Vertical Playbooks · established evidence
The Restaurant Discovery Stack: Reviews, Delivery Platforms, and the Off-Premises Majority
Restaurant discovery used to be a single question: where does the restaurant rank when a nearby guest searches for a place to eat. That framing no longer fits the data. Industry research places roughly three-quarters of restaurant traffic off-premises, in takeout, drive-thru, and delivery, which means the moment of choice increasingly happens inside a delivery app, a review aggregate, or a synthesized AI answer rather than on a classic results page. So being found is now a stack of distinct surfaces, each with its own evidence and its own rules: the star rating that independent restaurants feel more sharply than chains, the local map pack that still captures the largest share of local-intent clicks, the delivery-platform listing that can intermediate the order itself, and the AI answer that often disagrees with the map pack about who to name. Reading that stack layer by layer, and knowing which finding is established fact and which is still emerging, is the starting point for any restaurant that wants to be chosen.
Restaurant discovery is now a stack, not a single ranking
For most of the search era, a restaurant's visibility problem could be stated in one line: appear high when a hungry person nearby searches. That statement assumed the decisive event was a query typed into a search box and answered by a list of links. The structural data on how people actually eat has quietly dismantled that assumption.
The National Restaurant Association's 2025 research finds that roughly 75 percent of restaurant traffic now occurs off-premises, across takeout, drive-thru, and delivery, and that a substantial share of adults order delivery weekly. When the majority of transactions happen away from the table, the decisive event is no longer a single search result. It is a comparison made inside whichever surface the guest happens to be holding: a delivery marketplace, a maps app, a review page, or a chatbot answer.
The useful way to read this is as a discovery stack. Each layer, the rating, the map pack, the delivery listing, and the AI answer, is a separate gate with its own selection logic, its own governing evidence, and its own failure mode. A restaurant can pass one gate and fail the next. The analytical task is not to find the one lever that matters but to know what each layer is, how strong the evidence for it is, and where a specific restaurant is losing.
The off-premises majority reframes the question
The off-premises shift is the most established fact in this domain, and it is worth stating precisely because it does most of the reframing work. The National Restaurant Association's 2025 research reports that about three-quarters of restaurant traffic is off-premises and that roughly 37 percent of adults order delivery at least weekly. Off-premises is not a pandemic residue; it is the default mode of a large and durable share of restaurant demand.
What is less settled is where that off-premises demand transacts. Recent industry survey data shows consumer platform preference splitting, with a reported lean toward ordering directly from a restaurant's own app or site over third-party marketplaces, in the neighborhood of 46 to 58 percent depending on the study, while third-party marketplaces retain a large share and one player leads that third-party segment. The single most cited driver of platform choice in this research is promotions and loyalty rewards rather than brand loyalty alone.
Two operational implications follow directly, and only from established or clearly-labeled data. First, a restaurant's own bookable and orderable surfaces, its Google Business Profile, its site, its own ordering path, are not a nice-to-have alongside the delivery apps; for a meaningful and possibly majority slice of off-premises guests they are the preferred destination. Second, because the exact direct-versus-third-party split varies by study and age cohort, any confident claim that direct ordering has already overtaken the marketplaces should be treated as emerging, not proven.
The rating is a revenue lever, and independents feel it most
If the off-premises shift tells a restaurant where the decision happens, the reputation literature tells it what the guest reads at the moment of decision. The foundational study here is Michael Luca's analysis of Yelp, which used the platform's star-rounding as a natural experiment.
Luca found that a one-star increase in a restaurant's Yelp rating produced roughly a 5 to 9 percent increase in revenue. The finding that matters most for an independent operator is the second one: the effect was concentrated in independent restaurants and did not appear for chain-affiliated restaurants, plausibly because consumers already carry a quality prior for a known brand and so lean less on the review signal. Consumers also responded more strongly to ratings backed by more visible and more numerous review signals.
The implication is structural, not motivational. An independent restaurant has more to gain from reputation work than a chain does, because for the independent the rating is one of the only pre-purchase quality signals a guest can read. That is an economic asymmetry, and it is exactly why review acquisition and review response are treated in this vertical as measurable engineering rather than a vague reputation gesture. The caveat matters too: Luca's study is a 2011 restaurant study, later revised, and no equivalent large-sample replication across every high-consideration local vertical was located in this research pass, so the precise 5 to 9 percent figure should be read as established for restaurants specifically, not silently transplanted.
The AI answer and the map pack do not agree on who to name
The newest layer of the stack is the synthesized AI answer, and its most important property for restaurants is that it is not a restatement of the map pack. The two surfaces measurably disagree about who gets recommended.
BrightLocal's 2025 to 2026 research reports that a business ranking in Google's top local-pack results has less than even odds of also appearing in AI local recommendations, and describes visibility in ChatGPT's local recommendations as far harder to obtain than a Google map-pack ranking, citing a roughly thirtyfold visibility gap. The same research indicates that Google's AI Overviews and AI Mode draw primarily on the Google Business Profile as their local data source, with Yelp cited in roughly a third of AI local searches, used specifically for review synthesis.
This is single-vendor proprietary research, so it belongs in the emerging tier and warrants a second corroborating source before its exact figures are treated as fixed. But the direction is consistent with the broader classic-versus-generative divergence, and it carries a concrete instruction for restaurants: ranking in the map pack is not evidence of being named in the AI answer, and the two have to be measured separately. A restaurant that assumes its map-pack position guarantees its ChatGPT presence is making an inference the data does not support.
Why the profile and the menu carry disproportionate weight
If AI Overviews lean on the Google Business Profile and review aggregators lean on visible review text, then the machine-readable substrate a restaurant controls, an accurate profile with correct hours and categories, and a menu published as live text rather than trapped in an image or a PDF, is what these systems can actually quote. An engine cannot recommend a gluten-free pasta or a late-night patio it has no readable text for. This is an inference from the sourcing behavior above rather than an independently measured effect, and it is labeled as such.
Who is asking an engine now, and how much do they trust it
A discovery layer only matters in proportion to how many guests use it and how far they act on it. Both variables are moving, and both should be read with their evidence tiers attached.
On usage, BrightLocal's Local Consumer Review Survey reports that 45 percent of consumers had used an AI tool such as ChatGPT, Gemini, or Perplexity to find a local business recommendation in the trailing year as of its 2026 edition, up from 6 percent a year earlier. That single-source year-over-year jump is large enough to warrant independent verification and is labeled emerging. Pew Research Center's June 2026 survey, whose methodology is disclosed, provides the more conservative anchor: 49 percent of US adults now use chatbots at all, up from 23 percent in 2023, and 42 percent of chatbot users use them specifically for information search.
On trust, the picture is deliberately cooling. Pew's 2026 survey finds that only about 29 percent of US chatbot users trust the information those tools give them a lot or some. In parallel, general trust in online reviews has declined over several years from a mid-2010s peak of around 84 percent to roughly half of consumers today. The synthesis is not that AI discovery is either irrelevant or inevitable. It is that a growing share of guests now consult these surfaces while extending them limited trust, which raises rather than lowers the value of the signals that read as genuine: real reviews, accurate profiles, and specific, verifiable menu detail.
Reviews are a regulated asset, not a growth hack
Because the rating carries real revenue weight and guests trust it less than they used to, the temptation to manufacture reviews is structural. It is also unlawful, and that legal fact is part of the discovery stack, not separate from it.
The Federal Trade Commission finalized a rule, effective October 21, 2024, that makes fake and deceptive consumer reviews and testimonials a specified unfair or deceptive act, following its 2023 revision of the Endorsement Guides. The rule reaches buying, suppressing, boosting, or organizing reviews to distort what consumers think, and it applies to every reviewed business, not only to the review platforms. For a restaurant, this means that review acquisition has to be engineered to solicit honest reviews from real guests only, without gating for sentiment, incentivizing selectively, or suppressing the negative ones.
This is the point where the evidence and the honesty constraint converge. The same literature that proves reputation moves restaurant revenue also defines the boundary of how it may be built. Compliant review acquisition is not a concession that weakens the tactic; under a rule that penalizes manipulation and a public that already discounts reviews, authenticity is the only version of the tactic that survives.
Reading the stack: what to prioritize
Assembled, the evidence supports a clear order of operations without overstating any single layer. The off-premises majority is established and reframes the whole question toward the surfaces where guests actually transact. The rating's revenue effect is established for independent restaurants specifically, which makes compliant reputation work a high-return, high-confidence investment for exactly this profile. The map pack's dominance of local-intent attention is established in direction and emerging in precise magnitude, so it belongs in the priority set and should be measured spatially rather than assumed.
The AI-answer layer and the AI-usage surge are genuinely important and genuinely less certain. They rest partly on single-vendor research and large single-source swings, so a restaurant should build the readable substrate they reward, an accurate profile and a quotable menu, while measuring its actual presence rather than paying for confident promises about it. The direct-versus-third-party delivery split remains study-dependent, which is a reason to strengthen a restaurant's owned ordering surface without declaring the marketplaces defeated.
The operator's real question is not which of these is the future. It is which of these surfaces a specific restaurant is losing on today. That is an empirical question with a measurable answer, and it is the place any plan should begin.
The evidence
Key findings, with their sources
-
Roughly 75% of restaurant traffic now occurs off-premises (takeout, drive-thru, delivery), and about 37% of adults order delivery at least weekly.
established National Restaurant Association, 2025 research; NCR Voyix / Restaurant Dive industry survey data, 2025-2026.
-
A one-star increase in a restaurant's Yelp rating produced roughly a 5 to 9% increase in revenue, an effect concentrated in independent restaurants and absent for chains.
established Luca, M., "Reviews, Reputation, and Revenue: The Case of Yelp.com," Harvard Business School Working Paper 12-016, 2011 (rev. 2016).
-
Consumer platform preference for off-premises ordering splits, with a reported lean toward ordering directly from a restaurant's own app or site (about 46 to 58% depending on study) over third-party marketplaces; promotions and loyalty rewards are the top driver of platform choice.
emerging National Restaurant Association, 2025 research; NCR Voyix / Restaurant Dive, 2025-2026.
-
Local searchers click the local three-pack a plurality of the time, more than standard organic or paid, and the first map-pack position captures the largest single click share; pack businesses receive materially more traffic and user actions.
emerging Aggregated Google local-search behavior studies as reported via SearchEngineLand / industry local-SEO research, 2025.
-
A business ranking in Google's top local pack has less than even odds of also appearing in AI local recommendations, with a reported roughly thirtyfold visibility gap for ChatGPT versus the Google map pack; AI Overviews draw primarily on the Google Business Profile, with Yelp cited in about a third of AI local searches.
emerging BrightLocal, "AI Search Makes Local Listings More Important Than Ever" and "How AI Is Impacting Local Search," 2025-2026.
-
45% of consumers reported using an AI tool to find a local business in the trailing year as of the 2026 survey, up from 6% a year earlier; Pew finds 49% of US adults use chatbots (up from 23% in 2023) but only about 29% of chatbot users trust their output a lot or some.
emerging BrightLocal, Local Consumer Review Survey 2024/2026; Pew Research Center, "Americans and AI 2026," June 17, 2026.
-
The FTC's final rule making fake and deceptive consumer reviews and testimonials a specified unfair or deceptive act took effect October 21, 2024, and applies to every reviewed business, not only review platforms.
established Federal Trade Commission, final rule on the use of consumer reviews and testimonials, 16 CFR Part 465, effective Oct. 21, 2024; FTC Endorsement Guides, 16 CFR Part 255 (rev. 2023).
Calibration
What is proven, what is promising, what is unproven
| Evidence tier | Tactics | What the evidence says |
|---|---|---|
| established | Compliant review acquisition and response; accurate, complete Google Business Profile; publishing the menu as live readable text; owned ordering surface for off-premises guests | Off-premises majority (National Restaurant Association 2025); ratings-revenue effect for independents (Luca 2011, rev. 2016); FTC fake-review rule (16 CFR 465). |
| emerging | Spatial map-pack measurement across the service area; building the AI-readable substrate the profile and review aggregates reward; strengthening direct ordering against the marketplaces | Local map-pack click and action magnitudes (secondary-sourced industry studies); AI-vs-map-pack citation gap (single-vendor BrightLocal); 6% to 45% AI-discovery jump (large single-source swing). |
| contested | Declaring direct ordering has already overtaken third-party marketplaces; treating a map-pack ranking as proof of AI-answer presence | Direct-versus-third-party split varies by study and age cohort; map pack and AI recommendations measurably diverge, so one does not imply the other. |
Reference
Glossary
- Off-premises
- Restaurant demand fulfilled away from the dining room: takeout, drive-thru, and delivery. Industry research places roughly three-quarters of restaurant traffic here.
- Local map pack
- The boxed cluster of three local business results shown with a map on a local-intent search. It captures a disproportionate share of local clicks and actions.
- Discovery stack
- The set of distinct surfaces a guest may use to choose a restaurant, the rating, the map pack, the delivery listing, and the AI answer, each with its own selection logic and evidence.
- Third-party marketplace
- A delivery platform such as DoorDash or Uber Eats that intermediates ordering, and sometimes the listing itself, in exchange for commission, as distinct from a restaurant's own ordering surface.
- Google Business Profile
- A restaurant's primary owned local listing on Google. Research indicates it is the main local data source AI Overviews and AI Mode read when synthesizing local recommendations.
Straight answers
Frequently asked questions
Is restaurant discovery now a delivery-platform problem more than a Google problem?
It is both, which is the point of reading it as a stack. With roughly three-quarters of restaurant traffic off-premises, delivery and marketplace surfaces clearly matter more than they once did. But the map pack still captures a large share of local-intent attention, AI answers draw heavily on the Google Business Profile, and the star rating drives revenue for independents. No single surface has replaced the others; the right question is which one a specific restaurant is losing on.
Do online reviews really change restaurant revenue?
For independent restaurants, the established evidence says yes. Michael Luca's Yelp study found that a one-star increase in rating produced roughly a 5 to 9 percent increase in revenue, an effect concentrated in independents and absent for chains, because a known brand already carries a quality prior that a small restaurant does not. That figure is established for restaurants specifically and should not be silently applied to other verticals.
If I rank in the Google map pack, will ChatGPT recommend me too?
Not necessarily. Vendor research reports that a top map-pack business has less than even odds of also appearing in AI local recommendations, with a large reported visibility gap between ChatGPT and the map pack. This is emerging, single-vendor data, but the direction is consistent with the broader divergence between classic and generative results. The reliable approach is to measure your AI-answer presence separately rather than assume your map-pack position covers it.
Should I focus on my own ordering site or the delivery apps?
Both, weighted by your own data. Industry surveys show a reported lean toward ordering directly from a restaurant's own app or site over third-party marketplaces, though the exact split varies by study and age group and the marketplaces still hold a large share. Strengthening your owned, readable ordering surface is a safe investment; declaring the marketplaces defeated is not yet supported by the evidence.
Can I ask happy guests for reviews, or is that against the rules?
You can ask, as long as you ask honestly. The FTC rule effective October 21, 2024 bans fake and deceptive reviews and prohibits buying, suppressing, boosting, or selectively organizing reviews to distort consumer perception. Soliciting genuine reviews from real guests, without gating for positive sentiment or incentivizing selectively, is both compliant and, given declining public trust in reviews, the only durable version of the tactic.
Provenance
Sources
- National Restaurant Association, 2025 research on off-premises dining and delivery behavior (established)
- NCR Voyix / Restaurant Dive, industry survey data on delivery platform preference, 2025-2026 (emerging on the exact direct-vs-third-party split)
- Luca, M., "Reviews, Reputation, and Revenue: The Case of Yelp.com," Harvard Business School Working Paper No. 12-016, 2011 (rev. 2016) (established)
- SearchEngineLand / aggregated industry local-SEO research on local map-pack click and action behavior, 2025 (established direction, emerging on precise magnitude)
- BrightLocal, "AI Search Makes Local Listings More Important Than Ever" and "How AI Is Impacting Local Search," 2025-2026 (emerging, single-vendor)
- BrightLocal, Local Consumer Review Survey, 2024 and 2026 editions (emerging on the 6% to 45% AI-discovery jump)brightlocal.com
- Pew Research Center, "Americans and AI 2026: Chatbots, Smart Devices and Views on Impact," June 17, 2026 (established, disclosed survey methodology)pewresearch.org
- Federal Trade Commission, final rule on the use of consumer reviews and testimonials, 16 CFR Part 465, effective Oct. 21, 2024; Endorsement Guides, 16 CFR Part 255 (rev. 2023) (established)ecfr.gov
Every figure above is attributed to a real, dated source and tagged with its evidence tier. Where a claim could not be verified to a primary source, it is not stated as fact.