- This checklist covers nine layers — strategy, technical, on-page, content, entities and schema, local, off-site, contested AEO/GEO tactics, and measurement — and it is not a complete list.
- Every item carries an evidence rating: Strong (sourced to a primary document or a large study), Weak/contested (plausible but untested or disputed), or None yet (no study opened, logical inference only).
- The strongest single correlate with AI visibility in a 75,000-brand Ahrefs study was branded web mentions, not backlinks or page count.
- Fix technical access before anything else. Nothing downstream matters if the crawler that would read your page can't reach it.
- Measure with the same four metrics every time — visibility, share of voice, sentiment, position — and treat one run as directional, never as a verdict.
What belongs at each layer?
A checklist this size is easy to treat as a to-do list to clear end to end. It isn't one. Some items here are backed by a primary document or a large study; others are common advice nobody has actually tested. The evidence column tells you which is which, so you can spend time on the items that are more likely to matter and skip, or deprioritize, the ones that aren't.
| Layer | Checklist item | Why it connects to an AI answer | Evidence |
|---|---|---|---|
| Strategy & intent | Write down the buyer questions you want to win, not just keywords | AI answers are assembled from several related queries run against your actual question (query fan-out), not a single keyword match | Strong — Google Search Central |
| Strategy & intent | Decide which engines you're actually prioritizing, and why | Engines source differently; in a July 2026 Fractl survey of 343 marketing decision-makers, 34% prioritize ChatGPT Search, 23% Google AI Overviews, 16% Gemini, and 1% Perplexity | Strong (marketer behavior, not a causal AI-mechanism finding) |
| Technical | Confirm AI crawlers can actually fetch your key pages | Retrieval-based answers can't ground anything in a page the crawler never fetched | Strong — OpenAI crawler docs; Google's retrieval-augmented mechanism |
| Technical | Check robots.txt bot by bot, not with one blanket "block AI" rule | Training bots and search bots are controlled separately; blocking the wrong one removes you from answers without the training opt-out you wanted | Strong — OpenAI crawler docs |
| Technical | Serve primary content in the initial HTML response, not only after client-side JavaScript runs | A crawler that only sees an empty shell has nothing to ground an answer in | Strong — standard crawling constraint, consistent with the retrieval mechanism above |
| On-page | Put one sentence near the top of each core page that directly answers its main question | Plausible AEO practice, consistent with "answer-first" convention | Weak/contested — no study we opened isolates this as its own tested lever; Google's guidance only asks for helpful, people-first content generally |
| On-page | Keep FAQ content genuinely useful, not padding for schema's sake | Same reasoning as above | Weak/contested |
| Content | Publish original information — your own data, a real test, a documented experience | Google explicitly flags "scaled content abuse" from AI-produced commodity variations; Ahrefs' own site-pages correlation with AI visibility is weak (about 0.194 across ChatGPT, AI Mode and AI Overviews) | Strong — Google Search Central; Ahrefs, 75,000 brands |
| Content | Refresh factual claims — pricing, features, availability — on a schedule | Outdated facts get cited as current; this is the "brand described wrongly" pain point surveys keep finding | None yet — no controlled study opened; logical inference only |
| Entities & schema | Keep your brand name and one-line description consistent across your site and the profiles engines pull from | Branded web mentions were the single strongest correlate Ahrefs tested with AI visibility (0.66–0.71 across platforms) | Strong, correlational — Ahrefs explicitly cautions this isn't causation |
| Entities & schema | Add or fix Organization/Product schema where you already use it for rich results | Google's guide says structured data "isn't required" specifically for its generative AI features | Weak/contested |
| Local | Keep local listings and profiles accurate | Plausible for location-based prompts | None yet — no primary source opened on local AI-answer behavior |
| Off-site | Earn real mentions on third-party sites engines already cite, especially YouTube | YouTube mentions were Ahrefs' single strongest correlate across platforms (0.737) | Strong, correlational |
| Off-site | Treat backlinks and domain authority as a minor factor, not the lever | Ahrefs found backlinks (0.218) and Domain Rating (0.266–0.326) correlate far more weakly than branded mentions | Strong (as a "this is weak" finding), still correlational |
| Contested AEO/GEO tactics | Publish an llms.txt file | Google: "doesn't use" it; an Ahrefs study of 137,210 domains found 97% of published files got zero traffic in a month | Weak/contested — see llms.txt: what the evidence says |
| Contested AEO/GEO tactics | Break content into tiny "AI-friendly" chunks | Google's guide states directly there's "no requirement to break your content into tiny pieces for AI to better understand it" | Contradicted by the primary source |
| Measurement & testing | Track visibility rate — how often you're named across repeated runs — not a single rank | SparkToro and Gumshoe.ai found under a 1-in-100 chance of the same brand list appearing twice; frequency held up better than order | Strong |
| Measurement & testing | Use the same four metrics every time (visibility, share of voice, sentiment, position); treat any one run as directional only | Same study, plus ContextOwl's own measurement standard | Strong |
This is not every possible item in any of the nine layers — it's the items we could put a real source or a real "no evidence yet" against. If a tactic you've heard about isn't here, that's more likely an omission than a verdict.
How does each item connect to AI answers, and what counts as evidence?
Three ways an item earns its evidence rating:
- Strong — traced to a primary document (Google's own guide, OpenAI's crawler documentation) or a large, named study (Ahrefs' 75,000-brand analysis, SparkToro and Gumshoe.ai's run-variance study). Correlational findings are still labeled Strong when the sample is large and the source is primary — but correlation is not causation, and we say so inline.
- Weak/contested — a plausible, widely repeated piece of advice with no study we opened that actually isolates it as a lever, or an item where sources disagree.
- None yet — no primary or tier-2 source was opened on this specific point this session. That's a gap in our research, not proof the item doesn't matter.
What should you do first?
Technical access, every time. If OAI-SearchBot, PerplexityBot or ClaudeBot can't fetch your key pages, nothing else in this checklist can matter — a perfectly original, accurately priced, well-schema'd page that a crawler can't reach is invisible regardless. The AI search audit checklist walks through exactly this step in more detail, and the technical AI-readiness checklist goes deeper on the rendering and infrastructure side. After that, the Strong-rated content and entity items (original information, consistent branding) earn their place before the Weak/contested ones (schema for AI specifically, answer-first phrasing) and the explicitly contested AEO/GEO tactics, which are optional, low-cost, and not proven levers. If your team is still arguing about which of these is "SEO's job" versus "a new GEO job," SEO vs AEO vs GEO: what should your team actually do covers that split directly.
How do you measure progress?
With the same four metrics on every check: visibility (how often you're named across repeated runs), share of voice (your slice of every brand mention for a prompt set), sentiment (positive, neutral or negative when named), and position (where you land when named — the least stable of the four). Run each prompt more than once before drawing a conclusion; a single run is a data point, not a measurement.
A worked example
A fictional small SaaS brand, "Ledgerly," ran this checklist on 9 October 2026:
| Item | Finding | Action |
|---|---|---|
| Crawler access | robots.txt disallowed GPTBot and OAI-SearchBot together under one old "block bots" rule | Split the rule: keep GPTBot blocked (a real training-data preference), allow OAI-SearchBot |
| Original information | Pricing page only; no original data anywhere on the site | Published a short internal usage-benchmark post as the first original-information asset |
| Branded mentions | Zero mentions on YouTube, three scattered forum mentions with outdated pricing | Corrected the forum threads where possible; logged YouTube as a gap to address later |
| llms.txt | None published | Left for later — rated Weak/contested, not worth doing before the three items above |
Four rows, one afternoon, and the fixes went in order of evidence strength, not alphabetically down the table.
FAQ
Is this checklist complete? No. It covers nine layers with items we could source one way or another. Tactics without a rating here weren't tested, not ruled out.
Do I need to do every "Weak/contested" or "None yet" item? No. They're reasonable to do if they're cheap (schema, llms.txt), but don't deprioritize a Strong-rated item — crawler access, original content, branded mentions — to make time for one of these.
What's the actual difference between the SEO, AEO and GEO items here? We didn't split the table that way on purpose. Most items serve more than one label at once — fixing crawler access is "technical SEO" and also the precondition for any AEO or GEO outcome. The checklist organizes by what you're doing, not by which buzzword claims it.
How often should I re-run this? Quarterly, or immediately after a robots.txt change, a CMS migration, or a rebrand — any of those can flip a passing item to a failing one without you noticing.
What to do next
- Run the Technical layer first and fix any crawler-access gap before touching anything else on the list.
- Work down the Strong-rated items in Content, Entities & Schema, and Off-site before spending time on the Contested AEO/GEO tactics layer.
- Set up one repeatable way to check visibility rate across a prompt set, and re-run the whole checklist next quarter.
ContextOwl sells one-time AI visibility reports. The AI Bot Access Checker is a free way to check the Technical layer's crawler-access item right now.
Sources
- Google Search Central, "Optimizing your website for generative AI features on Google Search" (last updated 10 July 2026): https://developers.google.com/search/docs/fundamentals/ai-optimization-guide
- OpenAI, "Overview of OpenAI Crawlers": https://developers.openai.com/api/docs/bots
- Ahrefs, correlation study of AI brand visibility factors across ChatGPT, AI Mode and AI Overviews (12 December 2025; 75,000 brands, Spearman correlation via Ahrefs Brand Radar): https://ahrefs.com/blog/ai-brand-visibility-correlations/
- SparkToro and Gumshoe.ai, AI recommendation consistency study (600 volunteers, 12 prompts, 2,961 runs, November–December 2025), as reported by Search Engine Journal: https://www.searchenginejournal.com/ai-recommendations-change-with-nearly-every-query-sparktoro/566242/
- Fractl, survey of 343 US marketing decision-makers (21 July 2026), as reported by PPC Land: https://ppc.land/marketers-still-call-it-seo-as-81-reject-geo-pitch-fractl-finds/
- Ahrefs, "We Analyzed 137K Sites: 97% of llms.txt Files Never Get Read" (15 June 2026; 137,210 domains): https://ahrefs.com/blog/llmstxt-study/
More in Playbooks
Put the answer in the first paragraph, ask the question in the heading, and stop writing once one question is answered. That's most of it.
AI answers about you start from what's findable under your own name. A clean homepage, claimed profiles and healthy branded search are the base layer everything else sits on.
List every source an AI engine cites for your competitors but never for you. That list is your content and outreach plan — the GEO equivalent of a keyword gap analysis.