Rule out the three basics first
Start here because these are the most common and, when present, invalidate everything downstream. Skipping this step is the most frequently wasted effort in the whole process.
- Is robots.txt blocking the major AI crawlers
- Can target pages be fetched at all, or do they depend on JavaScript rendering with no server-side fallback
- Do pages return real content, or a login wall, captcha or geo block
- The first two are directly checkable with the crawler check.
Then ask whether the content can be quoted
Being fetchable is not being citable. Retrieval systems need to assemble part of your page into an answer. If the page is one block of promotional prose, there is nothing cleanly extractable.
- Is the conclusion stated first, within the opening lines
- Is there an explicit question-and-answer structure, or only product copy
- Are key facts specific enough to verify: numbers, dates, versions, conditions
- Does each second-level heading work as a standalone, quotable claim
Check whether the question was the right one
'AI never mentions us' is often accurate and still not a problem to fix. Buyers may not ask about you that way at all. They may ask 'how do I solve problem X' and never use a brand name.
Compare against real sales conversations. If your product name rarely appears in how buyers actually ask, being absent from AI answers is expected rather than a defect.
Only then question the approach
Re-examine method and investment after the layers above are clean. The order matters: a large share of cases that look like 'the method does not work' turn out to be an execution gap in one of the first two layers.
- Was work limited to a few pages while the pages that drive decisions were untouched
- Was budget spread thin across many low-value pages
- Was a baseline ever established, leaving no way to tell whether anything changed
- Was a conclusion drawn in under one full cycle
Triage in the order of crawler access, quotability, question framing, then investment. That order filters out most false conclusions. Sources reviewed on 2026-10-17. This article records a method with verifiable boundaries; it does not claim that any search or AI platform will index, cite or recommend anything.