Which Business Listings Actually Feed AI Engines

One number reframes the whole question. In our scrape, the directory threebestrated.ca was cited 116 times by Claude and almost never by anyone else, while google.com was cited 130 times by ChatGPT and barely registered inside Claude. Two listings, two engines, two parallel worlds that hardly touch. The standard advice tells you to list your business on the same ten directories and assures you they all "feed AI." Our citation data says that advice is built on a false premise. The engines do not share a feed. They read largely different webs, and the listing that wins one can be invisible to the next.

This article is the map the checklists skip. Rather than a flat roster of sites, it pairs each major listing with the specific engine that actually pulled from it, drawn from a first-party, compute-verified dataset rather than assertion. We cite the real domains each AI named, then triangulate that local-Ontario evidence against three large independent studies, the Yext analysis of 17.2 million citations, BrightLocal's 800-search study, and Local Dominator's 267,280-citation report, that reach the same conclusion from far larger samples. The takeaway is uncomfortable for anyone selling a universal listing package: there is no universal listing. There is only the right list for the engine where your customers ask.

The listing-to-engine map, in one table

Here is the whole argument before the prose unpacks it: each engine has a listing it reaches for first, and those listings barely overlap. The table below pairs the source layer with the engine that leaned on it hardest in our Ontario scrape, alongside the per-engine citation counts that earned each pairing its place. Read it as a routing diagram. The same business name has to travel four different roads to reach four different answers.

Which listing feeds which engine (Formative Digital scrape, 1,732 citations, nine Ontario cities, May 2026)

Engine	The listing it feeds on	Top domain & times cited	What the listing rewards
ChatGPT (OpenAI)	Google Business Profile, then your own website	google.com · 130	Complete Maps profile, reviews, consistent NAP
Claude (Anthropic)	Editor-curated directories	threebestrated.ca · 116	Being selected onto a human shortlist
Perplexity	Review aggregators (Yelp wired via Fusion)	homestars.com · 17	Breadth across many review surfaces
Gemini (Google)	No public listing, a grounding pipe	vertexaisearch.cloud.google.com · 384	Crawlable, schema-marked content

The counts are not interchangeable, and that is the point. Gemini's 384 is not a publisher at all; it is Google's Vertex AI Search wrapper, cited more than any real directory in the dataset because it masks the pages underneath. ChatGPT's 130 is Google's structured local data flowing through google.com. Claude's 116 is one Canadian directory editor's curation. Perplexity's lead domain sits at a modest 17 because it spreads its citations widest of the four, so no single source dominates. Four engines, four feeds, and a methodology note worth stating plainly: across 583 distinct domains over 44 city-and-vertical cells, only 95, or 16.3%, were cited by two or more engines. The other 83.7% were unique to one. We unpack that headline in the cornerstone study on why AI engines recommend different local businesses; this page zooms in on the listings themselves.

Two pieces of established research explain why the feeds matter so much. The peer-reviewed GEO paper by Pranjal Aggarwal, Vishvak Murahari, Ameet Deshpande and co-authors (arXiv:2311.09735, KDD 2024) showed that content-level and citation-level signals, reviews, statistics, quotable structured detail, lift a source's visibility inside generative answers by up to 40 percent, while page rank does not. And Kevin Indig's early-2026 Growth Memo found that about 44 percent of AI citations come from the first 30 percent of a page. A directory that opens with a clean ranked list hands the engine exactly the early, extractable content it rewards, which is a large part of why directories punch above their weight in the data that follows.

What feeds ChatGPT: Google Business Profile and Bing Places

The listing that feeds ChatGPT is your Google Business Profile, read through google.com. In our Ontario scrape, google.com was the source ChatGPT cited most by a wide margin, 130 times, ahead of every individual business website and every third-party directory combined. When ChatGPT named the best dentists in Toronto, it pulled The Richmond Dental Centre, Chaplin Dental, Clinton Dental and others, every one attributed to google.com. The engine is reading Google's Maps listings, the review counts, and the Knowledge Graph card, then writing them up in full sentences. Whatever Google knows about your business is most of what ChatGPT will say about it.

This produces the single most common confusion in local AI search, so it is worth stating directly: a top Google organic ranking does not feed ChatGPT. ChatGPT reads Google's Maps and Knowledge Graph data, not Google's ranked blue links. A business that ranks first organically but has a thin or inconsistent Google Business Profile can be skipped entirely, while a competitor with a complete profile and steady reviews gets named. The lever is the profile, not the ranking, and it is the highest-return single fix on this entire map.

Bing Places deserves a mention precisely because of a paradox. ChatGPT's web retrieval has historically run on Bing's index, which would suggest Bing Places should matter as much as Google. The data says otherwise. Even with Bing under the hood, the answer still resolves to Google's local data far more often, because Google's Maps corpus is richer and more complete than Bing's. Keeping a Bing Places listing accurate is cheap insurance and worth doing, but if you have one hour to spend on the ChatGPT layer, spend it on Google Business Profile. The two listings are not equal feeders despite the plumbing.

What ChatGPT mostly ignores is just as useful to know. Standalone consumer directories carry little weight here. BrightLocal's study of 800 manual ChatGPT local searches across 20 verticals and 20 US cities found directories supplied only 15 percent of sources overall, with business websites at 58 percent and business mentions at 27 percent. Within that thin directory slice, Three Best Rated and Expertise led, while Yelp, Facebook and Google Maps did not appear as directory sources at all. The signal lines up with ours: for ChatGPT, fix Google first, keep your own site clean and well-structured second, and treat most other directories as a low priority for this one engine.

What feeds Claude: editor-curated directories like Three Best Rated

The listing that feeds Claude is the editor-curated directory, and Three Best Rated sits at the centre of it. Claude cited threebestrated.ca 116 times in our scrape, more than it cited any business website, review platform or other directory, followed by custom-contracting.ca, yelp.com, furnaceprices.ca, goodcaring.ca and opencare.com. These share a defining trait: a human editor has already chosen a shortlist. Three Best Rated publishes a vetted top-three per category per city. Expertise.com runs a similar editorial selection. FurnacePrices.ca curates "what the data says" roundups. Claude trusts that editorial work and reuses it.

The practical consequence is sharp. On the Claude layer, being merely listed is not enough; being selected onto the shortlist is what gets you named. A business can have a profile on fifty directories and still miss Claude entirely if none of those directories editorially picked it. Conversely, earning a genuine Three Best Rated or Expertise placement carries real weight inside Anthropic's engine and almost none inside ChatGPT's google.com layer. That is the whole reason a per-engine map beats a flat checklist: the same effort buys visibility on one engine and nothing on another. Our companion piece on how Claude leans on curated directories for local answers breaks the pattern down query by query.

Notably, Claude is also where Yelp showed up as a meaningful feeder in our Ontario data, cited 31 times, more than ChatGPT pulled from it. This cuts against the common assumption that Yelp is primarily a Perplexity or ChatGPT signal. The engines do not agree even about a platform as large as Yelp, which is exactly why "get on Yelp, it feeds AI" is too blunt to be useful. The honest statement is narrower: Yelp feeds Claude and Perplexity in our data, and barely feeds ChatGPT.

What feeds Perplexity: the review layer and the Yelp Fusion hard-wire

The listing that feeds Perplexity is the review aggregator, and unlike the others, no single one dominates. In our scrape Perplexity cited homestars.com (17), opencare.com (9), dangeloandsons.com (9) and bbb.org (9) near the top, alongside recurring firm sites like preszlerlaw.com and mcleishorlando.com. Its lead domain sits at a modest 17 because Perplexity spreads its citations more evenly than any other engine. Breadth is the currency. A business visible across HomeStars, Opencare, the Better Business Bureau and several niche review surfaces has many chances to be retrieved; a business dominant on one and absent elsewhere has few.

Perplexity also has something the other engines do not: a contractual feed. Perplexity integrated Yelp's Fusion data directly into its answers, pulling Yelp maps, reviews, photos and review quotes through a data-licensing program rather than incidental web crawling. That makes Yelp a wired-in local source for Perplexity, not a hopeful one. The distinction matters for how you treat your Yelp profile: on Perplexity it is plumbed into the answer surface, so accuracy and review volume there feed the engine directly.

The larger studies confirm Perplexity's review-heavy diet from a much bigger sample, and they sharpen one point our Ontario numbers only hint at: Yelp is enormous in aggregate. Local Dominator's report of 267,280 AI citation mentions, monitored across ChatGPT, Gemini, Perplexity, Claude, Grok and Google AI Mode in April 2026, found combined Yelp citations of 71,512 outpaced Google's 55,977 as a data source. Yext's analysis of 17.2 million citations found that listings and directories represent the largest share of distinct citation source URLs at 54.53 percent, and that ChatGPT shows the widest swing by industry: in hospitality, third-party and review-managed platforms supplied roughly 45 percent of its citations, the heaviest such reliance among the engines Yext tracked in that sector. The review layer is not a minor input across the field; it is frequently the largest one. Perplexity is simply the engine that wears that diet most openly.

What feeds Gemini: there is no listing, only the Vertex pipe

Gemini is the exception that proves the rule: there is no listing that feeds it, only a pipe. The most-cited domain in our entire dataset was vertexaisearch.cloud.google.com, cited 384 times, more than any real publisher across all four engines. Vertex AI Search is Google's grounding service, so what the citation data shows is the redirect wrapper, not the pages underneath. You cannot "list your business on Vertex." There is no profile to claim, no directory to join, no editor to win over. The feed is a black box with Google's crawler on the other side.

That changes the lever entirely. Because the pages are masked behind the wrapper, the only thing you can reliably influence is whether Google's grounding pipeline can fetch, read and parse your content cleanly. Crawlability and structure are the whole game here. Clean Schema.org markup, fast and accessible pages, and clear entity definition are how you make that fetch reliable. You feed Gemini indirectly, by being legible to the machine that fills the pipe, not by occupying any listing. We open up the wrapper and its tracking implications in how Gemini's Vertex grounding shapes local citations.

The independent research adds a useful contrast that confirms Gemini's distinct behaviour. Yext's 17.2-million-citation analysis found that while ChatGPT in hospitality skews toward third-party sites, Gemini skews heavily toward brand-controlled domains. That fits what the Vertex pipe implies: Gemini grounds in what its crawler can structurally verify, which often means a business's own well-marked pages and Google's own data rather than the curated directories Claude prefers. The lesson for an Ontario business is that the Gemini layer rewards the least glamorous work, technically sound, schema-rich pages, and ignores the listing-collecting entirely.

Where the feeds overlap, shift by industry, and what to claim first

The feeds do overlap in a few specific places, and those overlaps are where your effort compounds across engines. The domains that surfaced in more than one engine in our data are the aggregators with genuine cross-engine reach: homestars.com appeared across all nine cities and two engines, opencare.com likewise, and furnaceprices.ca bridged Claude and Gemini. These are the rare listings that crack the 16.3% shared core. If a directory feeds two engines at once, it earns priority, because the same placement is doing double duty. Most listings do not clear that bar, which is why a blanket "list everywhere" strategy wastes effort on sources only one engine, or no engine, reads.

The feeds also shift by industry, and naming the right vertical directory is half the work. Our home-services queries surfaced HomeStars and FurnacePrices.ca; our health queries surfaced Opencare and RateMDs; our legal queries surfaced firm sites plus Best Lawyers and BBB. The public research shows the same intent-sensitivity at national scale. A Search Engine Land report on a Peec AI study of 30 million sources found Reddit, YouTube and LinkedIn most-cited overall, with Yelp and G2 specifically surfacing in recommendation-style queries, evidence that the feeding directory set changes with what the user is asking for. The implication is concrete: a dentist, a roofer and a SaaS company should not be working the same list, because their customers' questions route through different directories.

One scope note, so this page stays in its lane. This article is about which named listing each engine pulls from, the source domain that ends up in the answer. It is not about the proximity and "near me" mechanics that decide ranking once an engine is already reading a given listing, the address radius, service-area definition and category signals that govern local intent. Those sit one layer up from the directory question, and we treat them separately in the local signals behind AI near-me answers. Read this page for the listing-to-engine map; read that one for how an engine decides you are local enough to surface in the first place.

This sequencing maps directly onto two of Formative Digital's 12 Vectors. Vector 5, Cite, is the discipline of earning placement in the specific third-party sources each engine trusts, which is precisely the listing-to-engine matching this whole article describes. Vector 10, Localize, is the discipline of making your local entity unambiguous to every retrieval system at once, so the same business resolves cleanly whether it is being read through Google's Knowledge Graph, a curated directory, or a review aggregator. We run both through the Formative Forces, our orchestrated multi-agent system, so one business is worked across all four feeds in parallel rather than one listing at a time. None of it is guesswork, because every placement is verifiable against the same kind of citation scrape that produced these numbers.

One honest caveat belongs here, because this is work owners pay real money for. Outcomes depend on your industry, your competition and your existing digital presence, and listing work is not a switch that flips on identically for everyone. Our Brantford retail client Mattress Miracle grew from roughly 1,000 to more than 82,400 monthly organic visits (SEMrush, April 2026) through sustained structured-content and local work, and as their owner Brad put it, "In 40 years of advertising I've never seen anything like this. It's a completely new business." That reflects one industry and one starting point; yours will differ, which is why we diagnose before promising anything. The pattern that does hold across every business we have measured is the one this article maps: the listings feed different engines, so the strategy has to feed all four on purpose, then verify each one landed where it was supposed to.