How Perplexity Sources Its Local Recommendations

A homeowner in Brantford, Ontario opens Perplexity and types something ordinary: "who are the best roofers near me." She expects a tidy top-five, the kind Google's local pack hands over. What comes back is more revealing. The answer names D'Angelo & Sons and a couple of local firms, but the numbered citations underneath point all over the place: a HomeStars contractor page, a Better Business Bureau listing, a GAF manufacturer directory, and the roofers' own websites. Four citations, four different kinds of source, in one paragraph.

That spread is not noise. It is the most distinctive thing about how Perplexity works. Where one engine reads Google and another reads a single curated shortlist, Perplexity casts the broadest line in the water and reels in whatever it can ground. We measured it across nine Ontario cities, so this page rests on first-party evidence, not the generic retrieval explainer every competing article recycles.

Where does Perplexity actually get its local recommendations?

Perplexity gets them from a blend of third-party directories and the businesses' own websites, retrieved live and then summarised, rather than from any single ranked index. Ask it for the best dentists or HVAC contractors in a city and it runs a web search, reads the business names inside the pages it can fetch, and writes a short list with a numbered footnote behind each claim. The footnotes are the tell. They expose which sources fed the answer, and in our data they came from strikingly different layers of the web at once.

Formative Digital's scrape pushed the same local-intent prompt, "who are the best {vertical} in {city}, Ontario," through the live engines using DataForSEO against our matrix.db dataset in May 2026, covering nine Ontario cities and five service verticals. The figures below are Perplexity's grounding fingerprint: the domains it reached for most across that run.

The pattern jumps out of that list. The top is directories, exactly what you would predict, but sitting alongside HomeStars and BBB at the same citation frequency is dangeloandsons.com, a single contractor's own website. Perplexity did not treat the directory and the firm site as different tiers of trust; it cited both, across the same cities. That refusal to pick one lane is the whole story.

This is evidence national studies cannot reach. Kevin Indig's early-2026 Growth Memo analysis of ChatGPT citations found that about 44 percent of AI citations come from the first 30 percent of a page, which confirms that engines cite real, crawlable pages and reward placement near the top of them. We took that principle local and Canadian, into the long tail where a single review listing or a well-structured firm page decides whether a Cambridge roofer gets named at all.

Why is Perplexity the most diversified citation engine?

Perplexity is the most diversified engine because its architecture is retrieval-first and source-agnostic: it grounds every answer in a fresh web search and rewards breadth of presence rather than dominance of one index. The other major engines each funnel through a preferred layer. In the same FD dataset, ChatGPT leaned on google.com, Claude reached for curated directories like threebestrated.ca, and Gemini wrapped almost everything through Vertex AI Search. Perplexity alone declined to concentrate, scattering its citations across review platforms, rating bodies, vertical aggregators, and the firms themselves.

The cross-engine numbers make the contrast concrete. Across all four engines we logged 583 distinct domains, and only 16.3 percent were cited by two or more of them; the other 83.7 percent were unique to a single engine. Perplexity fed that long tail heavily. The full cross-engine picture is in our study of the missing consensus between AI search engines on local businesses.

This is also why Perplexity scores comparatively well on faithfulness. The Columbia Journalism Review's Tow Center tested eight AI search engines across 1,600 queries; Perplexity had the lowest failure rate of the group, though it still answered incorrectly in 37 percent of cases. The lesson for an owner is double-edged: the wide, retrieval-bound approach makes Perplexity careful, but no engine is infallible, so the sources it grounds against are worth controlling rather than trusting blindly.

Which directories does Perplexity lean on, and why does HomeStars dominate?

Perplexity leans hardest on broad, review-rich directories, and HomeStars dominates because it pairs wide geographic coverage with the structured, quote-friendly listing format the engine prefers to ground against. HomeStars was the single most-cited domain in our entire Perplexity run: 17 citations, present in all nine cities, spanning roofing and HVAC. Opencare played the same role for dentistry across all nine cities; BBB showed up across eight. These are the platforms that already aggregate the reviews, ratings, and business details Perplexity finds easiest to extract and attribute.

The reason these directories win is structural, and it lines up with the peer-reviewed research. In "GEO: Generative Engine Optimization" (arXiv:2311.09735, ACM SIGKDD 2024), Pranjal Aggarwal and colleagues showed that adding citations, statistics, and quotable structure to a source can raise its visibility inside generative answers by up to 40 percent. A HomeStars page is built exactly that way, a ranked list of contractors each with a rating, a review count, and a one-line description, far easier for an engine to fetch and cite than a firm's homepage written as marketing prose. The directory wins because its format is the citation-ready format.

How does Perplexity blend directory pages with individual firm websites?

Perplexity blends them by treating a well-structured firm site as a peer of the directory, not a poor substitute for it, and frequently cites both inside a single answer. The clearest proof in our data is dangeloandsons.com: D'Angelo & Sons, a roofing and home-improvement company, had its own website cited 9 times across all 9 cities, matching BBB and Opencare citation for citation. Perplexity was not falling back on the firm site because a directory was missing; it cited the firm site because the site earned it, with clear, city-specific, extractable pages the engine could ground.

The legal vertical shows the blend even more vividly. Alongside the directories, Perplexity repeatedly cited individual firm domains: preszlerlaw.com in seven cities, mcleishorlando.com in six, diamondlaw.ca in five. These firms recur city after city not because they bought placement but because their content travels, each maintaining distinct, well-built location pages that read cleanly to a retrieval engine. A single answer for "best personal injury lawyers in Mississauga" might cite a Best Law Firms directory page plus neinstein.com and diamondlaw.ca in one list.

This dual pull reframes the optimisation question. Earning presence in two places at once, the directory ecosystem and your own crawlable site, is more demanding than chasing one index, but a business strong in both is hard for any single algorithm change to dislodge. This page is the measurement study behind that claim, the source-mix data rather than the step-by-step playbook; if you want the action checklist that turns these findings into tasks, our guide to optimising for Perplexity citations covers the how-to, and which directory and proximity cues push a business into a near-me reply goes deeper on the directory side.

Why do Perplexity's local picks diverge so far from Google's?

They diverge because Perplexity is broader and more independent of Google than either ChatGPT or Google AI Overviews, so its citations split off most from Google's own results on local queries. ChatGPT built its local lists largely from google.com, and Google AI Overviews draw on Google's own index; Perplexity sits apart from both, as the table sets out.

Independent benchmarks confirm where the divergence concentrates. BrightEdge's analysis, reported by Search Engine Land, found Perplexity citations overlap Google's top 10 organic results about 60 percent of the time overall, but the overlap swings hard by vertical, from roughly 82 percent for healthcare down to 27 percent for restaurants. Local service and hospitality queries are exactly where Perplexity stops echoing Google and starts pulling the directories and niche sources our scrape captured.

The citation ecosystems differ structurally, which is why each engine has to be optimised on its own terms. Profound's analysis of roughly 680 million citations found Reddit accounting for 46.7 percent of Perplexity's top-10 sources, a profile distinct from Wikipedia's dominance inside ChatGPT and the YouTube-and-Reddit mix inside Google AI Overviews. Three engines, three different webs. A plan that wins one does not automatically win the others.

How do you earn a spot in Perplexity's widest-in-class citation net?

You earn it by being present and extractable in both places Perplexity looks at once: listed on the directories it trusts, and structured on your own site so the engine can fetch, read, and attribute a claim about you in seconds. The firms recurring across our scrape, D'Angelo & Sons, Preszler, McLeish Orlando, Neinstein, are the ones already doing both halves. The GEO research agrees that targeted, structured content lifts generative visibility measurably, so this is engineerable, not luck. None of it is magic ranking dust, and no one can promise a given engine names a given business on a given day.

One honest caveat belongs here, because this is advice owners spend real money on. Outcomes depend on your industry, your competition, and your existing digital presence; AI visibility is not a switch that flips the same way for everyone. 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, 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 run a per-engine Perplexity citation diagnostic before promising anything.

Be honest about the ceiling too. Even the most diversified engine is fallible. Nieman Lab, reporting on the same Tow Center study, noted that content-licensing deals between AI firms and publishers did not guarantee accurate citation. The reading is plain: you cannot buy your way into a correct Perplexity citation, and you cannot edit the engine. What you control is the source material it grounds against, which is why earning broad, accurate, structured presence is the only reliable lever.

A few questions readers keep asking

Does Perplexity use Google or Bing for its local data? Its retrieval draws on the Bing search index and Bing Places data, supplemented by its own PerplexityBot crawler that fetches pages directly. That partial independence from Google is a big reason its source mix diverges from Google's results on local service queries.

Can a local business get cited directly, or only through a directory? Directly, and our data proves it: D'Angelo & Sons' own domain was cited 9 times across 9 cities, matching the major directories. A firm site that is crawlable, city-specific, and structured for extraction can be cited as a primary source, not just listed inside someone else's page.

The throughline is the one the scrape keeps returning. Perplexity does not reward picking a lane. It rewards showing up, accurately and extractably, in as many of the places it searches as you can reach. For how directories and review platforms feed every engine, not just this one, our companion analysis of how AI search surfaces Ontario injury law firms traces the same mechanics inside a regulated vertical.