Ask an assistant where to eat tonight and you get two or three names, not twenty links. If yours is never among them, that is not bad luck. It is usually one of a small number of causes, and most of them you control.
A PDF menu or a menu rendered as an image is invisible. The model cannot read your dishes, cannot tell that you do handmade pasta or a serious vegetarian tasting, and therefore cannot recommend you for any of it.
Menus belong in real text on a real page. This is the single most common cause and the easiest to fix.
Hours on your site, different hours on Google, a third set on a delivery platform, an old phone number on a directory nobody updated. Every contradiction lowers a model's confidence, and low confidence means it names a competitor it feels surer about.
Diners ask for a specific thing: somewhere quiet enough to talk, a place open late near the theatre, good gluten-free options, a table for eight on a Thursday. If your site never addresses those, you cannot be the answer.
You know these answers. Your host answers them nightly by phone. Write them down.
Models weight independent corroboration heavily. Local press, food writers, a neighbourhood blog, a genuine guide listing. One substantive mention from a credible source does more than a page of your own copy claiming excellence.
Yes, but volume matters less than people assume. Consistency and specificity in what reviews say tends to matter more than raw count, because specific language gives a model something concrete to ground a recommendation on.
Be careful. Most of the work is unglamorous: put your menu in text, make your hours agree everywhere, answer real questions on your site. Anyone promising guaranteed AI placement is selling something that doesn't exist.
Trace where it got it. Usually a stale directory or an outdated aggregator listing. Correct the source and the model updates on its next crawl.
Want to know what AI currently says about your business? I run a free 30-minute diagnostic.