyses (Seer, Ahrefs) consistently show AI-referred sessions converting at multiples of organic search, driven by high pre-formed purchase intent.

Average order value by AI source vs other channels.

Assisted revenue: AI citations that influence a purchase completed through another channel later. Model this in multi-touch attribution; last-click will undercount it.

Revenue per cited product / page.

Citation-to-session-to-revenue funnel: citation frequency, then referral sessions, then revenue, so you can see where the drop-off is.

ROAS where paid shopping runs alongside.

PART 3: LEAD GEN / SERVICE DIVERGENCE TRACK

Schema (lead gen / service)

Organization and Service schema [Moderate].

LocalBusiness schema [Moderate]: for local service; with area served and NAP.

FAQPage and HowTo [Moderate].

Person schema for practitioners [Low to Moderate]: credential and author-entity signal.

Review schema [Moderate].

Cited content formats (lead gen / service)

How-to and process content [High].

Definitional and problem/solution content [High].

Cost and pricing explainers [High]: "how much does X cost" is a heavily cited query class; publishing real ranges gets you into the answer.

Provider / approach comparison content [High].

Case studies with concrete outcomes and data [Moderate to High].

Query intent (lead gen / service)

Informational, solution-seeking, and provider-evaluation queries ("how to," "what is," "how much does X cost," "best [service] in [location]," "X vs Y provider," "do I need X").

Off-site surfaces (lead gen / service)

Category review and comparison platforms [High]: G2 / Capterra for SaaS, Clutch for agencies, industry-specific directories for regulated services.

Google Business Profile completeness and accuracy [High for local]: with NAP consistency across citations.

Local citation consistency [Moderate for local].

Industry publications and association listings [Moderate].

Reddit and niche forums [Moderate to High].

Amplified trust and credential signals for YMYL [High]: legal, medical, and financial services face heavier E-E-A-T weighting. Named, credentialed authors and verifiable expertise matter more here.

Practitioner / author entity recognition [Moderate].

Measurement / KPIs (lead gen / service, pipeline-based)

AI-source-attributed leads: form fills and calls from sessions referred by AI engines.

Lead-to-MQL and MQL-to-SQL rate by AI source: measure quality, not only volume.

Cost per lead by source where paid runs in parallel.

Pipeline value influenced by AI-sourced first touch or assist.

Close rate and revenue per closed deal by originating source: AI-sourced leads frequently arrive further down the evaluation path with a pre-formed shortlist, which shows up as higher close rates. Track it to prove the channel's value.

Assisted conversions: AI citation as an early or mid-funnel touch in multi-touch attribution. Last-click will bury this; use a position-based or data-driven model.

Citation-to-session-to-lead funnel: same drop-off diagnosis as ecommerce, ending in leads rather than orders.

PART 4: NONPROFIT / PURPOSE-DRIVEN DIVERGENCE TRACK

This track carries only what genuinely diverges. A purpose-driven org inherits the lead-gen / service informational content formats and the amplified YMYL trust and credential signals already listed in Part 3. What follows is nonprofit-specific schema, the charity-trust off-site surfaces no other model has, a dual-audience query split, and a measurement layer that fractures into donor-facing conversions and mission-outcome (beneficiary) reach.

Schema (nonprofit)

Organization with nonprofit properties [Moderate]: nonprofitStatus (e.g. Is501c3), legalName, taxID / EIN, foundingDate.

NGO type [Low to Moderate]: schema.org NGO subtype of Organization where applicable.

FAQPage and HowTo [Moderate]: how to donate, how to volunteer, how to get help.

Event schema for fundraisers and galas [Low to Moderate].

Person schema for leadership and founders [Low to Moderate]: trust and author-entity signal.

Note: donation-action markup has weak and inconsistent support. Do not over-invest there.

Cited content formats (nonprofit)

Inherits the lead-gen informational and definitional formats. The nonprofit-distinct additions:

Cause / issue explainers grounded in original data or statute [High]: the citation magnet. A mission org that publishes the authoritative, data-grounded or statute-grounded reference on its cause becomes the cited source for the whole topic.

Impact reports with concrete outcomes and figures [Moderate to High].

Transparency and financial-accountability content [Moderate]: where donations go, program-to-overhead ratios. Feeds the donor-trust query directly.

"How to get help" resource content [High]: the beneficiary-facing counterpart, structured to be the answer for someone in need.

Query intent (dual audience)

The defining divergence. Purpose-driven orgs serve two populations with different intents, plus advocacy.

Donor-facing: trust and evaluation plus transactional: "is [org] legitimate," "best charities for [cause]," "where does my donation go," "how to donate to [cause]."

Beneficiary-facing: informational and solution-seeking, where being the cited answer means a person in need finds help: "how do I get help with [problem]," "free [service] near me," "how to stop [crisis]."

Advocacy-facing: "how to help with [issue]," "how to contact my representative about [issue]."

Off-site surfaces (nonprofit)

Charity rating and watchdog platforms [High]: Charity Navigator, Candid (GuideStar), CharityWatch, BBB Wise Giving Alliance. Engines lean on these for legitimacy and "best charity" queries. Ratings, transparency scores, and profile completeness on these platforms are controllable inputs that shape whether you are cited favorably.

Public regulatory records [Moderate]: Form 990 and 501c3 status are public and machine-readable, feeding trust and legitimacy answers.

Government and coalition resource directories [Moderate to High for beneficiary reach]: being listed on .gov and cause-specific resource pages drives citation for "how to get help" queries.

Cause-specific coalitions and directories [Moderate].

Local news and earned media on impact [Moderate to High].

Wikipedia / Wikidata [Moderate]: heavier for advocacy and larger orgs where notability warrants it.

Reddit and cause-community forums [Moderate].

Measurement / KPIs (nonprofit, split by audience)

Revenue metrics are appropriate on the donor side. The beneficiary side cannot be reduced to revenue, which is the core reason this needs its own track.

Donor side (revenue-based):

AI-source-attributed donation volume and value.

Donor acquisition by AI source.

Recurring / monthly-donor conversion by source.

Average gift size by AI source.

Assisted donations (AI citation as an early or mid-funnel touch).

Email / list captures from AI referrals: the nonprofit funnel is usually list-first, so the list signup is often the true first conversion.

Grant, major-gift, and partnership inquiries influenced.

Beneficiary / mission side (outcome-based, non-revenue):

Beneficiary reach: sessions and actions from people seeking help via AI referral.

Intake-form, hotline, and service-signup completions from AI sources.

Citation frequency on beneficiary-intent queries ("how to get help with X").

Resource-page engagement from AI referrals.

Advocacy side:

Advocacy actions from AI sources: petition signs, representative contacts, event RSVPs.

For many purpose-driven orgs the mission outcome is the primary KPI and the donation is the means that funds it. Score both. Reporting only donor revenue misrepresents the value of the beneficiary-reach work.

Turning this into a working score

Score each engine you care about separately. A blended number hides your weak engine.

Weight the shared core layers by their confidence tags. Gate factors (Layer 1) are pass/fail, not weighted. If a retrieval crawler is blocked, the page scores zero for that engine regardless of everything else.

Add the relevant divergence track for the business model.

Treat Layer 9 (corpus consensus) as a monitored output, not an on-page input you can directly grade. You influence it through Layers 4 and 7 and observe it through measurement.

The precise per-engine weightings are [Unknown] and proprietary. Any tool claiming a definitive score is asserting confidence it cannot have. Use the score for prioritization and trend tracking, not as ground truth.

Sources grounding the confidence tags: Aggarwal et al., GEO: Generative Engine Optimization (KDD 2024); SE Ranking referring-domain analysis (2025); Muck Rack + Seer earned-media study (2026); Profound structured-element analysis (2026); ConvertMate / Growth Memo citation-position benchmark (2026); Lantern content-type analysis (2026); Seer and Ahrefs AI-referral conversion studies (2025 to 2026); AI-crawler user-agent documentation from OpenAI, Anthropic, Google, Perplexity (current 2026).

Blog

GEO / AEO Readiness Factor Taxonomy