Most GEO checklists are opinion lists. This one has twelve items, and every item comes with something you can verify on your own site this week — a file you can open, a request you can run, or a number you can look up. If an item cannot be verified, it does not belong on a checklist.
Work through them in order. The first two are prerequisites: no amount of content work matters if the crawler is blocked or the answer only exists after hydration.
The 12 items
1. Let the answer-engine crawlers in
Generative engines can only cite pages they are allowed to fetch. Many sites inherited a robots.txt that blocks GPTBot, ClaudeBot or Google-Extended from a template or a security review nobody revisited.
How to verify: Open your own /robots.txt and read every Disallow line against the documented crawler tokens. Then check your logs for at least one successful 200 from each engine in the last 30 days.
2. Serve the answer in server-rendered HTML
Crawlers that do not execute JavaScript see only what the server returns. Content injected after hydration is invisible to them.
How to verify: Fetch your page with curl and no JavaScript, then search the output for the sentence that answers the page's core question. If it is missing, the crawler cannot see it either.
3. Answer one question per page, in the first 100 words
Assistants extract passages, not pages. A direct answer placed early is far more quotable than one that arrives after six paragraphs of context.
How to verify: Read the first paragraph aloud. If it does not answer the title as a complete sentence, rewrite it.
4. Use question-shaped headings
Headings that match how people phrase questions give the extraction step a clean boundary around the relevant passage.
How to verify: List your H2s. At least half should be answerable in two sentences without reading anything else on the page.
5. Ship valid structured data
FAQPage, HowTo, Article and Organization markup make the meaning of a page machine-legible and are cheap to maintain.
How to verify: Run the page through Google's Rich Results Test and confirm zero errors, not just zero warnings.
6. Publish and maintain a real sitemap
A sitemap that lists every public URL with a lastmod is the fastest path to discovery for new content.
How to verify: Request /sitemap.xml, confirm it returns 200 with XML, and spot-check that your newest page is present with an accurate lastmod.
7. Keep one canonical URL per topic
Duplicate or parameterised variants split the signal and make it ambiguous which URL an engine should reference.
How to verify: Confirm every page emits a self-referencing canonical and that no two pages target the same question.
8. Date your content and keep the dates honest
Freshness is disproportionately important for anything with a year in the query. Fake refresh dates get you nothing and cost trust.
How to verify: Check that your visible published date matches the datePublished in your structured data, and that updates carry a real change.
9. Make claims sourceable
Passages with a linked, checkable source are safer for an engine to reuse than unsupported assertions.
How to verify: Count the numeric claims on the page. Every one should have an outbound link to a primary source, or be deleted.
10. Fix your internal link graph
Engines follow internal links to understand which page is the hub for a topic and which are the supporting details.
How to verify: Confirm each cluster has one hub page linked from every member, and that no member is orphaned.
11. Keep the page fast and stable
Slow or unstable pages get fetched less thoroughly, and human referral traffic from assistants bounces hard on a slow first paint.
How to verify: Run Lighthouse on the page and confirm the performance and SEO scores are where you want them before you optimise anything else.
12. Measure crawls, citations and referred sessions
GEO without measurement is decoration. You need crawl coverage, cited landing pages, and AI share of visits per engine.
How to verify: Confirm you can name, today, the three pages assistants fetch most and the three that send you the most AI referrals.
What to do after the checklist
Items 1, 2, 5 and 6 are one-off engineering work. Items 3, 4, 8, 9 and 10 are editorial habits that only pay off if they become defaults in your writing process. Item 12 is the loop that tells you whether any of it worked — see the AI citation tracking pillar for the measurement model and the share-of-voice guide for the executive-facing metric.
One caution on tooling: a checklist that tells you to add llms.txt, buy a prompt-simulation seat and rewrite every page in the same week is selling activity, not outcomes. Do items 1 and 2, verify against the crawler registry, then measure for a month before you spend anything. Google’s structured data documentation remains the best free reference for item 5.