The GEO Checklist: What's Actually Verified to Work

Most "GEO checklists" repeat the same dozen bullet points without saying where any of them came from. Some are backed by real research. Some are Google's own documented guidance. And some — like the claim that Google AI Overviews "preferentially pulls quotes from <blockquote> tags" — are just SEO folklore that got repeated until it sounded true. It isn't documented anywhere, and it doesn't belong in a serious playbook.

This piece is a reference, not a deep dive — a single scannable page that consolidates what's actually verifiable about generative engine optimization (GEO) right now, organized into six categories: crawlability, entity clarity, evidence density, content structure, freshness, and measurement. Every item below is tied to a primary source or a named study. Where the common wisdom is unproven or has been quietly walked back, that's called out explicitly rather than smoothed over.

Crawlability comes first — and it's still just SEO

Direct answer: Generative engines can't cite what they can't retrieve. Google has been unambiguous that AI Overviews and AI Mode are fed by the same crawler as regular Search: "AI is built into Search and integral to how Search functions... robots.txt directives for Googlebot is the control for site owners to manage access to how their sites are crawled for Search," per Google's AI features documentation. There's no separate "AI crawler" to configure for Google's own surfaces — if Googlebot can index a page, it's eligible.

Two things people conflate that are actually separate controls:

• Google-Extended governs whether your content can be used to train future Gemini models — it's a directive in robots.txt, not a crawler with its own traffic. Disallowing it doesn't remove you from Search or from AI Overviews eligibility.
• llms.txt is not a Google standard, and adoption data doesn't support treating it as load-bearing. Independent crawl-traffic analysis found AI bot requests for /llms.txt accounted for roughly 0.1% of all AI crawler traffic on a monitored domain, and Google's own John Mueller has compared it publicly to the discredited keywords meta tag — see the state-of-llms.txt roundup for the traffic data and quotes. It costs nothing to add, but don't treat it as a visibility lever.

Baseline crawlability hygiene — no noindex/robots blocks on pages you want cited, a submitted sitemap, and pages reachable without login walls — is covered in Google's robots.txt guide and Googlebot documentation.

Entity clarity: tell machines what you are, unambiguously

Direct answer: Generative engines assemble answers by retrieving and synthesizing across sources — which means unambiguous entity signals (who wrote this, what organization stands behind it, what this page is actually about) do real work. Google's own guidance here is structured data plus E-E-A-T, not anything AI-specific.

For structured data, Google's general structured data guidelines are explicit: markup must describe visible on-page content (no marking up information the user can't see), should use the most specific applicable schema.org type, and needs validation via the Rich Results Test before you trust it. Organization markup on your homepage or about page, and Article markup with clear author attribution, are the two highest-leverage types for a B2B content site.

One nuance worth flagging: FAQ structured data used to produce rich dropdown results in Google Search, but Google deprecated FAQ rich results entirely as of May 2026 — the schema type still exists and isn't harmful to keep, but it no longer buys you a SERP feature (Search Engine Journal coverage). If your checklist still lists "add FAQ schema for the rich snippet," that line is out of date.

For E-E-A-T, Google's Search Quality Rater Guidelines update added "Experience" as a fourth pillar in 2022 specifically to reward content demonstrating firsthand use of a product or topic, not just secondhand summarization. Byline author pages, visible credentials, and genuine first-hand detail are the concrete implementation of this — not a ranking factor Google scores directly, but a documented input to how raters (and therefore Google's algorithm tuning) judge quality, especially for YMYL topics.

Evidence and citation density — the one area with an actual study behind it

Direct answer: This is the best-evidenced category in GEO, because it's the one an actual peer-reviewed study measured. Princeton, Georgia Tech, IIT Delhi, and the Allen Institute for AI published "GEO: Generative Engine Optimization" at KDD 2024 (ACM proceedings), building a 10,000-query benchmark and testing nine content-modification strategies against a generative-engine simulation, with the strongest tactics validated on a live deployed system.

The headline, stated in the paper's own abstract: these optimization strategies "can boost visibility by up to 40%" in generative engine responses. Directionally, the strategies that helped were adding citations to credible external sources, adding direct quotations, adding statistics, and improving fluency/readability — all consistently positive across the tested domains. Keyword stuffing was the one strategy that underperformed the unoptimized baseline. The paper is also honest about limits worth repeating here: effect size varied by domain, meaning a tactic that moves the needle in one content category won't necessarily transfer to another.

Treat this as the one item on this checklist with real experimental backing — and treat any claim about exactly which HTML element or formatting trick AI Overviews prefers (the blockquote claim being the canonical example) as unverified until Google documents it, because Google hasn't.

Content structure: extractable, not gimmicked

Direct answer: Google itself is the least prescriptive on structure. Its own position, stated plainly in the AI features documentation, is that "there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary" beyond fundamental SEO. That's worth sitting with — a lot of "AI content structure" advice is a rebrand of long-standing featured-snippet practice (direct answers under question-phrased headers, short scannable paragraphs, lists and tables where the content is genuinely list- or table-shaped) rather than anything new.

Where there is newer evidence is around Google's query fan-out technique, which both AI Overviews and AI Mode use: rather than answering the literal query, Google's systems generate related sub-queries and retrieve across them, per Google's own description in the AI features documentation and its AI Mode launch announcement. Independent analysis of this behavior — not Google's own claim, so treat it as directional rather than definitive — found that pages structured to comprehensively answer a topic's adjacent sub-questions get cited more often than pages narrowly targeting one query, because they're more likely to surface in one of the fanned-out sub-searches rather than just the original (query fan-out analysis). Practically: cover the follow-up questions a reader would actually ask next, in the same piece, rather than optimizing a single narrow query.

Freshness: real signal, easy to fake badly

Direct answer: Google uses multiple signals to infer when content was published or updated — structured data (datePublished/dateModified), visible on-page dates, and sitemap lastmod values — because any single one can be wrong or gamed. Google's John Mueller has stated directly that changing a dateModified value without making a substantive change is "noise" that doesn't help rankings or visibility (documented in Search Engine Land's byline dates guide). The signal only works if the update is real: revised facts, updated figures, corrected claims — not a timestamp bump on unchanged text.

Freshness appears to matter more for retrieval-augmented generative systems than for classic ranking, since the retrieval step in RAG-based answer generation applies recency as a filter across candidate sources when several cover the same ground — though this is an industry-observed pattern rather than a documented Google mechanism, so hold it more loosely than the crawlability or structured-data items above.

Measurement: track citations, not just rankings

Traditional rank tracking doesn't capture whether an LLM-based answer engine mentions or cites you — a page can rank on page one and never surface in an AI Overview or chatbot answer, and vice versa. The metrics practitioners have converged on are mention rate (share of tracked prompts where your brand is named), citation rate (share of responses linking to a domain you own), and share of answer (your prominence relative to other brands named in the same response) — defined and benchmarked in LLM Pulse's share-of-voice methodology. Build a fixed set of category and comparison prompts, run them repeatedly across the answer engines that matter to your buyers, and track the trend — a single spot-check isn't a measurement system, since these responses are probabilistic and vary run to run.

The checklist

What to leave off your checklist

Direct answer: Two things worth naming explicitly, since the discipline here is as much about what to stop repeating as what to add: specific HTML-element preferences (like a "blockquote boost") that no engine has documented, and llms.txt as anything more than a low-cost, low-priority addition. If a claim about GEO can't be traced to a primary source or a named study, it doesn't belong in a checklist that's supposed to be the reliable summary of everything else on the site.

Sources:

• AI Features and Your Website — Google Search Central
• Robots.txt Introduction and Guide — Google Search Central
• What Is Googlebot — Google Search Central
• General Structured Data Guidelines — Google Search Central
• Organization Schema Markup — Google Search Central
• Article Schema Markup — Google Search Central
• Google's E-E-A-T Quality Rater Guidelines Update
• GEO: Generative Engine Optimization (arXiv, Princeton/Georgia Tech/IIT Delhi/AI2, KDD 2024)
• GEO: Generative Engine Optimization — ACM KDD 2024 Proceedings
• Google Drops FAQ Rich Results From Search — Search Engine Journal
• Byline Dates in SEO: What Google Actually Uses — Search Engine Land
• The State of llms.txt in 2026
• Google's Query Fan-Out Technique and What SEOs Should Know
• Share of Voice in AI Search: How to Calculate It — LLM Pulse