On October 20, 2026 we counted every Google listing for plumbers, electricians, painters, roofers and HVAC companies within about 12–16 miles (20–25 km) of downtown Houston, Chicago and Phoenix: 10,348 listings in all (method). We recorded Google's 3-pack for each trade and city. Then we asked ChatGPT and Perplexity "Who are the best {trade} in {city}?" for the same 15 searches, plus a ChatGPT re-run of the plumber question in each city: 33 answers (method).
| What we measured (US, October 20, 2026) | Result | Base |
|---|
| ChatGPT picks also in Google's 3-pack | 22.8% | US picks |
| Perplexity picks also in Google's 3-pack | 15.3% | US picks |
| ChatGPT picks Perplexity also named | 60.8% | 79 picks |
| Review counts shown by ChatGPT that matched Google | 8 of 35 | 35 checks |
| Searches with Yelp on Google's first page | 13 of 15 | 15 searches |
| Searches with an AI Overview | 1 of 15 | 15 searches |
Source: Alfred, The AI Recommendation Gap 2026, US edition (method).
The two assistants agreed with each other far more in the US than in the UK: Perplexity named 60.8% of ChatGPT's US picks (method). Both still mostly passed over the businesses in the 3-pack. Ranking on the map does not get you named in the answer, and the reverse holds too.
The review bar is real even in a city where few businesses clear it. In Chicago, only 9.7% of 2,704 trade listings had 100 or more Google reviews (method), and the businesses the assistants named were drawn mostly from that small group. Across nine consumer industries in Chicago the share ran from 16.5% for dentists to 69.0% for restaurants, and across four cities 90.0% of the AI picks with a known review count had 100 or more (method). Every US pick we looked up and found on Google, 33 in all, was a claimed profile with a website (method).