Awesome GEO

SEEN: a 41-item GEO checklist you can score today

By · Updated

SEEN is a four-layer framework for Generative Engine Optimization (Structure, Evidence, Entity and Notability) with a 41-item checklist for making an entity easy for AI to find, read, verify and recommend. This page is written by Dmitry (Dee) Kargaev, who maintains Awesome GEO and wrote the SEEN working paper.

Other GEO checklists and frameworks

Other published GEO checklists and frameworks are listed here, each as its own page describes it.

  • Incremys publishes a page-by-page checklist of technical, content, authority and monitoring controls. Each criterion is scored from zero to two, and each gap is scored on impact, ease and confidence.
  • Plain Intelligence uses a four-layer factor model of access, passage, entity and authority, plus a measurement loop. It says to run the full audit quarterly, the passage tier on every publish and the measurement tier monthly.
  • Customer Impact lists 25 points in five areas for about half a day of work, and says it needs no tooling and no baseline. The page also spreads the same order over about 90 days as a roadmap.
  • WikiAP splits the first 30 days into four weeks: measurement, technical diagnosis, content improvement and re-measurement. The page is in Korean.
  • AMEC, the International Association for the Measurement and Evaluation of Communication, asks questions on strategy, information quality, AI-generated representation, measurement and governance. It calls its checklist a starting point for discussion rather than a formal assessment.
  • Wellows describes an audit for agencies built on a prompt set of 10 to 30 prompts, a baseline report and a 30, 60 and 90 day roadmap.
  • Layer3Labs groups its checklist into technical setup, content optimization, off-page signals, measurement and engine-specific checks. It says to run the full technical checklist once, then revisit it quarterly.
  • GEO Checklist calls itself the official reference site for GEO and lists six endpoints to implement, among them /.well-known/ai.txt and /sitemap-llm.xml.

The four SEEN layers and the studies cited for them

SEEN treats GEO as corpus engineering: owned content, platform-hosted proof and third-party corroboration made coherent enough for machines to use. The SEEN working paper (2026) says traditional SEO treats the ranked web page as the unit of work and rank position as the measure of visibility. It adds that generative search weakens that model because it retrieves passages, issues derived queries, synthesizes answers, attaches citations and recommends entities without a stable ranked list. Its failure mode is called blanking: an AI answers in your category and acts as if your entity does not exist, absent rather than lower-ranked.

Structure makes the answer extractable: definitions, pricing, scope, comparisons and core facts sit in visible HTML with real hierarchy. For this layer the SEEN site cites ALM Corp / Indig (2026): 44.2% of ChatGPT citations came from the first 30% of webpage content.

Evidence makes claims verifiable: numbers, outcomes and positioning are paired with methodology pages, case studies, bylines and primary sources. The SEEN site cites Wan, Wallace and Klein (ACL 2024) here: LLMs score evidence primarily on direct relevance to the query.

Entity makes the entity unambiguous: one canonical name, one description, consistent profiles, schema and a clean sameAs graph. MultiHal (2025) is the study cited for it: knowledge-graph-grounded retrieval improved hallucination scores across evaluated models.

Notability gets the wider web to corroborate the recommendation through third-party mentions, platform-hosted reach, links and organic community presence. For this layer the SEEN site cites Otterly.AI (2026): community platforms captured roughly 52.5% of AI citations in a sample of more than 1 million citations.

The full 41-item SEEN checklist

The items split 10 for Structure, 10 for Evidence, 10 for Entity and 11 for Notability. Score each item 1 if it is clearly true today and 0 if it is missing, vague or only partly true, then fix the lowest layer first. The checklist is also available as a PDF.

Structure

  1. Key pages open with a 30 to 60 word answer-first summary.
  2. H1, H2, and H3 hierarchy describes meaning, not visual layout.
  3. Core facts are visible HTML, not images, PDFs, or JS-only state.
  4. Top competing categories have explicit comparison pages or tables.
  5. Major pages answer real buyer or user questions.
  6. Product or service pages answer what it does, who it is for, price, and next step.
  7. Schema exists on key pages and matches visible copy.
  8. Pages render usefully without JavaScript, or have SSR / prerender fallback.
  9. Long-form pages have anchors for sub-answer extraction.
  10. llms.txt and llms-full.txt exist where an owned site is part of the corpus.

Evidence

  1. Major quantitative claims link to a method page, case study, or source.
  2. Original research, methodology, or a whitepaper lives under a durable URL.
  3. At least one case study uses names, numbers, and dates.
  4. Independent reviews or public product signals exist outside the site.
  5. Substantive articles have a credentialed byline.
  6. Pricing or pricing logic is public enough to cite.
  7. Testimonials are attributed with name, role, company, and link where possible.
  8. Press, podcasts, or talks are listed with reachable links.
  9. Statistics cite primary sources, not aggregator summaries.
  10. An About or Methodology page explains how the work is done.

Entity

  1. The canonical name is identical across owned and platform-hosted surfaces.
  2. One sentence describes the entity consistently across profiles.
  3. Person, Product, or Organization schema links the sameAs graph.
  4. The About page is about the entity, not only a marketing pitch.
  5. Founder, company, and product relationships are explicit.
  6. Wikidata exists or is planned where notability allows.
  7. Local or service entities have a consistent Google Business Profile.
  8. Name collisions are disambiguated in titles, descriptions, and About copy.
  9. Profile copy is dated or updated within the last 12 months.
  10. Photos, logos, and bios match across important surfaces.

Notability

  1. Independent third-party publications mention the entity recently.
  2. The entity appears in relevant best-of, top-N, or alternatives pages.
  3. Organic discussion exists in subject-relevant communities.
  4. A meaningful number of links come from high-authority domains.
  5. Wikipedia presence or mentions exist where notability standards allow.
  6. There is a speaking, podcast, guest-writing, or public expertise trail.
  7. Platform-hosted properties show visible engagement and active cadence.
  8. Partnerships or integrations appear on partner-owned pages.
  9. Customers or collaborators publish their own proof where possible.
  10. A key person has a public trail outside the entity’s owned channels.
  11. Claimed awards come from editorial sources with real criteria.

Weakest-layer rule and quick audit

The SEEN site says the weakest layer decides the outcome: a 9 out of 10 in Structure with a 2 out of 10 in Notability does worse in practice than 6 in all four.

For a quick audit, ask an AI system to explain your entity, who it is for, why it should be trusted and which sources support the answer. Then see what breaks: extraction, proof, identity or outside corroboration.

What is not known

The SEEN working paper does not claim that the framework guarantees citations, recommendations, rankings or traffic. It presents the 41 items as a practitioner checklist and sets out a validation agenda for future empirical work.

The author’s other working paper, The SEO-to-GEO Gap (2026), is an exploratory comparative synthesis of direct SEO and GEO studies, and its three findings are provisional. One of them is that the overlap literature indicates GEO still operates on top of organic search rather than replacing SEO outright. The paper argues for a reweighting model rather than a replacement model.