We have measured 123,040 real small business websites across 31 trades and 11,088 towns. Of the newest 20,000 read in full, 96.9% carry no machine-readable answers to the questions their own customers ask. Not answered badly. Not answered at all, in any form software can parse.
When someone asks an AI assistant for a local business, the answer is short. Across 4,454 assistant answers we have recorded, an answer names 3.7 businesses on average. Not a page of results to scroll. A few names, and everyone else is invisible.
An assistant deciding who to name leans on what it can actually read and verify. A page that answers its customers’ questions in a structured, machine-readable form gives an assistant something to stand on. A page that only looks good to a human gives it nothing.
Our own monthly measurements point the same way: in evaluation rounds run on 18 August and 20 August 2026, the businesses assistants named scored 15 to 20 points higher on readiness checks than the businesses they ignored. That is an association, not proof of cause, and we say so everywhere it appears. It is also the pattern anyone selling certainty should be asked to show.
The questions a business answers on the phone ten times a day, written on its website, each with a short answer, and marked up so software recognises them as questions and answers. A bakery answers “do you take custom orders”; a plumber answers “do you charge a call-out fee”. No redesign, no new platform, usually an afternoon of work.
The gap is not a technology problem. These sites were built to be read by people, and they do that job well. Something else is reading them now, and deciding who gets recommended.
Business listings come from OpenStreetMap. We keep the ones with a website and read each site, checking for structured question-and-answer content among other machine-readability practices. Sites we could not read are counted, not dropped, because dropping them would flatter the figure. We know of 413,546 businesses and are working through them, re-reading everything every three months, and no figure is published from fewer than 30 businesses.
The figures on this page are computed from the live measurement each time it renders, so the numbers you read are current, not the numbers from the day of first publication. To cite this finding: The FAQ gap, Faindable Research, first published 23 August 2026, faindable.xyz/research/faq-gap.
measures your own site against the same checks, free, in about a minute. Get your score, or see the full research page.