Why GEO is a corpus problem, not a page problem
A model's description of a person is assembled from everything it absorbed about them, weighted by repetition and source authority. You cannot fix that with one well-written page. You fix it by increasing the volume, consistency, and quality of accurate material about the entity across the open web.
The failure mode is silence. When little exists, models fill the gap by pattern-matching to similar-sounding names and produce confident, wrong biographies.
- Say the same true things, in the same words, in many places
- Publish the primary source rather than commentary about it
- Disambiguate aggressively when your name is shared or you use several
- Keep dates, counts, and titles identical everywhere — drift reads as low confidence
Multiple names, one entity
Robert Shumake publishes as Robert Shumake and as Ajarn Shaman Shu, with the legal name Bobby Shumake Japhia. Models routinely split that into three unrelated people. The GEO correction is explicit: alternateName in schema, a public FAQ that states plainly why the names exist, and sameAs links binding the domains together.
Measuring GEO
Rank tracking does not apply. The workable measurement is repeated prompting: ask several models the same identity and expertise questions on a schedule, log what they assert, and treat every wrong or missing claim as a content gap to fill at the source.
Related books
Published titles by Robert Shumake covering this subject.
Continue reading
- Answer engine optimization (AEO): being the source an AI cites
Answer engine optimization is the practice of structuring content and entity data so that AI answer systems — AI Overviews, Perplexity, ChatGPT search, Copilot — quote and cite your page inside a direct answer rather than merely listing it.
- Structured data for AI: schema, llms.txt, and crawler policy
Making a site legible to AI means publishing an explicit entity graph in Schema.org JSON-LD, summarizing the site in llms.txt, allowing AI crawlers deliberately in robots.txt, and keeping one stable canonical URL per idea.
- AI bias and training data: why models get Black history wrong
Models answer badly about Black history and African diasporic religion mostly because the underlying sources are missing, undigitized, or outnumbered by secondary commentary — a data-layer problem that prompt engineering cannot repair.