We audited 1,721 restaurant websites in Canada and counted what is missing. Here is how this trade scores — and what it gets wrong — when an assistant decides who to recommend.
From a corpus of 20,806 sites, rebuilt on 9 September 2026. Every figure here is measured, not estimated.
“Where should I eat tonight in Montréal?” is now typed into an assistant as often as into a search box, and the assistant answers with three names and no list to scroll. Being the fourth-best restaurant in town used to mean a place on page one; it now means not being mentioned.
The contrast is the whole story: 69/100 on conventional SEO, 6/100 on answer-engine readiness. These sites are not bad; they are optimised for the previous decade.
The site cites no outside source — no association, no supplier, no certification body. A model weighs a page that stands alone less than one that agrees with something it already trusts.
The questions customers actually ask — price, hours, area covered — are either missing from the site or written in a form an assistant cannot lift.
Text is split into fragments that make no sense on their own. An assistant quotes a paragraph that stands up out of context, and there is none.
Nothing on the page says when it was written or last changed, so a model has no way to prefer it over a competitor's older page.
The site says what the business does but not enough about how, for whom, or at what price to be the best available answer to anything.
The reputation lives entirely on Google and nothing about it is visible where a model reads the business.
Share of the 1,721 audited sites affected by each defect.
The audit is free and needs no account. You will see your numbers beside your trade's.
Audit my site →Or read the 1,721 audits one by one: the trade benchmark.