Who AI recommends, and how concentrated it is: 12 categories, 5 models
240 brand-free buyer questions were asked to ChatGPT, Perplexity, Gemini, Claude, Grok between 2026-09-07 and 2026-09-07. This report aggregates the 1188 completed answers: which providers get named, how concentrated each category is, how each model behaves, and which sources the answers cite. Every number traces back to a per-industry dataset you can open.
Five things the data says
- 3 of 12 categories already have a "default answer": Project management software (monday.com, 38%); Skincare brands (La Roche-Posay, 35%); Electric bikes (Aventon, 28%). All of them are product categories. In every local-service category the most-named provider appears in 13% of answers or fewer.
- Across 1188 answers, the share that named at least one specific provider ranged from 38% to 64% by category. The rest gave guidance or pointed to directories — an answer that names nobody is a gap nobody has filled yet.
- In 6 categories, at least 20% of answers sent the buyer to a directory or review site (Furniture brands 28%, Hotels 32%, Restaurants 31%, Dentists 29%, Pet supplies 23%, Law firms 21%). Directory listings are a precondition for being named there, not an alternative to it.
- Models behave differently: Gemini names 3.5 providers per answer on average, Perplexity 2.1. Perplexity attaches citations to 100% of answers, Claude to 85%. A share-of-voice number averaged across models hides this.
- Of 11763 citations, 15% went to named providers' own sites, 4% went to communities (reddit, quora…), 4% went to directories & review sites. reddit.com was cited in 12 of 12 categories.
Which categories already have a default answer
"Top share" is the share of completed answers naming the most-named provider; "Top 3" adds the next two. Sorted by top share.
| Category | Type | Most-named provider | Top share | Top 3 | Providers | Answers naming ≥1 | Points to directories |
|---|---|---|---|---|---|---|---|
| Project management software | Product | monday.com | 38% | 101% | 103 | 64% | 0% |
| Skincare brands | Product | La Roche-Posay | 35% | 81% | 88 | 56% | 1% |
| Electric bikes | Product | Aventon | 28% | 62% | 65 | 51% | 6% |
| Furniture brands | Product | IKEA | 22% | 50% | 143 | 53% | 28% |
| Real estate agents | Local · Denver, Colorado | Compass | 13% | 24% | 127 | 50% | 17% |
| HVAC & plumbing services | Local · Phoenix, Arizona | Parker & Sons | 12% | 28% | 131 | 49% | 9% |
| Gyms & fitness studios | Local · San Diego, California | Planet Fitness | 11% | 26% | 126 | 51% | 18% |
| Hotels | Local · Miami, Florida | Four Seasons Hotel at The Surf Club | 10% | 29% | 129 | 54% | 32% |
| Restaurants | Local · Seattle, Washington | The Pink Door | 8% | 21% | 230 | 56% | 31% |
| Dentists | Local · Austin, Texas | KoLa Dentistry | 8% | 21% | 114 | 38% | 29% |
| Pet supplies | Product | PetSafe | 8% | 22% | 119 | 56% | 23% |
| Law firms | Local · Chicago, Illinois | Regas & Dallas P.C | 7% | 17% | 183 | 49% | 21% |
How each assistant answers a buying question
Averaged over all categories where the model completed answers. "Ranked list" is the share of answers that ordered the named options.
| Model | Answers | Name ≥1 provider | Providers / answer | With citations | Citations / answer | Ranked list | Points to directories |
|---|---|---|---|---|---|---|---|
| ChatGPT | 240 | 48% | 2.80 | 97% | 8.58 | 8% | 19% |
| Perplexity | 240 | 39% | 2.13 | 100% | 18.88 | 24% | 12% |
| Gemini | 228 | 73% | 3.50 | 100% | 5.90 | 27% | 13% |
| Claude | 240 | 48% | 2.41 | 85% | 5.77 | 25% | 16% |
| Grok | 240 | 55% | 2.73 | 100% | 10.18 | 18% | 30% |
Where the answers come from, category by category
Share of citations by source type. Only sources marked inside the answer text are counted.
| Category | Citations | Named providers' own sites | Directories & review sites | Communities (Reddit, Quora…) | Research, government & universities | News & media | Video | Other sites |
|---|---|---|---|---|---|---|---|---|
| Project management software | 1002 | 29% | 2% | 2% | 0% | 1% | 0% | 66% |
| Skincare brands | 1038 | 4% | 3% | 3% | 1% | 7% | 0% | 83% |
| Electric bikes | 993 | 19% | 1% | 7% | 1% | 3% | 2% | 68% |
| Furniture brands | 1046 | 15% | 3% | 6% | 0% | 4% | 0% | 72% |
| Real estate agents | 913 | 27% | 4% | 3% | 6% | 0% | 0% | 60% |
| HVAC & plumbing services | 976 | 18% | 4% | 4% | 2% | 0% | 0% | 72% |
| Gyms & fitness studios | 957 | 18% | 7% | 8% | 2% | 1% | 1% | 63% |
| Hotels | 935 | 3% | 10% | 3% | 1% | 2% | 1% | 79% |
| Restaurants | 933 | 4% | 5% | 8% | 4% | 3% | 0% | 76% |
| Dentists | 943 | 22% | 4% | 5% | 5% | 1% | 0% | 62% |
| Pet supplies | 1022 | 9% | 3% | 2% | 4% | 4% | 0% | 78% |
| Law firms | 1005 | 19% | 4% | 2% | 5% | 0% | 0% | 70% |
| All categories | 11763 | 15% | 4% | 4% | 3% | 2% | 0% | 71% |
Domains cited across the most categories
| Domain | Type | Categories | Citations |
|---|---|---|---|
| reddit.com | Communities (Reddit, Quora…) | 12 | 504 |
| alibaba.com | Other sites | 3 | 73 |
| linkedin.com | Other sites | 3 | 53 |
| pmc.ncbi.nlm.nih.gov | Research, government & universities | 3 | 28 |
| nytimes.com | News & media | 3 | 28 |
| forbes.com | News & media | 3 | 25 |
| tripadvisor.com | Directories & review sites | 2 | 75 |
| youtube.com | Video | 2 | 31 |
| yelp.com | Directories & review sites | 2 | 25 |
| zillow.com | Directories & review sites | 2 | 22 |
| businessinsider.com | News & media | 2 | 17 |
| medium.com | Other sites | 2 | 15 |
| nypost.com | Other sites | 2 | 15 |
| resources.rework.com | Other sites | 2 | 13 |
| ones.com | Other sites | 1 | 46 |
Questions teams ask before they start
Which categories does AI already have a default recommendation for?
Categories where one provider appears in at least a quarter of answers: Project management software — monday.com (38%); Skincare brands — La Roche-Posay (35%); Electric bikes — Aventon (28%).
Which AI model names the most brands?
Gemini named at least one provider in 73% of its answers, averaging 3.5 providers per answer across 12 categories.
Where does the data come from and can I check it?
Each category has its own page with the full question list, per-model counts and cited domains, and the provider table is downloadable as CSV. 1188 answers were collected between 2026-09-07 and 2026-09-07 with the same engine the product uses for customer sites.
How can a brand use this?
Find your category, read which questions produced no names at all (those are open), check which source types the answers cite, and then measure your own brand on the same questions. Concentrated categories need displacement content on the cited sources; open categories reward being the first consistent answer.
How the data was produced
- Market: United States. 12 categories, 7 of them local services with a sample city each (assistants name local providers only when a place is given).
- 20 brand-free buyer questions per category, generated to cover research, comparison, evaluation, purchase and trust intents; each asked once to ChatGPT, Perplexity, Gemini, Claude, Grok with web search enabled, collected 2026-09-07.
- 1188 of 1200 answers completed; failed answers are excluded from every percentage rather than counted as zero. Providers were extracted by a separate parser from the full answer text; directories, marketplaces, review sites, insurers and equipment vendors are never counted as providers.
- Citations are counted only from bracketed markers inside the answer text that resolve to a real URL, never from the retrieval candidate pool. A named provider's own site is detected by matching its name against the domain.
- AI answers change between runs; treat this as a dated snapshot, not a ranking. The same engine measures customer sites in the product.
- Download the full provider table as CSV · Run a snapshot for your own brand
Turn AI visibility into a repeatable growth system.
Create a project, confirm your market and start measuring the questions that matter.