Lane 1 · The answers themselves
You cannot repair a gap you have never looked at. Do this first and keep the file; it becomes the thing every later month is measured against.
- Prompt set written: 10 to 20 real buying questions, covering category, best-for, versus, how-to and pricing
- Run across ChatGPT, Perplexity, Gemini, AI Overviews, Copilot and Claude
- Screenshots dated; named / not-named and citations recorded per engine
- Login conditions noted, since answers differ between signed-in and signed-out sessions
Lane 2 · Access: fix before everything else
If the crawlers are blocked, every other lane is wasted effort. This is the one that is usually broken by accident, in a file nobody has read in two years.
- robots.txt fetched and read block by block
- Search discovery checked for OAI-SearchBot, PerplexityBot, Claude-SearchBot, Googlebot and Bingbot
- User-request retrieval checked separately for ChatGPT-User, Perplexity-User and Claude-User
- Model-development controls checked separately for GPTBot, ClaudeBot, CCBot and Google-Extended; Google-Extended does not control Google Search, AI Overviews or AI Mode
- Per-bot verdict written as allowed, blocked or unverified, with the reason; access is not reported as a citation, mention, recommendation or accurate answer
- Bot-protection and firewall layers checked for challenges aimed at AI crawlers
- Raw-source test: the real copy is present in the initial HTML of every money page, not injected later by script
- Sitemap present, listed in robots.txt, and actually containing the money pages
- No stray noindex and no misaimed canonical on any money page
- Ten-second access check run at optimize.suedeai.ai
Lane 3 · The money pages
An engine has to be able to lift an answer out of your page. The test is whether a claim still holds up once it is pulled out and dropped somewhere else, alone.
- A plain definition in the first paragraph of every money page
- Key claims pass the extraction test: true, complete and attributable when read alone
- Answers-first FAQ, phrased the way buyers actually ask
- A comparison table wherever the buying question calls for one
- Headings written as signposts rather than slogans
- Supported schema only (Organization, Product, FAQPage, Article, HowTo), validated, and matching the visible content
- Visible, truthful updated dates
- llms.txt shipped, with claims kept modest
Lane 4 · Evidence
Answer engines and skeptical buyers converge on the same signals: the ones that are expensive to fake. Count your claims against your receipts. Earning the third-party half of those signals is the AI PR agency lane.
- Claims-versus-receipts count run across the money pages
- Named authors with real bios, and Person schema on key content
- Every number carries a date, a source, or a stated methodology
- Disclosures present anywhere a skeptic would want one
- Review-platform presence accurate, and happy customers actually asked
- Offsite trail checked by asking an engine: "What do people say about [company]? Include sources."
Lane 5 · Entity clarity
Machines have to be able to tell that all of your surfaces are the same company. Being one consistent thing everywhere is most of this lane.
- One consistent organization name and description across every surface you control
- Organization and Person schema connecting the people, the product and the site
- "What do you know about [company]?" asked across engines, and the coherence of the answers graded
Then · The rhythm
Not a lane, but the habit that makes the lanes worth running. One reading is a data point; the trend is the finding.
- Repairs logged with their ship dates
- Inputs verified live before each repeat measurement: the schema block really in production, the deploy really not reverted
- A 30-minute monthly measurement on the calendar: same prompts, same file
- Deltas read as trend, displacement and input correlation, with single-month wobble ignored