>
>

Who ChatGPT recommends in Toronto: 375 answers, 25 categories, one $100-a-year directory

By
Published
November 8, 2026
Data collected
October 17, 2026

What this study found

We asked ChatGPT "Who is the best [category] in Toronto?" 375 times across 25 service categories and three model versions. One directory, Best in Hood, was cited in 37% of answers; Google's own pages were cited in none. Eleven categories had one business named in at least 12 of 15 answers. The same question asked five times gave the same first pick all five times in 13% of cases. The study did not test whether any business paid for its directory placement.

375
ChatGPT answers analysed across 3 model versions
25
Toronto service categories
37%
of answers cited one directory that sells its top spot
13%
of 5-run groups gave the same first pick every time (10 of 75)

What we expected

The working assumption in local SEO is that review volume on Google decides who gets recommended. If that carried over to AI assistants, ChatGPT's answers to "who is the best [category] in Toronto" should cite Google Business Profiles, Yelp and local news, and the businesses named should be the ones with the most Google reviews. This study recorded which sources ChatGPT actually cites and which businesses it actually names, so the assumption could be checked against counts rather than examples.

The category list was locked at 07:50 ET on 17 October 2026, before any answers were collected.

How we measured it

  • Prompt: "Who is the best {category} in Toronto?" with no other wording, no system prompt, no location setting.
  • Categories (25): dentist, family doctor, physiotherapist, chiropractor, optometrist, accountant, family lawyer, real estate agent, mortgage broker, financial advisor, plumber, electrician, HVAC company, roofer, general contractor, hair salon, barber, personal trainer, yoga studio, daycare, veterinarian, dog groomer, auto mechanic, SEO agency, wedding photographer.
  • Tool: DataforSEO AI Optimization, LLM Responses endpoint, llm_type chat_gpt, web search on, default temperature. Models requested: gpt-6.1-sol, gpt-5.6-sol, gpt-5.5 (resolved by the API to gpt-5.5-2026-04-23).
  • Runs: 5 per category per model, 375 in total, all collected 17 October 2026.
  • Extraction: every answer's text was read and each named business, named practitioner and non-business entity (regulator, directory, government page) was coded by hand into a structured record, then checked programmatically so every recorded name appears verbatim in the answer text. Name variants were merged to one canonical name per business within a category. Directories, review sites, regulators and government pages were not counted as businesses.
  • Citations: taken from the URL annotations the API returns with each answer, grouped by domain after removing "www." and tracking parameters. A site counts once per answer however many times that answer cites it.
  • Stability measure: for each category-and-model group of five runs, the share of runs whose first-named business matched the group's most common first-named business.

Where the data came from

Every figure in this study, with its source

Figure Source Pulled Sample Raw files
Businesses named, first picks, per-category counts Hand-coded extraction of each answer's text 2026-10-17 375 answers extracted/*.json
Cited sources (bestinhood.com 137, etc.) URL annotations returned by the DataforSEO ChatGPT endpoint 2026-10-17 375 answers, 251 distinct domains answers.jsonl
Best in Hood pricing and ranking description bestinhood.com homepage and /pricing, opened directly 2026-10-17 2 pages Sources 1–2
Best in Hood Toronto list order and labelling Six Toronto category pages on bestinhood.com, opened directly 2026-10-17 6 pages Sources 3–8
Model behaviour (clarifying questions, words, citations per answer) Computed from the flattened answer set 2026-10-17 375 answers answers.jsonl

All 375 raw API responses, the flattened answer set, the coded extractions and the scripts are held in the project archive and can be supplied to a reviewer on request.

What happened

One directory was cited in 37% of answers

Best in Hood (bestinhood.com) was cited in 137 of the 375 answers. Its homepage says each list is built from public Google Maps ratings and review counts. Its pricing page sells a $50 one-time Basic listing and a $100-a-year Premium listing described as "Top spot on the page, guaranteed for one year." On the six Toronto category pages we checked, no listing was labelled as paid, and page order did not follow star rating alone. This study did not test whether any named business paid for a placement. DaSilva Consulting has no affiliation with Best in Hood.

Every answer carried at least one citation link, and 251 different sites were cited in total. Google's own pages were cited in zero answers, Yelp in five, BlogTO in one and the Toronto Star in none. HomeStars was second at 65 answers and Reddit threads third at 40.

Bar chart of the sites ChatGPT cited in 375 answers about Toronto businesses: bestinhood.com 137, homestars.com 65, reddit.com 40, bests.ca 27, threebestrated.ca 24, reviews.birdeye.com 23, verytoronto.com 23, clutch.co 20, trustanalytica.org 20, toronto.ca 17, yelp.com and yelp.ca 5, blogto.com 1, thestar.com 0, google.com 0.
Number of the 375 answers citing each site at least once. DataforSEO ChatGPT API, collected 17 October 2026.

The same data as the chart above, as a table

Site Answers citing it (of 375) Share
bestinhood.com 137 37%
homestars.com 65 17%
reddit.com 40 11%
bests.ca 27 7%
threebestrated.ca 24 6%
reviews.birdeye.com 23 6%
verytoronto.com 23 6%
clutch.co 20 5%
trustanalytica.org 20 5%
toronto.ca 17 5%
yelp.com + yelp.ca 5 1%
blogto.com 1 0%
thestar.com 0 0%
google.com 0 0%

Some businesses came up almost every time; some categories had no consensus

Downward Dog Yoga Centre was the first-named yoga studio in all 15 answers. Spadina Auto Service was named in all 15 auto mechanic answers and led 13 of them. Search Engine People was named in all 15 SEO agency answers. Outline Financial appeared in 14 of 15 mortgage broker answers and The Cabinet Salon in 14 of 15 hair salon answers. In 11 of 25 categories, one business was named in at least 12 of 15 answers. In three categories (daycare, family doctor, financial advisor) no business reached half the answers; across 15 family doctor answers, no clinic or physician was named more than twice.

In 11 of 25 categories, one business was named in at least 12 of 15 answers. Downward Dog Yoga Centre: named first in all 15 yoga studio answers. Spadina Auto Service: named in all 15 auto mechanic answers, first in 13. Search Engine People: named in all 15 SEO agency answers. Outline Financial: named in 14 of 15 mortgage broker answers. The Cabinet Salon: named in 14 of 15 hair salon answers. In daycare, family doctor and financial advisor, no business reached half the answers.
Businesses named in almost every answer. Source: this study.

Independent businesses dominated: of 1,648 business mentions, 92 (6%) were national or multi-location chains.

Who ChatGPT recommends, by category. "Named in" counts answers that listed the business as an option; "first pick" counts answers that led with it. Each category has 15 answers.

Category Most-named business Named in Most common first pick First pick in Distinct businesses named
Accountant Gondaliya CPA 12 Stern Cohen LLP 6 33
Auto mechanic Spadina Auto Service 15 Spadina Auto Service 13 19
Barber High Altitude Barbershop 10 High Altitude Barbershop 9 44
Chiropractor Ossington Chiropractic and Rehabilitation 8 Ossington Chiropractic and Rehabilitation 6 35
Daycare Kids Way Daycare 6 Kids Way Daycare 5 19
Dentist The Richmond Dental Centre 9 The Richmond Dental Centre 5 25
Dog groomer Juju Grooms 11 Juju Grooms 7 23
Electrician SureFire Power Group Limited 12 SureFire Power Group Limited 4 17
Family doctor Dr. Saeed Riaz (tied with 8 others) 2 Women's College Hospital Family Practice Health Centre 2 18
Family lawyer Martha McCarthy (tied with 1 other) 11 Herschel Fogelman 6 30
Financial advisor Objective Financial Partners 4 Jason Heath 3 11
General contractor Northern Level Architecture & Construction 8 Astaneh Construction 3 27
Hair salon The Cabinet Salon 14 The Cabinet Salon 9 22
HVAC company Tempasure Heating & Air Conditioning 8 Cozy World Inc. 5 22
Mortgage broker Outline Financial 14 Ratehub.ca (CanWise Financial) 10 18
Optometrist Spadina Optometry 12 Dr. Archie Chung & Associates 5 27
Personal trainer Fit Squad Training Gym 12 Fit Squad Training Gym 7 31
Physiotherapist Axis Therapy & Performance (tied with 1 other) 9 Axis Therapy & Performance 5 23
Plumber John The Plumber 8 WaterWorks Plumbing & Drains 3 17
Real estate agent Teuta Guci 8 Teuta Guci 7 19
Roofer Home Vision Inc. (tied with 1 other) 12 Alpine Roofing 8 14
SEO agency Search Engine People 15 dNOVO Group 8 12
Veterinarian Central Toronto Veterinary Referral Clinic 10 Kahu Animal Hospital 8 25
Wedding photographer Purple Tree Photography 13 Purple Tree Photography 8 27
Yoga studio Downward Dog Yoga Centre 15 Downward Dog Yoga Centre 15 24

The same question rarely gave the same first answer

Across 75 category-and-model groups of five runs, all five runs agreed on the first-named business in 10 groups (13%). On average, the most common first pick appeared in 59% of runs. Of the 582 businesses named, 369 (63%) were named by only one of the three model versions and 55 (9%) by all three. 356 of 375 answers (95%) said there was no single best option.

Four statistics. The same first pick all five times in 13% of 75 groups of five identical questions. 95% of 375 answers said there is no single best. 63% of 582 businesses named were named by only one model. 9% were named by all three models.
How much the answers varied. Source: this study.

How the three models differ

Comparison of three ChatGPT models across 125 answers each. Ended by asking the user a question: gpt-6.1-sol 100%, gpt-5.6-sol 9%, gpt-5.5 27%. Businesses named per answer: 2.6, 5.4, 5.1. Said there is no single best: 99%, 86%, 100%. Citations per answer: 2.8, 4.8, 5.1. Words per answer: 145, 180, 256.
How the three models answered. Source: this study.

Per-model behaviour across 125 answers each

gpt-6.1-sol gpt-5.6-sol gpt-5.5 (2026-04-23)
Named at least one business 115 (92%) 123 (98%) 119 (95%)
Businesses named per answer, average 2.6 5.4 5.1
Said there is no single best 124 (99%) 107 (86%) 125 (100%)
Ended by asking the user a question 125 (100%) 11 (9%) 34 (27%)
Citations per answer, average 2.8 4.8 5.1
Words per answer, average 145 180 256

Worked example, dentist: gpt-6.1-sol named no dentist in three of its five answers, the University of Toronto Faculty of Dentistry clinics in one, and Baywell Dental Centre and Bay Bloor Dental in one, each time asking which neighbourhood the user was in. gpt-5.6-sol named The Richmond Dental Centre in all five answers and led with it in four.

Where the names overlap with Best in Hood

In the mortgage broker category, ChatGPT's three most-named brokers (Outline Financial, True North Mortgage, Ratehub.ca) are Best in Hood's top three. In auto repair, ChatGPT's three most-named shops (Spadina Auto Service, EuroMechanic, Central Import) are all on Best in Hood's five-name list. In hair salons, one of ChatGPT's top three (PERFECT X HAIR) is on the Best in Hood list; The Cabinet Salon and Take Khair Beauty came from other sources. Best in Hood was cited in 18 of 25 categories, most often auto mechanic (15 of 15 answers), then barber, mortgage broker, personal trainer and roofer (10 of 15 each).

‍

Where we departed from the registered method

  • 20 of the 375 answers were collected in a second pass, about four hours after the rest. All 15 optometrist runs and the 5 chiropractor runs on gpt-5.5 failed on the first attempt with HTTP 402 (account balance exhausted) and were collected between 16:06 and 16:19 UTC the same day, after a top-up, with the same prompt, models and settings. Figures computed on the first 355 alone gave the same headline results (37% Best in Hood, Google cited in zero answers).
  • One test call was made before the category list was locked (dentist, gpt-6.1-sol) to confirm the tool returned citations. It is not included in the 375.

What this does not answer

  • Whether any named business paid for a Best in Hood listing.
  • Whether results differ by user location, logged-in account, or the consumer ChatGPT app versus the API.
  • What share of Toronto residents use ChatGPT for local recommendations.
  • Whether being on a cited directory causes a business to be named, or whether both reflect the same Google review data.

What we test next

The same 25 prompts rerun at the next ChatGPT model release, to see whether the cited sources and the named businesses move together. A separate study of Canadian online shopping prompts, collected 24 October 2026, is published alongside this one.

Edition history

Every substantive change to this page since publication

Date What changed Why
2026-10-27 First published —