How a Brooklyn restaurant doubled AI-driven reservations in 4 months.

When someone asks ChatGPT "where should I eat in Park Slope tonight," the response is a list of three to five specific restaurants. There's no scrolling through ten blue links, no comparison shopping. Whoever is in that list of three to five gets the reservation. Whoever isn't, doesn't. The economics of AI search are more brutal in restaurants than in almost any other vertical — but the work to land on that list is also more tractable than in most other verticals. This is a case study of one restaurant doing that work over four months.

The restaurant in this case study is a 32-seat new-American spot that opened in 2023 in Park Slope, Brooklyn. The owner asked to stay anonymous because some of the numbers below are sensitive, so we'll call it "the restaurant" throughout. Every number, every tactic, every implementation detail is real. The pattern is replicable for any independent neighborhood restaurant in a competitive urban market — Brooklyn, Oakland, East Austin, Wicker Park, the Mission. The specifics adapt; the structure transfers.

The starting point (February 2026)

The restaurant was doing about 280 covers per week at an average check of $68, generating roughly $19,000 in weekly revenue. Most of that volume came from neighborhood regulars and walk-ins. Reservation flow split roughly: 55% direct (Resy), 25% Google reservation button, 10% Yelp, 10% word-of-mouth or walk-in. Detectable AI-search referrals? Effectively zero.

When we ran "best restaurants in Park Slope for date night" through ChatGPT in late February 2026, the response named six restaurants and the case-study restaurant wasn't one of them. Same query on Perplexity returned a TripAdvisor list page first and named five restaurants — also no inclusion. Google AI Overviews returned a "top restaurants in Park Slope" carousel pulled directly from Google Maps results; the restaurant ranked 12th on Maps for that area, so it was in the carousel but well below the fold.

Initial Reffed score: 38 out of 100. The diagnostic surfaced four big problems, in priority order:

1. The restaurant's website was almost entirely visual — a homepage photo gallery, a one-paragraph "about" section, a PDF menu link, and a Resy reservation button. The full menu existed only as a PDF, which AI crawlers can't reliably parse. There was essentially zero text content for retrieval to find.
2. The Google Business Profile was undermanaged — no posts in 9 months, no menu items uploaded to Google's structured menu tool, photos hadn't been added since opening, dish-level reviews weren't being responded to.
3. No Restaurant schema, no Menu schema, no MenuItem data. The site shipped only default WordPress Organization schema.
4. Zero coverage in food publications — no listicle inclusions, no neighborhood guides citing the restaurant, no food-blog reviews. The restaurant existed in Yelp and Google but nowhere else AI engines retrieve from.

Why restaurants are structurally easier than B2B SaaS

Before walking through the plan, a brief detour on why this vertical is more tractable. B2B SaaS GEO work fights against established competitors with millions of monthly visitors and years of content investment. Restaurants compete in hyperlocal pools — there are maybe 40 serious dining destinations within a 1-mile radius in most urban neighborhoods. Reaching the top 5-10 of those is achievable in a few months of focused work, where the equivalent in SaaS would take a year or more.

The other reason: AI engines weight directories heavily for "best X in [neighborhood]" queries. Google Maps, Yelp, TripAdvisor, Eater, and food publications feed retrieval. A B2B SaaS site has to win on its own authority. A restaurant can win partly by being in the right directories, the right neighborhood guides, and the right roundup articles. That's a much shorter list of levers to pull.

The 4-month plan

We structured the work in four monthly sprints. Each sprint had a defined technical track and an off-site track. The owner is busy running a restaurant — total time commitment was capped at about 4 hours per week, which she did on Tuesday mornings before service.

Month 1: Web foundation and the structured menu

The single highest-leverage change was getting the menu out of a PDF and into HTML. We rebuilt the menu page with each dish as its own card containing name, description (2-3 sentences with specific ingredients), price, and any allergen or dietary flags. Sixteen entrees, eight starters, six desserts, and the cocktail list — about 4,200 words of new menu copy. The owner wrote the dish descriptions herself; she knew the food better than any copywriter would.

On top of that HTML, we layered Restaurant, Menu, and MenuItem schemas. Each menu item got name, description, menuAddOn where applicable, and suitableForDiet flags for vegetarian, gluten-free, and dairy-free dishes. We also added nutrition for the dishes where she had nutritional data from her chef.

Why this matters more than it sounds: AI engines answering "best vegetarian restaurants in Park Slope" need to know which restaurants have substantial vegetarian options. Without schema and structured menu text, the answer comes from third-party sources — Yelp tags, Google Maps descriptions — that may or may not be accurate. With schema, the answer comes from your own structured data, which is more trustworthy and more cite-able.

The Reffed score moved 38 → 51 by the end of Month 1, but more importantly the restaurant started appearing in AI responses for some specific queries. "Vegetarian dinner Park Slope" started returning the restaurant in Perplexity in week 3. ChatGPT didn't pick it up yet — Bing indexing lag.

Month 1 cost: $0 in tools, plus 6 hours of the owner's time across 4 weeks.

Month 2: Google Business Profile and the local-pack lever

Google AI Overviews for restaurant queries lean heavily on Google Maps results. Improving the Google Business Profile (GBP) is the single most direct lever for that surface. We did six things:

• Uploaded the full menu using GBP's structured menu tool — every dish, every price, every description. This isn't optional in 2026; restaurants without menus in GBP get visibly down-ranked.
• Added 47 new photos across food, interior, exterior, and team. Most importantly, we tagged the food photos with the actual dish names from the menu so they show up when someone searches "[dish name] Park Slope."
• Wrote 12 weekly Google Posts in advance — seasonal menu changes, holiday hours, new cocktail launches, chef's pick of the week. Posts are a freshness signal for the local algorithm.
• Responded to every review from the past 18 months that hadn't been replied to. 89 reviews total. Response engagement is a measurable ranking signal in Google Maps.
• Added every Q&A visible on the GBP profile — about 15 owner-answered FAQs covering reservations, dietary options, parking, wheelchair access, kid-friendliness, dog policy.
• Fixed NAP consistency across 8 directory sites — Yelp, TripAdvisor, Yellow Pages, OpenTable, Resy, Eater, Foursquare, Apple Maps. The address format was inconsistent across these (some had the apartment number, some didn't, two had the wrong zip), which weakens entity matching.

By end of Month 2, the restaurant's Google Maps rank for "restaurants Park Slope" moved from 12 → 7. Maps rank is the single biggest input to Google AI Overviews placement for restaurant queries.

The Reffed score moved 51 → 64. ChatGPT started returning the restaurant for niche queries — "Park Slope restaurant with garden seating," "best new American Brooklyn under $80." Not the bigger queries yet.

Month 2 cost: $0 in tools, 5 hours of owner time. The Yelp/TripAdvisor cleanup was the most tedious work in the entire project.

Month 3: Editorial coverage and third-party citations

The most powerful citation signal for restaurants is being named in published food coverage. Eater, Time Out, Infatuation, Resy's editorial, neighborhood blogs, food substacks. AI engines pull from these heavily for "best X" queries because they're the editorial vetting layer the model trusts.

Three concurrent tracks here:

Outbound to local food writers. The owner identified 11 NYC food writers, journalists, and substack authors who covered Brooklyn restaurants. We drafted a short, specific pitch email — not "please review us" but "we're doing a seasonal menu change on April 1 and would love to invite you for a tasting, here's what's new and why it might be interesting." Six replied. Three actually came in. One wrote about the restaurant in their newsletter (~8,000 subscribers). Two ended up doing one-line mentions in a "spring menu roundup" article they were already planning.

Submitted to neighborhood guides. Brooklyn Magazine, Greenpointers (cross-neighborhood), Time Out New York, BKLYNER. We pitched specific angles — "10 Brooklyn restaurants with the best vegetarian tasting menus," "Park Slope date night spots that aren't Italian." Time Out included the restaurant in a vegetarian listicle in week 9.

Eater inclusion attempt. Eater NY is the most-cited food publication in NYC GEO retrievals. Getting included in their listicles is high-value but they're notoriously hard to break into. We submitted a press kit to one of their freelance contributors and got nowhere this round. Eater inclusion was the only meaningful goal we didn't hit.

By end of Month 3, the restaurant was appearing in ChatGPT responses for the headline query "best restaurants in Park Slope" (3rd or 4th of 5 mentioned, varying run-to-run). Perplexity ranking was now first or second for "Park Slope vegetarian" and "date night Brooklyn under $100." The Time Out listicle citation showed up in retrievals across multiple engines.

Reffed score moved 64 → 76. Month 3 cost: $0 in tools, but $312 in food and drink covered for the three food-writer tastings (real cost of goods, not retail). About 8 hours of owner time across the month, much of it on hosting the tastings.

Month 4: Long-tail menu content and seasonal positioning

The fourth month focused on capturing long-tail queries — the specific dish-and-context searches that AI engines field constantly but most restaurants ignore. "Where can I get [dish] in [neighborhood]." "Restaurants with [specific dietary need] in [neighborhood]." "Best [cuisine type] for [occasion] in [city]."

We added four blog-style pages to the site, each 1,200-1,600 words, each targeting a specific long-tail cluster:

1. "The Park Slope Date Night Guide: 7 Restaurants We Actually Recommend" — written from the owner's perspective recommending six other Park Slope restaurants alongside their own. AI engines weight first-person editorial restaurant recommendations heavily, and the goodwill of recommending competitors paid off in two of those restaurants reciprocating.
2. "What's New in Vegetarian Tasting Menus in Brooklyn (Spring 2026)" — seasonal trend piece that mentioned five other Brooklyn restaurants doing notable vegetarian work, plus their own offering. Same logic.
3. "Brooklyn Restaurants With Real Outdoor Seating (Not Just Sidewalk Tables)" — niche post that targeted a high-intent seasonal query. Listed 8 restaurants with actual gardens, patios, or rooftops.
4. "Why Our Mushroom Tartine Is Different (And Why Ours Costs $24)" — single-dish deep-dive on their signature dish. Specific, opinion-driven, exactly the kind of content AI engines cite when someone asks "what's a good mushroom dish in Brooklyn."

By end of Month 4, the restaurant was appearing in AI responses across about 15 distinct query patterns we tracked. Direct AI-search referral traffic to the website doubled compared to Month 1 baseline (though absolute numbers are still small — we'll cover that next). Reffed score: 76 → 84.

Month 4 cost: $0 in tools, 6 hours of owner time for the four blog posts (she wrote first drafts; we edited).

The results, plainly

Four months of work, total out-of-pocket cost: $612 (food and drink for journalist tastings, plus one round of professional photography for the new menu items). Total owner time: ~25 hours across 4 months, roughly 90 minutes per week.

Reservation flow measured against the February 2026 baseline:

• Direct Resy reservations grew 18% (455 → 537/month). This is the AI-driven indicator — people searching the restaurant by name after seeing it in an AI response, then booking direct.
• Google reservation button clicks grew 47% (95 → 140/month). This is the Google AI Overviews + Maps surface, where the restaurant moved from 12th to 4th in Park Slope rank.
• AI-search referrer traffic to the website doubled from 26 to 52 sessions per week. ChatGPT (chat.openai.com), Perplexity (perplexity.ai), and Google AI Overviews together accounted for ~85% of that referrer traffic.
• Weekly covers grew 280 → 340, a 21% increase. Average check stayed flat at $68. Weekly revenue: $19,040 → $23,120, an additional $4,080 per week.

Annualized: roughly $212,000 in incremental revenue against a one-time cost of $612 and ~25 hours of owner time. The payback period was measured in days, not months. But — and this is important — most of that revenue isn't directly attributable to AI search alone. The Google Maps rank improvement does most of the lifting on raw cover counts, and Maps rank is influenced by both traditional local SEO and the AI-search work intertwined. The cleanest AI-only metric is the doubling of AI-referrer sessions, which is small in absolute terms.

What would have failed

Two things we tried that didn't work, worth surfacing so you don't repeat them.

Paying for inclusion in roundup articles. Several "best of Brooklyn" content sites offered paid inclusion in their listicles for $400-1,200 per piece. We didn't pay, but I checked back on a few of them three months later — AI engines clearly weight these lower than editorially-curated sources. The paid lists rarely surface in retrievals. Skip them.

Adding extensive recipe-style content. Early in Month 1 we tried publishing a "How We Make Our Mushroom Tartine" recipe-style post, thinking AI engines would cite it for "Park Slope mushroom dish" queries. They didn't. Recipe content competes against a massive corpus of recipe sites and food blogs that have years of authority. The single-dish opinion piece we ended up with in Month 4 — short, opinion-driven, specifically anchored to the restaurant — performed much better than the recipe version would have.

What transfers to other restaurants (and what doesn't)

The pattern that transfers across most independent neighborhood restaurants:

• HTML menu instead of PDF, with full dish descriptions and dietary flags
• Restaurant + Menu + MenuItem schema (more impactful in this vertical than in most others)
• Comprehensive Google Business Profile maintenance — structured menu, weekly posts, review responses, photo uploads, NAP consistency across directories
• Editorial outreach to 8-15 local food writers and neighborhood publications, with specific pitch angles
• Four to six long-tail content pieces that recommend other restaurants alongside your own — this counterintuitively builds your own citation rate faster than self-promotional content

What doesn't transfer cleanly:

• Tourist-heavy markets (Times Square, Fisherman's Wharf, the Strip) — AI engines weight tourist guides differently and the directory effects are stronger
• Fast-casual or chain-affiliated concepts — the brand-level entity confusion makes individual-location work harder to translate into citation gains
• Markets with fewer than 5-6 food publications covering them — the editorial-coverage track has nowhere to land outbound pitches

The four-month timeline, summarized

If you're working through this yourself, here's the compressed version of what to do when:

Weeks 1-4: Get the menu out of PDF, into HTML, with full dish descriptions. Layer Restaurant + Menu + MenuItem schema. Don't worry about anything else yet.

Weeks 5-8: Comprehensive Google Business Profile cleanup — structured menu upload, photo refresh, weekly Posts schedule, review response sweep, NAP consistency audit across the 8-10 directories that matter in your city.

Weeks 9-12: Outbound to local food writers with specific pitches. Submit to neighborhood listicles. Pursue editorial coverage. Plan a tasting event for 2-4 writers if your costs allow.

Weeks 13-16: Four long-tail content pieces. Each one should be specific, opinion-driven, and reference other restaurants generously. The goodwill compounds, and the AI-citation gains are larger than self-promotional pieces.

Total budget for an independent restaurant: $500-1,200 in real costs (food and drink for tastings, maybe one round of new photography), plus owner time. The pattern works if you have a real product to build authority around. It doesn't manufacture quality where there isn't any.

Where Reffed fits

The audit step at the start of this case study — the diagnostic that surfaced the four big problems — is exactly what Reffed runs automatically. We crawl your site, check your Bing and Google index health, check Restaurant/Menu schema, audit your Google Business Profile state, and run a category-specific set of prompts against ChatGPT, Claude, Perplexity, Gemini, Copilot, and Google AI Overviews. The audit report orders findings by likely citation impact, not by "what's easiest to fix" — which means for most restaurants the top priorities are HTML menu content and GBP maintenance, not whatever the latest GEO advice on Twitter is talking about.

Run a free audit on your restaurant site to see where you stand. If you want the full playbook including the engine-specific tactics for each of the 6 major AI engines, the Quickstart course covers it ($147 founding).