Everybody's looking for pizza. Nobody can find yours.

It's 7:14pm on a Friday. A guy two miles from your shop opens his phone and types "pizza near me." His wife asks ChatGPT for "best pizza in [your city]." Their teenager pulls up a Reddit thread someone posted last August: "best slice in [city]?" They compare. They pick. They order.

Your pizza is great. That's not the question. The question is: of the four or five places they checked in those 90 seconds, how many of them know you exist?

For most independent pizza shops I audit, the answer is one. Sometimes two. Almost never four.

The Friday-night tournament

A hungry customer doesn't run one search. They run a tournament. Usually five gatekeepers, each checking different signals before showing your shop to the person on the couch:

Google Maps "Map Pack". The three businesses with pins at the top. Per Sistrix's public click-through data, the #1 Map Pack spot averages 28.6% CTR; #3 drops to 11%.
ChatGPT or Perplexity. "Best pizza in [your city]". BrightLocal's 2025 AI search study found AI tools cite Yelp in about 1 in 3 local-intent queries, plus "best of" listicles from local news, Eater, Thrillist, and Reddit.
Yelp app. Millennials still open Yelp first for restaurants. The data has been steady on that for ten years.
An old listicle from your local news. "Top 10 pizza in [your city] 2024". Cached forever. Still cited by ChatGPT.
r/[your city] from August. "Best slice in [city]?", 47 comments. ChatGPT reads Reddit constantly. About 40% of its local queries touch a Reddit page.

Each surface has a different gatekeeper. Your pizza being great is necessary. It's not enough.

What ChatGPT actually does when someone asks for pizza

A customer types "best pizza in Edison NJ" into ChatGPT. Behind the scenes:

ChatGPT searches the web via Bing.
It pulls 6-15 source pages.
It picks 3-6 to cite by name.
It writes the summary the customer reads.

The sources it pulls most often for restaurant queries:

Yelp pages with 30+ reviews and complete data
"Best of" articles from local news, Eater, Thrillist, ThreeBestRated, Expertise.com
Reddit threads (organic mentions, not your own posts)
TripAdvisor for travel-intent queries
Your own website, but only if it carries enough specific, structured content

What ChatGPT does NOT primarily use: your Google Business Profile data directly. Google data reaches ChatGPT through Bing's index, but it gets no special weight there.

Why your shop is invisible to half of these surfaces

Three patterns show up in almost every audit:

1. You're not on any of the local "best pizza" lists

This is the biggest one. Local journalists, food bloggers, and lifestyle sites publish "Top 10 Pizza in [city]" lists every year. The Aggarwal et al. GEO paper (2024) found ChatGPT over-cites listicles by a 3:1 ratio versus long-form articles. Listicles are structured, comparative, easy for a model to extract from. ChatGPT loves them.

If you've never been featured on a local food blog, never been added to an Eater "where to eat pizza in [city]" map, never applied to ThreeBestRated or Expertise.com, you don't exist in the source material ChatGPT is reading.

The fix: search "best pizza in [your city]" and look at the first two pages of results. For every listicle that shows up, find the submission process. Most are free. Some charge a small "verified business" fee. Apply to all of them in one afternoon.

2. Your Yelp page is incomplete or has wrong info

Yelp is the single most-cited source for local restaurant queries. That BrightLocal stat is the biggest signal in the dataset. If your Yelp page:

Has a different phone number, address, or hours than Google
Has fewer than 30 reviews
Hasn't been touched in over a year
Has no menu uploaded
Doesn't list "Pizza" as a category

…then ChatGPT sees stale or contradictory data and weights you down. AI tools handle inconsistency by trusting the source that's most internally consistent. That's almost never you if you've never claimed Yelp.

Fix: claim the page, match your NAP (name, address, phone) to Google character for character, upload your current menu, ask three regulars to review you this month.

3. Reddit and the local food blogs have nothing about you

For the share of ChatGPT queries that touch Reddit (about 40%), an absent community signal hurts. If r/[yourcity] never mentions your shop, ChatGPT has nothing to ground a recommendation in.

You can't pay your way onto Reddit. Astroturfing gets caught. Reddit's anti-marketing immune system is harsh and it will burn you. What works: when a regular asks "anyone tried [your shop]?" in r/[city], make sure your loyal customers see it and answer honestly. Organic mentions compound.

Same logic for your local food blogger. Invite them in. Buy them a slice. Don't pitch. Just be a place they remember when they write their next piece.

What about Google's Map Pack?

Still matters more than anything else by volume. Google search drives roughly 191 billion referrals per month versus around 1.1 billion for all AI tools combined (Searchable 2026). Don't ignore Google because ChatGPT exists. Optimize both.

Map Pack ranking for pizza shops comes down to four signals, in this order:

GBP category set correctly. "Pizza Restaurant" as primary. Not "Restaurant" or "Italian Restaurant"; those compete you against everyone in town. Pizza shops that switched from "Restaurant" to "Pizza Restaurant" routinely jump 4-8 spots in Sterling Sky's controlled studies.
Review velocity. Not total reviews. Velocity. Your top map-pack competitor is averaging 8 reviews a month; you're averaging 2. Close that gap and you move.
Photos updated weekly. Specifically interior, food close-ups, and a few customer-uploaded shots. Whitespark's 2026 Local Search Ranking Factors survey put photo recency back in the top-tier signals after several years of decline.
NAP consistency across the top 30 directories. Yelp, BBB, Apple Maps, Bing Places, TripAdvisor, OpenTable, Grubhub, DoorDash. Mismatches anywhere bleed trust.

The 14 things we check at localpicks cover both Map Pack and AI surfaces. See the full breakdown for restaurants →

How to know if it's working

Pick 10 prompts a hungry customer might actually type, and check monthly:

"best pizza in [your city]"
"pizza delivery near me [city]"
"where to get a good slice in [city]"
"best New York-style pizza in [city]"
"kid-friendly pizza in [city]"
"late night pizza [city]"
"wood-fired pizza [city]"

Run each in ChatGPT with web search on. Note which shops get cited. If your name isn't there, look at what was cited and read those pages. The gap is your roadmap.

We do this automatically (it's our 14th audit), but the manual version takes about an hour the first time and tells you exactly which surfaces to fix first.

FAQ

Can I pay ChatGPT or OpenAI to be cited? No. ChatGPT's web search results aren't pay-to-play. Optimization runs through earned sources (Yelp, listicles, Reddit, your own site).

What about Perplexity, Bing Copilot, Gemini? Same general logic, slightly different weights. Perplexity leans on News and Reddit. Gemini favors brands with strong Wikidata presence (Andrea Volpini at WordLift has good research on this). Bing Copilot weights Bing Places and your own site. We check all of them.

How fast does it move after I fix things? Yelp re-indexes within a week. Bing Places about the same. ChatGPT and Perplexity citation patterns shift in 30-60 days as their crawlers re-read updated source pages. Map Pack ranking changes from a category fix can show in 2-3 weeks.

Does menu schema on my site actually help? Yes, especially for specific-dish queries like "best Neapolitan in [city]" or "where to get a Detroit-style square". Schema.org Restaurant + Menu + MenuItem markup lets AI tools answer at the dish level instead of the shop level.

If you want a check on which prompts ChatGPT does and doesn't mention you for, run our free 5-minute audit. It's the same 14 checks we run on every pizza shop that signs up.