The GEO checklist: 30 checks before your next content sprint

Content sprints fail quietly in GEO. A team writes six good pages, waits a month, sees no movement in AI answers and concludes the discipline is hype, when the actual culprit was a firewall rule from 2024 or a score computed from one lucky ChatGPT run. The checklist exists to catch those failures before the writing starts. It is ordered by dependency: crawlability gates content, content gates citation, third-party presence corroborates it all, and measurement is how you find out which of the first three is your bottleneck.

Every check cites its evidence, most heavily the Princeton GEO study, which tested nine optimization methods across 10,000 queries and found statistics, quotations and cited sources lifting visibility in generative answers by up to 40 percent while keyword stuffing fell below baseline. If the discipline itself is new to you, read what is generative engine optimization first, and GEO vs SEO for how these checks relate to the SEO audit you already know. Then work the groups in order and mark each check pass, fail or unknown; the unknowns are usually where the answers went.

Everything downstream depends on retrieval. AI engines pull sources from search indexes and their own crawlers, so a page they cannot fetch or parse does not exist for them, however brilliant the writing. These eight checks fail silently and often; the fix is usually minutes once found.

Retrieval puts your page on the engine's desk; structure decides whether the answer uses it. The Princeton GEO study is the evidence base here: statistics, quotations and cited sources lifted visibility by up to 40 percent while keyword stuffing fell below baseline. These eight checks turn that finding into an editing pass.

Engines compose verdicts from sources you do not own: 5W Research puts Wikipedia and Reddit alone at over a quarter of U.S. ChatGPT citations. Half of GEO is an away game, and these seven checks are its fixture list.

AI answers are probabilistic, so measurement discipline separates GEO programs from GEO theater. These seven checks define the honest loop: stable questions, repeated runs, receipts, trends, and a recheck after every fix.

Teams that run the audit tend to land in one of three profiles. The blocked publisher passes the content checks and fails crawlability: strong pages, invisible to engines, usually via a bot-protection default or a robots.txt copied from a paranoid template. The fix is fast and the payoff is fast, because the content already exists. The self-referential brand passes crawlability and content but fails the third-party group: everything the engines could say about them is something the brand said about itself, so cautious engines say nothing. That profile needs the slow work of reviews, communities and earned citations, started now because it compounds.

The optimist is the most common profile: reasonable passes across the first three groups and a measurement section full of unknowns. They checked ChatGPT once, saw their name, and stopped looking. Given SparkToro's finding that identical runs agree less than 1 percent of the time, that single check carries almost no information. The optimist does not need more content; they need a baseline, which is the cheapest item on the entire checklist to fix.

Whichever profile you match, resist the urge to fix everything at once. The checklist is a diagnosis; the treatment is one bottleneck at a time, verified by the recheck in check 30, because a loop that verifies beats a sprint that assumes.

Some fashionable items are absent on purpose. llms.txt is the loudest: large-scale 2026 studies spanning hundreds of thousands of domains and hundreds of millions of bot events found no measurable citation lift from the file by itself, so it earns a footnote rather than a check. Add one if you like, it costs minutes, but never let it substitute for the crawl access and structure checks that actually gate retrieval. Prompt-injection tricks, hidden text addressed to the models, sit further down the list still: they are detectable, embarrassing when archived, and aimed at systems that update faster than the trick spreads.

Also absent: anything you would buy instead of earn. Paid placement in answer engines barely exists as of mid-2026, and the citation economics described in check 20 reward the slower route of being genuinely present in the sources engines already trust. Incentivized reviews and undisclosed community seeding belong in the same bin, since platforms prosecute both and engines quote the prosecution. The checklist is boring by design, because every check on it survives an engine update, an algorithm leak and a journalist reading your robots.txt.

Groups one through three are human work: log reading, editing, profile claiming, honest participation. Group four is machine work, and doing it by hand is how measurement quietly stops happening. Reachroller runs the entire measurement group as designed here: a fixed question list run on schedule across engines, mention rates with branded questions excluded, every raw answer and its citations stored as receipts, trends per question and per engine, and a recheck workflow after each fix. When a question is lost, it generates the fix page publish-ready, with slug, title tag, meta description and schema, which turns checks 9 through 16 into a template instead of a memory. Starter is $29 per month for 400 credits and 25 tracked questions; the honest caveat is that ChatGPT tracking is live today and the remaining engines are rolling out. How that stacks against other trackers is covered in the best AI visibility tools.

The sequence for your next sprint, then: run the audit, fix crawlability the same day, baseline your questions with a Reachroller check, and only then write, with checks 9 through 16 taped above the keyboard. Four weeks later the recheck tells you what the sprint actually earned, in mention rates rather than vibes. The free check gives you three days, 50 credits and every feature, no card, which is enough to complete the measurement group of this checklist before you write a single new page.