How New Guests Find Restaurants in 2026: 88% of Ad-Driven Orders Came From Strangers

At Marigold Maison in Phoenix, 375 of the 426 orders matched to August ads, 88%, came from people who were not searching the restaurant by name. At a Scottsdale steakhouse, 54.7% of $298,850 in ad-matched reservation revenue came from searches like “private dining Scottsdale” rather than the restaurant’s own name. On Roka Akor’s website, a visitor arriving from ChatGPT was as likely to take a tracked action as one arriving from Google organic search: 3.62% against 3.63%.

Three restaurants, three different channels, one pattern. The guests who grow a restaurant are strangers, and the ways strangers arrive are now measurable to the dollar. This piece lays out the numbers, what they say about where a marketing budget should go, and three checks you can run on your own restaurant this week.

What Discovery Revenue Is

Discovery revenue is money from guests who found your restaurant without already knowing its name: through a non-brand search, an AI assistant’s recommendation, or an ad shown to a cold audience, and then booked, ordered, or walked in. It excludes brand searches and repeat guests, who were coming anyway.

The distinction matters because most restaurant marketing reports blend the two. A Google Ads account that bids on the restaurant’s own name will show a spectacular return, because it is charging you for guests who had already decided. Strip those out and you see what the marketing actually found. The three restaurants below let us do that, because Bloom matches ad exposure and website visits to completed reservations and POS orders rather than to clicks.

Paid: The Strangers Pay the Bill

At both restaurants running Bloom’s Agentic Ad Manager, the majority of revenue verified against reservations and orders came from people who were not searching the restaurant’s name: 54.7% in Scottsdale and 88% at Marigold Maison. The channel that found them differed. The result did not.

In Scottsdale, $2,718.97 of Google Search spend over 90 days matched to $298,849.67 in completed reservations, about 110 to 1. The non-brand campaigns, built around occasions, private dining, brunch, and cuisine, produced 54.7% of that revenue at 71 to 1, on money spent entirely on people who had never typed the restaurant’s name. Brand search looks better per dollar. Non-brand search produced more dollars, and the only guests the account can take credit for finding.

Marigold Maison ran the same test through a different door. Its non-brand Google Search campaign returned $304 on $1,061 and was paused. Its Meta prospecting campaigns, shown to lookalike and broad Phoenix audiences, returned 35.8 to 1. Across both platforms, $978 in August ads turned into $30,688 in tracked sales, and 88% of the matched orders came from people the ads reached cold.

The lesson is not “buy Meta” or “buy non-brand search.” The Scottsdale steakhouse found its strangers on Google. The Phoenix Indian restaurant found them on Instagram. The lesson is that when you verify against the check instead of the click, strangers turn out to be the majority of what paid discovery earns, and the winning channel is a fact about your restaurant, not about the platforms.

Organic: AI Visitors Act Like Google Visitors

On Roka Akor’s website, visitors referred by AI assistants completed a tracked action 3.62% of the time, statistically identical to Google organic search at 3.63%, and ahead of paid search at 2.68%. A recommendation from ChatGPT is worth as much as a top Google result, and it is measurable today.

Roka Akor is a five-location Japanese steakhouse group that Bloom has optimized for search, answer engines, and voice. It ranks #1 on Google for “Japanese steakhouse Chicago” and third in San Francisco (SEMrush, September 2026), and prospects who ask ChatGPT for the best Japanese steakhouse in San Francisco consistently see it in the top three. Over the 90 days ending September 21, organic search brought 50,011 of 111,554 website sessions, 44.8% of all visits and the largest single channel. The full Roka Akor case study covers how the site got there.

The AI assistant channel is still small, 1,133 sessions, and that is exactly why the action rate matters. Operators tend to dismiss AI referrals as curiosity clicks. Roka’s data says the opposite: the person who arrives from an assistant has already been told this is the right restaurant, and behaves like it. Paid search visitors, who clicked an ad, acted less often than either.

AI Assistant Visits Grew 2.7x in One Summer

Roka Akor’s monthly website sessions from AI assistants rose from 152 in June to 387 in July and 405 in August 2026, a 2.7x increase in two months, with 277 more in the first three weeks of September. The channel is small, and it is compounding.

Two things drive a curve like that. Guests are asking assistants where to eat more often, and Roka’s website gives those assistants specific, verified facts to cite: real menu language, real hours, real review substance, consistent business data across every platform. Google’s own guidance on AI features says the same thing in different words: its generative features are “rooted in our core Search ranking and quality systems,” and the content it prefers is content created from first-hand knowledge. A restaurant that wins the Google result and the AI answer is usually winning them for the same reason.

The Brand Search Trap

Brand campaigns show the best return per dollar because they harvest demand that something else created. Judged on verified revenue, the prospecting and non-brand campaigns that created the demand are the growth engine, and cutting them starves the brand campaign a month later.

Here is the trap in numbers. In Scottsdale, brand search returned about $325 per $1 and non-brand returned $71 per $1, so an operator judging on return per dollar would move budget to brand. But brand spend was capped at $416.79 in 90 days by how many people searched the name, and the non-brand campaigns are what put the name in front of them. Cut the finders and the $325 shrinks with them. Growth lives in the channels that introduce you, not the ones that remind people you exist.

How to Measure It in Your Restaurant This Week

Three checks separate discovery revenue from brand revenue: split your search spend by brand and non-brand, compare the AI Assistant channel to organic in Google Analytics, and connect your ads to reservations or POS so returns are counted in checks, not clicks.

1. Check 1: Split brand from non-brand in Google Ads. Open the search terms report for the last 90 days. Add up spend and conversions for every term containing your restaurant’s name, then everything else. If you cannot produce that split in ten minutes, your account structure is hiding the answer. Ask whoever runs the account for two campaigns, brand and non-brand, so the split is permanent.
2. Check 2: Find the AI Assistant channel in Google Analytics 4. Since May 2026, GA4 groups referrals from ChatGPT, Gemini, Claude and similar assistants into a default channel called “AI Assistant,” with no setup required. Open “Reports,” then “Acquisition,” then “Traffic acquisition,” and compare that row’s key event rate to Organic Search. If it matches or beats organic, as it does for Roka, your AI visibility is already worth defending. One caveat: assistant visits that arrive without a referrer, from an in-app browser or a copied link, still land in Direct, so the AI row understates the channel.
3. Check 3: Count ad returns in reservations and checks. Send completed reservations and POS orders back to Google Ads and Meta as offline conversions, or have your platform do it. Until you do, every return figure you see is the ad platform grading its own homework with clicks. The full-stack attribution guide walks through each layer.

The restaurants growing in 2026 are not the ones with the most brand searches. They are the ones who can prove which strangers became guests, and buy more of them.

One Guest Data Layer Behind Both Channels

Bloom runs both halves of discovery from the same guest data: organic through restaurant SEO, AEO and voice services that make a website citable, and paid through the Agentic Ad Manager that plans and runs Google and Meta campaigns and counts only revenue matched to reservations and orders.

The reason the three findings above can be measured at all is that each restaurant’s guest data sits in one place. WiFi, POS, online ordering, reservations, website, reviews, surveys, and ad clicks flow into one guest profile, so a stranger who arrives from ChatGPT or a Meta ad can be recognized when the reservation completes and again when they return. Paid revenue attribution is live today. Organic revenue attribution is in beta. The restaurant advertising playbook covers the paid side in depth, and the restaurant SEO, AEO and voice guide covers organic.

Questions Operators Ask

Methodology and Sources

Scottsdale figures are from Google Ads reports for June 23 to September 20, 2026, with the “Bloom Purchase” conversion action matched to completed reservations; the restaurant ran no Meta ads in the period. Marigold Maison figures are from the published case study, August 2026, with orders matched independently by Google and Meta. Roka Akor figures are from the restaurant’s Google Analytics 4 property for June 23 to September 21, 2026, 111,554 sessions; “tracked action” is a GA4 key event on the site, and “AI assistants” is GA4’s default AI Assistant channel. Rankings are from SEMrush, pulled September 22, 2026. Industry return-on-ad-spend benchmarks are as cited in our restaurant advertising playbook. Named restaurants appear where a published case study exists. Results are from these restaurants and will differ at yours.