How Claude Recommends Local Businesses

The competitor articles that rank for this question describe Claude in the abstract. Claude uses Brave Search. Claude cites a few sources. Trust signals matter. All true, all unprovable from the outside, and all missing the one thing an Ontario business owner actually needs: which domains Claude reaches for in this province, and which businesses it surfaced when it did. We pulled that record. The table below is the first thing worth seeing.

Read the bottom row twice. The Google Maps three-pack, the source every other AI-local guide treats as the foundation of a recommendation, was cited zero times on its own. Google's business data reached Claude only when a Business Profile's details had been mirrored onto a page Claude could name. The directories did the work. This piece explains why, and how an Ontario company earns a Claude citation instead of waiting for one. The strongest evidence sits at the top for the same reason it matters to Claude: Kevin Indig's early-2026 Growth Memo citation study found that 44.2% of AI citations are drawn from the first 30% of a page.

Does Claude use Maps, or web search, to find a local business?

Claude finds local businesses through live web search, not a maps product. When web search is switched on, Claude sends the query to Brave Search, retrieves a short list of pages from that index, and composes its answer from what those pages contain. Anthropic is explicit that the result is grounded in nameable sources: its help documentation states that when Claude searches the web, "every response includes citations, so you can easily verify sources yourself," with source links and relevant quotes. That single choice is why a Maps panel never surfaces as the answer. Claude wants a page it can point at, and a map pin is not a page.

Most explanations get this backwards. They treat Claude as if it ranks businesses the way Google ranks listings. It holds no "best dentists in Hamilton" list anywhere in its weights. It runs a search, reads three to six results, and tells you what those results already say. Change the pages Brave returns, and you change the recommendation. And Brave's index is its own corpus, not a copy of Google's, which is why Claude's local answers can look nothing like the three-pack a searcher sees in the same city.

ThreeBestRated dominates Claude's Ontario citations

ThreeBestRated dominates because it gives Claude exactly the shape of source it is looking for: a short, pre-vetted, plainly worded shortlist. The site's entire format is "the three best" providers in a city for a category, each with hours, contact details, and a stated vetting process. When Claude searches "best roofing contractors in Guelph, Ontario," a page titled "3 Best Roofing Contractors in Guelph, ON" is close to an ideal retrieval target. The work of selecting and ranking is already done, in language Claude can lift and attribute. So it lifts it.

In our data, threebestrated.ca was cited 116 times, spanning eight of the nine cities and four of the five verticals we tested. The next individual business site, custom-contracting.ca, appeared 35 times. The gap is not subtle: ThreeBestRated was cited far more often than Yelp, HomeStars, and Opencare combined, 116 against their 74. For an Ontario service business, that listing is not a nice-to-have. It is one of the highest-probability paths into a Claude answer that exists.

This holds up theoretically as well as empirically. Aggarwal and colleagues, in the GEO paper presented at KDD 2024, showed that generative engines carry a systematic bias toward third-party authoritative sources, and that adding citations, quotations, and statistics to a source raises its visibility by 30 to 40% on their position-adjusted word-count metric. A curated directory is that source. It cites itself, quotes its picks, and structures the facts. Claude behaves exactly as the research predicts a generative engine should.

Does Claude ever cite Google Maps directly, or only mirrored pages?

Claude never cited Google Maps as a standalone source in any of the 1,732 citations we recorded. Google's local data did reach Claude, but only when a Business Profile's name, address, phone, and hours had been mirrored onto a page Claude could cite by URL, a directory profile, a review listing, or the business's own site. The map pin is invisible to this process. The page that repeats the pin's details is not.

That nuance matters for owners who have poured effort into their Google Business Profile and assume AI recommendations come from there. For Claude, the Profile is a consistency anchor, not a citation: its value is pushing accurate, identical details out across the web, where Claude's actual sources pick them up. A Profile that disagrees with a business's directory listings hurts, because it hands Claude conflicting facts to reconcile.

How many sources does Claude pull before it names a business?

Claude typically grounds a local answer in two to six distinct domains. The common pattern in our Ontario sample was a curated directory, sometimes two, plus one or two individual business sites, occasionally rounded out by a review platform. Surface Local, in its analysis of Claude's web-search behaviour, describes the same shape: a focused set of three to six sources per answer, usually the official site plus two to four directories or curated guides, rather than a long crawl of the open web.

The small number is the whole point. Because Claude reads only a handful of pages before answering, the contest for a citation is not "rank in the top fifty." It is "appear in the three to six pages Brave returns for this query." That narrower target is why a business can rank well on Google and still never be named by Claude. The two systems fill different slots, and Claude's are scarce.

Notice the texture of those answers. A directory anchors each one, and named businesses sit alongside it. The dentist row leans on vertical directories built for health, Opencare and RateMDs, because that is what Claude found for that category. The mix shifts by vertical, but the directory backbone holds across all of them.

The directories and review sites Claude trusts most in Ontario

Claude's most-trusted local sources in Ontario form a clear hierarchy, and it is not the one most owners would guess. Curated directories sit at the top, generic review sites in the middle, and vertical-specific directories cluster by category underneath.

Two findings stand out. First, ThreeBestRated outweighs Yelp and HomeStars combined, which inverts the usual assumption that the biggest review brands win AI citations. They do not, at least not with Claude in Ontario. Second, curated human-edited guides such as FurnacePrices.ca and TrueNorthForming punch above their size precisely because they read like recommendations, not listings. Claude is drawn to sources that have already done the judging.

Why does Claude favour directories over a business's own website?

Claude favours third-party directories because an independent source is more useful to it than a brand describing itself. When Claude justifies naming a business, "a curated directory ranked them in the top three" is a claim it can attribute. "Their own homepage says they are the best" is not. The engine grounds assertions in sources a reader could verify, and a self-published superlative fails that test. So the marketing copy is discounted, and the directory entry carries the weight.

This matches the wider pattern in generative search. Leigh McKenzie, writing in Search Engine Land in February 2026, frames generative engine optimization around earning independent third-party mentions, reviews, journalism, community discussion, and notes that Reddit, LinkedIn, and YouTube were among the top sources cited by leading large language models in October 2025. Earned media outranks owned media for AI citation. A business website still matters, because Claude cites official sites for facts like address and services. It just does not carry the recommendation alone. The point here is narrow, why Claude in particular reaches past your homepage; the broader case that pre-ranked listings beat owned pages on every engine sits in our pillar on pre-ranked listings as AI local infrastructure.

Brad's reaction came from watching organic visibility move at a pace traditional advertising never produced. The lever behind it is the one this article describes: getting independent sources to describe a business accurately, so the engines and the directories they read have something true to repeat. Results of that kind depend on industry, competition, and a business's existing digital presence, so no two cases move identically.

Why Claude's curated-directory slice diverges from the other engines

Claude's local sourcing is distinct because each major engine grounds its answers in a different slice of the web, and Claude's slice is curated directories. Across our full cross-engine dataset, 583 distinct domains appeared over 44 city-and-vertical cells, yet only 16.3% of them were cited by two or more engines. In other words, 83.7% of cited sources were unique to a single engine. The engines are not reading the same web, so they do not name the same businesses.

The takeaway is not that one engine is right. It is that a single tactic does not cover all of them. A business optimised purely for Google's Knowledge Graph might win ChatGPT and stay absent from Claude, because the directory layer Claude trusts never picked it up. We unpack the full divergence, and why national brands behave differently from local ones, in our companion analysis of the consensus gap between AI engines on local picks. For the engines that contrast most sharply on architecture, our breakdown of how ChatGPT and Perplexity differ as answer engines goes deeper on retrieval versus generation.

What does a business need on the web to get named by Claude?

A business needs three things on the web before Claude will reliably name it: a place in a curated directory, consistent identity across platforms, and plain factual descriptions on independent pages. Each one removes a reason for Claude to skip you. Without the directory, you are not in the result set Brave returns. Without consistency, Claude finds conflicting facts and hedges. Without clear third-party descriptions, the engine has nothing quotable to attribute.

The entity-consistency point deserves weight, because it is the one owners most often neglect. If your business is "Smith & Co Roofing" on one directory, "Smith and Company Roofing Inc." on another, and a third spells the street differently, Claude has to decide whether those are one business or three. Faced with that ambiguity, it tends to name the competitor whose identity is clean. Tightening that consistency is foundational work, and we cover the mechanics in our guide to keeping your name, address, and phone aligned for near-me answers.

Claude resolves conflicting sources by weighting consensus

Claude resolves conflicts by looking for agreement, and agreement is easiest to find when a business presents the same facts everywhere. When two sources say a roofer is the top-rated option in Oakville and a third says nothing, Claude weights the consensus and names the roofer. When sources disagree on basic details, the hours, the spelling, the service area, the engine grows cautious, qualifies its answer, or quietly leaves the business out in favour of one with a cleaner record. Consensus is the currency, and contradiction is the tax.

This is why a curated directory is doubly valuable. It does not just put a business in front of Claude; it supplies a confident, single-source ranking that other pages can corroborate. A ThreeBestRated entry that matches a business's own site and its review profiles forms exactly the kind of agreement Claude rewards. The businesses that get named consistently are usually the ones whose web presence tells one coherent story across every page Claude might read.

What should an Ontario business do to earn a Claude citation?

An Ontario business should start by earning the curated directory listing Claude already trusts, then make the rest of its web presence agree with it. The order matters. The directory gets you into the candidate set; consistency and clear copy turn that into a confident citation. None of this is a ranking shortcut. It is the patient work of giving an answer engine accurate, independent sources to repeat.

1. Get listed where Claude looks. A ThreeBestRated entry for your city and category is the single highest-probability path in our data. Add the vertical directories that fit you, Opencare or RateMDs for dental, HomeStars for trades.
2. Make your identity identical everywhere. One business name, one address format, one phone number, repeated byte-for-byte across your site, your Google Business Profile, and every directory.
3. Write city-specific, factual pages. Plain descriptions of what you do, where, and for whom give Claude quotable claims. Skip the superlatives it discounts anyway.
4. Earn independent mentions. Local press, community pages, and review profiles are the earned media that outranks your own marketing in Claude's eyes.
5. Measure on Claude specifically. Run your real queries through Claude and record which sources and businesses it names, because the answer differs from Google and from the other engines.

That final step is where most owners are blind. You cannot improve a Claude citation you have never measured, and Claude's answer will not match a Google search. Our walkthrough of recording what each engine names in its own answer shows how to capture those answers engine by engine. If results do not land within twelve months of work on an existing domain, Formative Digital keeps working at no further cost until they do; that is our Results Guarantee, a continuation of work rather than a refund.