Building a GEO Strategy From Scratch

One, check the crawl layer: a log analysis (request volume, status codes, response sizes per crawler) and confirmation of access status. The test is that retrieval crawlers receive complete HTML.

Two, build the question set and baseline: a frozen question set, layered by buying stage and separate per language, plus baseline data with its definition. Everything later is attributed against it.

Three, compute the page matrix: the enumerable object list, page types per object, and the derived page ceiling. Page count is derived, not declared.

Four, publish in stages: a first batch sampling each page type, plus the technical groundwork — routes, structured data, internal links.

Five, review: a band-movement table and at least one decision to cut.

Off-site and entity signals can run alongside steps two and three — third-party records and description consistency do not depend on content.

But off-site distribution should wait for something reportable. Publishing company news with no substance only occupies a publication record.

Long-tail exposure at the decision stage can land within the first two stages; head terms depend on accumulated authority and pay off later. State that order at the outset, or the first few months read as failure.

AI referral volume is small early on — cross-site studies average around 1% of traffic — so early tests should be band movement and mention rate, not referral volume.