How AI Engines Pick Roofers in Ontario

Has any AI engine ever actually named a specific roofer, or is the whole field just guessing? It has, repeatedly, and we logged exactly which ones. Across nine Ontario cities in May 2026 we asked ChatGPT, Anthropic Claude, Google Gemini and Perplexity the same question, "who are the top 5 best roofers in this city, Ontario," and recorded every business and every source each engine cited. The result is 1,732 citations of first-party evidence about how a roofer answer is really built, and it does not match the advice most agencies are selling.

That gap is the reason this piece exists. Almost every competing article on AI visibility for roofers gives the same generic counsel, optimise your content, add schema, build authority, without a single piece of evidence that any engine has ever named a real contractor. We are leading with the evidence instead. The data says something specific and a little uncomfortable for the trades: for a high-consideration, seasonal purchase like a roof replacement, AI routes the answer through review aggregators first, and your path to a citation runs through them before it runs through your homepage.

AI names review aggregators before it names a single roofer

Ask the four engines for the best roofer and they name a mix of two or three review aggregators and a short rotation of contractor domains that built extractable city pages. In our scrape, the single most cited contractor was Custom Contracting at 30 citations across five cities, followed by True North Forming at 19 and D'Angelo & Sons at 14 across all nine cities. But the aggregators sat above or alongside them: ThreeBestRated at 37 and HomeStars at 26 were cited more often than any roofing company except Custom Contracting. The answer is not a leaderboard of the best roofers. It is a list of whoever was easiest to retrieve and quote.

Front-loading that finding is deliberate, and it follows the evidence on how engines read. Kevin Indig's February 2026 Growth Memo citation study, which isolated 18,012 verified citations from roughly 3 million ChatGPT answers, found that 44.2 percent of AI citations come from the first 30 percent of a page. Engines reward sources that lead with a clean, ranked answer. A ThreeBestRated page titled "3 Best Roofing Contractors in Hamilton" opens with exactly that. A roofing homepage that opens with a hero image and a tagline gives the engine nothing to lift near the top, so the engine reaches for the directory instead.

Two patterns matter most. First, review aggregators carry the answer for trades, the same way ranking directories carry it for regulated professions. Second, the engines barely agree with each other. Across the full FD dataset of 583 distinct domains over 44 city and vertical cells, only 95 sources, or 16.3 percent, were cited by two or more of the four engines. For roofing specifically, ChatGPT's Maps-driven lists and Claude's directory lists name different firms entirely for the same city. There is no single answer to win. There are four, which is exactly why 83.7% of AI citations belong to just one engine in our cross-vertical dataset.

Why do review aggregators like HomeStars and ThreeBestRated dominate roofer answers?

They dominate because a roof is a rare, expensive, trust-heavy decision, and aggregators package exactly the proof an engine needs to ground a recommendation safely. A homeowner replaces a roof maybe twice in a lifetime and spends thousands doing it. The engine, like the homeowner, wants third-party validation rather than a contractor's own marketing claim. A HomeStars city page gives it a ranked list, verified review counts, star ratings and addresses in one retrievable place. That is far easier to quote and attribute than a roofing company's homepage written as a sales pitch.

This is also where the academic evidence lines up with the field data. The peer-reviewed GEO paper by Pranjal Aggarwal and colleagues, "GEO: Generative Engine Optimization" (arXiv:2311.09735, presented at KDD 2024), found that adding cited sources, quotations and statistics to a page can lift its visibility inside AI-generated answers by up to 40 percent, and that the effective tactics vary by domain. Aggregator pages are built from precisely that material: counts, ratings, quotes, structured detail. Your homepage usually is not. So the aggregator wins the citation even when your work is better, because it is the more groundable source.

The lesson is not that your website does not matter. It is that the aggregator layer is the first gate, and most roofers have not engineered for it. If your firm is missing from ThreeBestRated, HomeStars, Yelp and the Better Business Bureau, you have ceded the sources the engines pull from most. Custom Contracting, True North Forming and D'Angelo & Sons recur across cities precisely because they show up in both layers at once: present in the aggregators, and structured on their own domains. The directories-first reality of local AI search is a pattern we trace across verticals in our look at how directories shape AI answers for local queries.

Does ChatGPT use your Google reviews, or does it cross-check Yelp, BBB and Angi too?

ChatGPT leans heavily on your Google reviews, but the other three engines cross-check a wider review web, so Google reviews alone are not enough. In our data, ChatGPT assembled most of its Ontario roofer lists from google.com, pulling Maps cards complete with star ratings, review counts and addresses. For that one engine, a complete Google Business Profile with strong, recent reviews is close to decisive. Watch the sample ChatGPT answer for Burlington and you can see it: each roofer arrives with a Maps-style line like "Open now, Siding contractor, 4.8 (26 reviews)."

The moment you step outside ChatGPT, the review surface widens. Claude reached for ThreeBestRated, Yelp and HomeStars. Perplexity spread across HomeStars and the Better Business Bureau, with the BBB cited in eight of nine cities. Gemini wrapped its sources in Vertex but surfaced Birdeye, Houzz and RenovationFind underneath. None of those four engines is satisfied by a single review platform. A roofer who has 200 Google reviews and nothing on HomeStars, Yelp or the BBB is strong in one engine and invisible in the source layers the others trust.

So the honest instruction for a roofer is breadth, not depth on one platform. Claim and complete your profiles on HomeStars, Yelp, the Better Business Bureau, Houzz and ThreeBestRated, keep your name, address and phone identical across all of them, and gather reviews on more than just Google. Entity consistency across those surfaces is what lets each engine recognise you as the same business, a point we cover in our guide to keeping your business identity consistent for AI. The roofers winning multiple engines are not the ones with the single biggest review count. They are the ones present everywhere the engines look.

An AI citation and a #1 Google ranking are built from different inputs

Ranking first on Google and getting named by an AI engine are different jobs, because Google rewards your page while AI rewards whatever source it can retrieve and ground in the moment. A number-one Google position is built on backlinks, on-page relevance and a long tail of classic signals pointing at your domain. An AI citation is built on retrievability: can the engine fetch a page right now, find your firm named in a clean structure, and attribute a claim safely. Those are not the same asset, which is why a roofer can top Google and still be missing from three of four AI answers.

The click data is why this distinction now costs money. Pew Research Center found that in March 2025, Google users who encountered an AI-generated summary clicked a traditional search result only 8 percent of the time, against 15 percent when no AI summary appeared, and clicked a link inside the AI summary itself just 1 percent of the time. Being ranked below the answer is worth a fraction of what it was. Being named inside the answer is the new prize. Search Engine Land calls this shift generative engine optimization, the move from competing for a page-one position to winning a mention inside the generated reply.

This matters because owners keep assuming Google authority transfers automatically to AI, and for trades it largely does not. ChatGPT's reliance on google.com means your Google work pays off there. Claude, Perplexity and Gemini read aggregators and grounding wrappers that your Google rank never touches. We pull this divergence apart in detail in our piece on how AI recommendations differ from Google Maps results. The roofers who understand it stop measuring success by a ranking report and start measuring it by who the engines name.

What makes a roofing page extractable enough for AI Overviews and Perplexity to cite?

An extractable roofing page answers the buyer's question in plain, structured text near the top, backed by schema the engine can parse without guessing. The engines are not reading your page for beauty. They are scanning for a clean claim they can lift and attribute: what you do, where you do it, what it costs, what proves you are credible. The contractor domains that recur in our scrape, Custom Contracting, True North Forming, D'Angelo & Sons, all lead their city pages with exactly that kind of extractable detail rather than a slogan.

Price transparency is one of the most underused levers here. The sample Claude answer for Burlington pulled a concrete figure straight off a contractor page: "25-year warranty, roof replacement from $8,000." A stated cost range is a quotable, groundable fact, and engines prefer facts they can attribute. Credibility signals work the same way. The same answer surfaced manufacturer certifications by name: GAF Master Elite, CertainTeed ShingleMaster, IKO Roofing Preferred and Owens Corning Platinum Preferred. Putting those certifications, your warranty terms and a real price range in extractable text near the top of a city page gives the engine multiple safe claims to cite.

Schema is the structural half of this, and the brief from Google is narrower than most agencies pretend. Google says no special schema is required for AI features, yet LocalBusiness and FAQPage markup still earn their place by disambiguating your firm and exposing question-and-answer pairs the engine can read cleanly. The point is not to game a tag. It is to make the page legible, so the grounding step finds your name, your city and your claims without ambiguity. We go deeper on this in our explainer on building a schema graph that AI can actually read. This work maps onto Vector 6 of Formative Digital's 12 Vectors, Structure, which is about making a page machine-legible before it is anything else.

How does seasonality change the way homeowners search for roofers with AI?

Seasonality compresses roofing demand into spikes, which raises the stakes of being citable before the spike rather than during it. Ontario roofs get bought in bursts: spring melt that exposes winter damage, summer replacement weather, and the autumn scramble before snow. When a homeowner discovers a leak in April, they do not run a months-long search. They ask an engine for the best roofer in their city and act on the short list it returns. If you are not already present in the aggregators and Maps data that day, you are not in the answer, and there is no time to climb in.

This is what makes a roof a high-consideration but time-boxed purchase, and it changes the optimization calendar. Unlike an everyday service, the roofer query surges seasonally and unpredictably after storms, so the citable assets, your aggregator listings, your Google profile, your structured city pages, have to be built in the quiet months and maintained continuously. There is no benefit to a content sprint that lands after the season peaks. The compounding play is to be the source the engines already trust when demand arrives, which is the logic behind a steady, structured program rather than a seasonal blitz, and it pairs with the measurement discipline we describe next.

The broader Canadian context is that AI-mediated discovery is normalising fast. Statistics Canada reported that in the second quarter of 2025, 12.2 percent of Canadian businesses used AI to produce goods or deliver services, double the 6.1 percent a year earlier, with virtual agents and chatbots among the top applications at 24.8 percent. The homeowners asking an engine "who should re-roof my house" are no longer a fringe. For a seasonal trade, missing that channel during a demand spike is missing a measurable share of the year's work. If you want to know where your firm stands before the next season, you can have us check which engines name you at no cost.

A roofer controls four levers that AI engines actually read

Four levers map directly onto what the engines pull, and every one of them is buildable without gaming anything. The engines decide who to name from sources you can influence: third-party reviews, structured data, stated pricing and named credentials. None of these games the engine. Each one simply makes you a safer, more quotable source for a grounding step that is looking for exactly that material.

What a roofer cannot control is just as important to accept. You cannot edit the engine, you cannot rewrite a ThreeBestRated shortlist you are not on, and you cannot stop an aggregator from ranking a competitor above you. The work is entirely on the inputs: be present in the sources, be the more groundable listing, be the firm with a stated price and named certifications when the engine goes looking. Tools like Whitespark and Birdeye help manage the listing and review side at scale, but the strategy is the same whether you use them or do it by hand. Concentrate effort where you have real influence, and stop spending it where you have none.

How do you measure whether ChatGPT, Gemini and Perplexity are recommending your roofing company?

You measure it by running the real homeowner queries through each engine on a schedule and tracking who gets named, because AI visibility is four scores, not one. The single most expensive mistake we see is treating AI search as one channel with one ranking. It is four channels that happen to share a chat box, and because 83.7 percent of cited sources appeared in only one engine, checking ChatGPT tells you almost nothing about Gemini or Perplexity. Honest measurement is cross-engine and repeated, never a single snapshot.

In practice that means asking ChatGPT, Claude, Gemini and Perplexity "who are the best roofers in my city, Ontario" on a regular cadence, recording which firms each one names and in what order, and tracking your share of those mentions over time. Sample each engine more than once, because each wobbles run to run. Watch the source layer as your leading indicator: if you have newly appeared on HomeStars or ThreeBestRated, expect movement in Claude and Perplexity before you see it anywhere else. The citation is the lagging signal; the listing is the early one. We lay out the full method in our guide to measuring your AI search citations across engines.

Tracking the source layer also tells you where to act next. If you are named by ChatGPT but not the others, your Google presence is strong and your aggregator presence is thin, so the fix is breadth on HomeStars, Yelp and the BBB. If Gemini surfaces you through Vertex but Claude does not, your structured content is readable but you are missing from the curated shortlists Claude trusts. The diagnosis points straight at the gap, which is the whole reason we measure per engine rather than blending it into one comforting number.

What should an Ontario roofer fix first to start getting cited within 30 to 60 days?

Fix your aggregator presence and your Google Business Profile first, because those are the highest-authority sources the engines re-crawl most often, so they move fastest. Before touching your website, claim and complete your listings on HomeStars, Yelp, the Better Business Bureau, Houzz and ThreeBestRated, make your name, address and phone identical everywhere, and correct your Google Business Profile down to the category and service area. In our data those are the exact sources ChatGPT, Claude and Perplexity pulled from most, so a correction there has the shortest path to a changed answer.

Then make your own city pages extractable. Lead each one with a plain service-and-city line, a real price range, your named certifications and your warranty terms, and add LocalBusiness and FAQPage schema so the grounding step reads you cleanly. This is slower than the listing fixes because your domain is re-crawled less often than a major aggregator, but it is what lets Gemini's Vertex pipeline and AI Overviews cite you directly rather than only through a directory. The sequence matters: aggregators and Google first for speed, structured pages second for durability.

One honest caveat belongs here, because this is your money on the line. No engine guarantees a citation, the engines and aggregators change constantly, and outcomes depend on your competition, your city and your existing presence. Our Brantford retail client Mattress Miracle grew from roughly 1,000 to more than 82,400 monthly organic visits (SEMrush, April 2026) by building content the engines could read and cite, but that reflects one industry and one starting point, and roofing is its own seasonal, competitive market. If an existing domain shows no measurable organic search results after twelve months of work with us, our Results Guarantee means we keep working at no further cost until results land. What the data supports is direction, not a promise: be present in the sources, be the more groundable listing, and the citations follow.