How Do You Get a Small Business Cited by ChatGPT Instead of Just Ranked by Google?

AEO and GEO — answer engine optimization and generative engine optimization — are the practice of making a business the thing ChatGPT, Claude, Perplexity, or a Google AI Overview cites when someone asks a real question, not just another link in a Google search result. I ran the full checklist on widergood.com itself before recommending it to anyone, the same rule that applies to everything else here.

The first pass scored 8 out of 20 on that checklist — ten areas, two points each, including the live citation test that the free check on this site can't run because it needs a person at the keyboard. Four test queries, phrased the way an actual prospect would ask them and never using the business name, returned zero citations. The businesses that did show up became the real competitor set — not whoever ranks well in ordinary search, which turned out to be a different list entirely.

Most of what raised the score afterward was mechanical, not creative. robots.txt already allowed the nine named AI crawlers — a lot of sites block them by accident with an old blanket rule and never notice. ProfessionalService, Person, and FAQPage schema went in and passed Google's Rich Results Test, so an AI engine reading the page gets structured facts instead of guessing from paragraphs.

The FAQ content mattered more than any of it. Eighteen questions, phrased the way people actually ask them, each answered directly in the first sentence — that's the block an AI engine quotes back almost verbatim when someone asks the underlying question later. Page titles got the same treatment: what the business does and who it's for, not just a name.

None of this replaced good positioning. It made positioning that already existed legible to a machine instead of only to a person scrolling the page.

Updated 4 September 2026. Two things above have moved since this was written, and both are worth saying plainly rather than quietly editing.

The 20-point checklist is retired. It scored ten areas two points each, which meant a site could lose the same point for a missing canonical and for a missing testimonial — as if an engine being unable to read the page at all were the same size of problem as a thin proof section. The scan now runs the 42 checks a machine can read from a site and reports how many pass, one count, with the same count beside each area so the numbers add up. Behind that, every check is weighted by how well it is evidenced — what a platform documents outranks what a published spec checks, which outranks what practitioners agree on — and that weighting decides the order of the fixes rather than the headline. widergood.com passes every check on that list except the testimonial ones, which cannot be invented.

Nine named AI crawlers became twenty-two. That is not the site changing; it is the crawlers changing. Between them OpenAI, Anthropic, Google, Apple, Meta, Amazon, Microsoft, DuckDuckGo and Common Crawl now publish more than twenty distinct agents, and the ones that matter most are not the training crawlers everyone argues about — they are the retrieval crawlers that fetch a page to answer a question while someone is waiting. Blocking one of those hands the citation to whoever left it open. Any list of bot names is out of date within a couple of months, which is why the scan keeps its list as data with a vendor documentation link and a verification date beside every entry.

The newer half of the work is that a site now has a second reader. Not the engine that cites you, but the agent that acts for someone: it wants your content as Markdown when it asks for it, a robots.txt that names it, and a machine-readable description of anything it can call. Almost no small business has any of that today, which is exactly why it is cheap to be early.