The generative engine optimization (GEO) checklist

Content

1. You have a page for every comparison a buyer makes: "you vs each top competitor" and "alternatives to each top competitor". Check: search your own site for the competitor names.
2. Each page answers the question in the first 100 words, then shows the evidence. Check: read only the first paragraph. Does it answer?
3. Real prices, limits and trade-offs are stated, including where a competitor is better. Check: would a skeptical buyer trust it?
4. Claims carry a source and a date. Check: every number links to where it came from.
5. A "best [category] for [use case]" page exists for your three biggest use cases.
6. Pages have visible "last updated" dates, and the dates are true.
7. One topic per page, with a descriptive H1 and sensible H2s. Check: skim only the headings.
8. An FAQ block with real questions from sales calls, not invented ones.

Technical

1. robots.txt allows the AI crawlers you want: GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot, ClaudeBot, Google-Extended. Check: open /robots.txt.
2. Key pages return 200, are server-rendered and load without JavaScript. Check: `curl` the page and look for your text.
3. Canonical URLs are set and consistent. Check: view source.
4. JSON-LD for Organization and WebSite on every page, FAQPage where there is an FAQ, Article on guides, BreadcrumbList on nested pages. Check: parse the `ld+json` blocks; they must be valid JSON.
5. An XML sitemap lists every indexable page with a true lastmod. Check: /sitemap.xml.
6. An /llms.txt describes you in plain language and links your best pages. Optional, cheap.
7. Pages are fast on mobile and the layout does not jump. Check: Lighthouse.
8. Your brand name, domain and a one-line description are identical across your site, LinkedIn, Crunchbase and review profiles.

Third parties

1. You have listed the URLs cited for your top 20 questions and know who owns each.
2. You are present, accurately, on every review site that appears in those citations (G2, Capterra, Trustpilot, and so on).
3. A real-customer review ask goes out after onboarding, every month. Nothing bought, nothing written for them.
4. For each listicle that ranks for your category and does not name you, one honest, factual outreach email has been sent.
5. Your founders or team answer relevant community questions under their real names, with affiliation disclosed.
6. Wikipedia, Crunchbase and similar profiles are accurate. You may not edit your own Wikipedia page; fix the sources it relies on instead.

Measurement

1. A fixed list of 20 to 50 buyer questions, grouped into category, comparison and use case.
2. Run on at least ChatGPT with web search, Perplexity and Gemini. Check: one engine is not a measurement.
3. Recorded per answer: named or not, position, competitors named, sentiment, cited URLs.
4. Re-run on a schedule. Single runs are noisy: in our own runs the same company was named in 5, 5 and 2 of the same 10 questions on one day.
5. Cited and recommended are tracked separately.
6. A 90-day target agreed before the work starts, and a baseline from day one.

Missing most of these? Start with the first three technical items and the comparison pages. They take a week, and everything else builds on them. Want a second pair of eyes? Run the free audit or book a call.