How to Use This Checklist
The checklist is organized into four dimensions matching the components of the GEO score. Run them in order, because when an earlier item fails, the return on later work is largely cancelled out.
Each check gives two things: the pass criterion, meaning how to confirm it, and the usual failure reason, meaning why most sites lose points there. Record pass or fail only; scoring is unnecessary.
After one pass, sort failures by fix cost from low to high and start with the cheap ones. Most sites clear eighteen to twenty-two items on the first run, with the remainder concentrated in structured data and content trustworthiness.
Technical Access: 8 Checks
One, robots.txt allows retrieval crawlers: confirm OAI-SearchBot, PerplexityBot, and Claude-SearchBot are not matched by any rule. The usual failure is pasting a block-everything list found online.
Two, WAF and CDN are not blocking: check status-code distribution for these crawlers; over five percent 403 or 429 counts as a fail. The usual failure is a default bot-protection rule misfiring.
Three, core content renders server-side: load the page with JavaScript disabled and check whether price, specs, and body text are visible. The usual failure is a site-wide SPA architecture.
Four, key pages return 200 without redirect chains: check for more than three hops. The usual failure is leftover redirect chains from historical URL migrations.
Five, load performance is adequate: primary above-the-fold content should complete within three seconds. The usual failure is unoptimized images and render-blocking third-party scripts.
Six, sitemap.xml exists and is current: confirm it covers all indexable pages with accurate lastmod. The usual failure is it has not been regenerated in months.
Seven, llms.txt exists with valid links: confirm a 200 response, correct Markdown structure, and absolute URLs.
Eight, no AI-specific blocking elsewhere: inspect HTTP headers and firewall rules for explicit AI interception policies.
Content Trustworthiness: 8 Checks
One, core claims carry verifiable data: does each major conclusion give a specific number and source? The usual failure is qualitative description dominating quantitative.
Two, primary sources are cited: do you link to the original report, paper, or official dataset rather than to secondary coverage? The usual failure is citing only industry media.
Three, opinions are attributed: when quoting someone, is a name and title given? The usual failure is writing experts say.
Four, author identity is traceable: does each content page have a named author with a background page? The usual failure is the author being the site name or an editorial desk.
Five, dates are visible: are publish and last-updated dates shown? The usual failure is no date displayed anywhere.
Six, parameters and units are explicit: do specs, prices, and ranges state units and conditions? The usual failure is unverifiable phrasing like battery lasts a long time.
Seven, conclusions match evidence: is anything claimed beyond what the data supports? The usual failure is presenting correlation as causation.
Eight, comparative content is fair: does it acknowledge competitor strengths? The usual failure is listing only your own advantages.
Entity Consistency: 7 Checks
One, Organization markup is complete: name, alternate names, industry, logo, official site. The usual failure is only name being present.
Two, sameAs is configured: does it link to authoritative profiles such as Wikipedia, LinkedIn, and Crunchbase? The usual failure is an empty sameAs, which is the most overlooked item and among the most impactful.
Three, brand name is consistent everywhere: is the same form used across site, social profiles, and third-party directories? The usual failure is mixing Chinese and English names or abbreviations.
Four, product naming is stable: does one product use one name across all pages? The usual failure is old and new names coexisting after a redesign.
Five, author entities are unified: does the same author appear with identical name and title across articles? The usual failure is inconsistent name forms.
Six, structured data types are correct: does the type match the page's nature? The usual failure is marking every page as Article.
Seven, markup matches visible content: are marked-up values actually present on the page? The usual failure is declaring information the page does not contain.
Monitoring and Iteration: 7 Checks
One, a fixed prompt set exists: are ten to fifteen business-relevant prompts defined? The usual failure is inventing prompts each time, which makes results incomparable.
Two, sample count is fixed: does each round use the same number of samples? The usual failure is single sampling, which cannot distinguish noise from trend.
Three, multiple AI platforms are observed: does coverage include ChatGPT, Perplexity, and Gemini? The usual failure is testing one platform only.
Four, competitor performance is recorded: do you track key competitors' visibility in the same window? The usual failure is watching only yourself, making industry-wide movement indistinguishable from your own.
Five, metric definitions are explicit: are mention rate, recommendation rate, and citation rate defined consistently? The usual failure is conflating brand mention with page citation.
Six, a review cadence is set: is there a fixed weekly or monthly review? The usual failure is measuring only when someone remembers.
Seven, changes are logged against results: does each modification record when it happened and what changed? The usual failure is frequent edits with no way to attribute effects.
Turning the Audit Into a Routine
A one-off audit has limited value. Technical access decays as CDN and WAF configurations change, and content trustworthiness fluctuates as pages are updated. Use this checklist as a quarterly baseline; each pass takes roughly one to two hours.
About a third of the items can be automated: reachability checks for technical access, syntax validation for structured data, and cross-platform sampling statistics. The remainder, content trustworthiness and entity consistency, requires human judgement and is exactly where the largest gaps sit.
The site audit in Broccoli AI GEO reports along comparable dimensions and ranks action items by expected impact against implementation cost. Using automated output alongside this manual checklist covers the semantic blind spots tooling cannot reach.