Google's AI is now choosing where people eat

A customer opens Google Maps on a Friday night and types something like: "a cosy Italian with a table for two around 7, not too far from the city." A couple of years ago that query returned a scrollable list of ten or twenty results, and the diner picked through them by eye — photos, star ratings, a squint at the menu. Today, Google's new Gemini-powered Ask Maps reads that whole sentence, cross-references reviews, photos and saved places, and hands back a short, curated answer. Sometimes it's two or three places. Sometimes it reads like it's betting hard on just one. Either way, the diner isn't scrolling ten links anymore — they're reading whatever the AI decided to surface.

Ask Maps is live in the US and India right now, not Australia. But it's the sharpest signal yet of where Google is heading, and the shift is already showing up in how Australians search generally. The list is shrinking. The question worth asking isn't whether this reaches you. It's whether your venue is one of the places these systems can actually read.

Why it matters

Google announced Ask Maps in March 2026: type a natural-language query instead of "restaurants near me," and Gemini cross-references Google's database of 300+ million places and 500+ million contributor reviews and photos to generate an answer. A Google product manager described the output as "several options" that satisfy the query — not a guaranteed single pick, but a small, curated set drawn from a database most of us never think about being read by anything other than a human scrolling on their phone.

This isn't a fringe behaviour shift. In Australia, 12% of people now name an AI tool as their primary way of finding things online — double the 5% recorded a year earlier. Separately, an estimate from Roy Morgan-sourced data put 58% of Australians aged 14+ as having used an AI tool in an average four-week period in early 2026. A moderate-authority industry survey also suggests around a third of Australian diners discover new venues specifically via Google Search, out of the 95% who use some kind of online resource to find somewhere to eat. However you weigh the exact numbers, the direction is the same: more people are letting an AI read the internet for them and hand back an answer, rather than reading it themselves.

That's a problem if the AI can't read your internet presence properly.

What Google's AI is actually reading

Here's the part most venues get wrong: a Google Business Profile is necessary, but it was never the whole picture, and it's becoming less of one. AI Overviews are, in Google's own framing, "grounded by design" — they're synthesised from whatever Google's normal search index has already crawled, retrieved and ranked, not from some separate parallel data source. Ask Maps draws on Maps data specifically (reviews, photos, saved and searched places), and personalisation comes only from a user's Maps-linked history, not Gmail or other Google services, according to Google VP Miriam Daniel. Google hasn't disclosed exactly how it ranks or selects which businesses make the cut, and it hasn't announced any advertising inside the feature.

What that means in practice: these systems are pattern-matching against structured, crawlable information — a real web page with your hours, your menu, your address written in a machine-readable format, consistent across every place your business appears. An Instagram bio is not that. A caption isn't structured data an AI can reliably extract; it's a sentence written for a human scrolling a feed, not a system trying to answer "which of these 40 Italian places actually has a table at 7." If your hours live only in an Instagram highlight and your menu only exists as a photo of a chalkboard, an AI reading the web has nothing to parse.

SEO practitioners who've studied this converge on a similar shortlist of signals: structured schema.org markup, a genuinely complete Google Business Profile, reviews with real specificity and recency, and — critically — consistency of your name, address, phone number, hours and website across every platform that mentions you. This is practitioner analysis, not something Google has confirmed as an official ranking factor, but it lines up with everything Google has said about how these features are grounded. A mismatch between what your website says and what your GBP says doesn't just look sloppy to a human — it reads as a conflicting signal to a system trying to decide whether to trust you at all.

The Providence test: being loved isn't the same as being findable

The clearest evidence of this isn't theoretical. A comparative experiment testing "best restaurant in Providence" across Google, ChatGPT, Gemini, Perplexity and Bing found something uncomfortable: the restaurants that showed up weren't necessarily the best-loved ones. Some rooms that were full every single weekend appeared in zero AI answers across all five tools tested — because their web data wasn't structured or consistent enough for any of the systems to confidently surface them. Meanwhile, less adored spots with a claimed, complete profile and consistent structured data across the web did get recommended.

That's the whole argument in one experiment. The AI isn't ranking who makes the best carbonara. It's ranking who gave it something it could actually read.

A prioritised checklist for this week

None of this requires a rebuild. Most venues already have the raw information — hours, menu, address, photos — it's just scattered across a GBP listing, an Instagram bio, and maybe an old PDF menu nobody's updated since a price rise. In order of impact:

• Check your hours, address and phone number match everywhere. Your website, your Google Business Profile, any directory listing you're on — if one says you close at 9 and another says 10, that's a conflicting signal, not a rounding error.
• Get your menu onto a real page, not just a PDF or an Instagram photo. A page an AI can crawl and parse beats a scanned document every time.
• Claim and completely fill out your Google Business Profile — categories, attributes, photos, the lot. It's still the anchor these systems reference, even if it's no longer the whole story.
• Write with specifics, not vague marketing copy. "Best coffee in town" tells an AI nothing. "Single-origin filter, all-day brunch, dog-friendly courtyard" is the kind of specific, occasion-matched language these systems are pattern-matching a query against.
• Make sure your site is actually structured, not just pretty. A real page with proper markup for your business type, address and hours gives every AI system something concrete to ground an answer in — the same underlying idea as the schema.org LocalBusiness markup practitioners keep pointing to.

None of these are exotic. They're housekeeping — the kind of thing that's easy to let slide when you're running service five nights a week, but expensive to ignore now that a shrinking shortlist is doing the work a full page of search results used to do. If you want the deeper version of "why this keeps slipping" for venues already stretched thin, we wrote about it here.

What to ignore

Don't try to reverse-engineer the exact algorithm — Google hasn't disclosed one, and chasing a black box is a waste of a Tuesday. Don't treat an Instagram-only presence as good enough; it's a great channel for a human audience and a dead end for an AI trying to extract structured facts. And don't panic and rebuild your entire online presence from scratch. This is a data-hygiene problem, not an existential one — the businesses that showed up in the Providence test weren't the flashiest, they were the most consistent.

The honest pitch

This is, admittedly, close to home for us. Every site built with FastPage gets a proper LocalBusiness schema.org entity generated automatically — your hours, address, menu, cuisine and rating, written in the structured format these AI systems are grounded in — without you needing to know what "schema.org" even means. It syncs with your Google Business Profile so your rating shows up correctly, and it gives customers a real booking form instead of a DM you'll forget to reply to. We're not claiming any special pipeline into Ask Maps — nobody has one, Google hasn't published a ranking method — but a real, structured, consistent web page is exactly the raw material these systems are reading, and a bio link isn't.

If your venue's information is currently scattered across an old PDF, a GBP listing nobody's touched since opening, and an Instagram bio, get started with FastPage and have that sorted this week — or have a look at pricing if you want the numbers first.