HIGH SALIENCE / RESEARCH / AI ANSWERS BY INDUSTRY
RESEARCH NOTE · PUBLISHED SEPTEMBER 26, 2026 · DATASET INCLUDED
AI Answers in Four Industries: Where ChatGPT and Google Get Their Sources
We asked ChatGPT and Google the same 100 buyer questions, 25 each in fintech, healthcare, legal and ecommerce, and recorded every source both systems pointed to. Google showed an AI Overview on 94 of the 100 questions. Only 27% of the sites ChatGPT cited ranked in Google's top ten for the same question, and only 20% also appeared in Google's own AI Overview. Ranking on Google and being cited by ChatGPT are, in every industry we checked, mostly different lists.
BY MIKE HAWLEY, FOUNDER
01The Finding
Two systems, two reading lists
| Industry | AI Overview shown | Overview links from sites in Google's top ten | ChatGPT cited sites in Google's top ten | ChatGPT cited sites also in the Overview |
|---|---|---|---|---|
| Fintech | 24 of 25 | 55% | 26% | 21% |
| Healthcare | 22 of 25 | 60% | 29% | 12% |
| Legal | 23 of 25 | 42% | 35% | 24% |
| Ecommerce | 25 of 25 | 48% | 20% | 22% |
| All four | 94 of 100 | 50% | 27% | 20% |
The pattern from our software teardowns holds outside software. Google's AI Overviews lean on the sites Google already ranks: 50% of their source links came from domains in the organic top ten for the same search. ChatGPT's citations overlapped with that top ten far less, and with the Overview itself less still.
Read the columns as shares of source links (Overviews) and of cited sites (ChatGPT), counted once per answer.
02By Industry
Who each system cites, industry by industry
| Industry | Most-cited by ChatGPT (answers citing) | Most-linked in AI Overviews (Overviews linking) |
|---|---|---|
| Fintech | nerdwallet.com (9), consumerfinance.gov (8), wsj.com (3), chase.com (3), cnbc.com (2) | youtube.com (15), nerdwallet.com (12), reddit.com (11), cnbc.com (6), usnews.com (5) |
| Healthcare | healthgrades.com (2), medlineplus.gov (2), nih.gov (2), nhs.uk (2), cms.gov (2) | youtube.com (10), reddit.com (3), clevelandclinic.org (3), zocdoc.com (2), justanswer.com (2) |
| Legal | americanbar.org (6), legalzoom.com (4), justia.com (3), avvo.com (3), superlawyers.com (3) | youtube.com (13), reddit.com (9), avvo.com (4), quora.com (4), superlawyers.com (3) |
| Ecommerce | rtings.com (8), wired.com (7), tomsguide.com (5), goodhousekeeping.com (4), techradar.com (3) | youtube.com (17), forbes.com (6), nytimes.com (5), reddit.com (5), rtings.com (4) |
Fintech: ChatGPT relied on personal finance publishers and regulators. NerdWallet was cited in 9 answers and the Consumer Financial Protection Bureau's consumerfinance.gov in 8.
Healthcare: ChatGPT spread its citations thinly and leaned on government sources: 17 of its 52 cited sites were .gov health sites or the UK's NHS. No single hospital or directory dominated.
Legal: the American Bar Association (americanbar.org) was ChatGPT's most-cited source, followed by legal directories and online legal services. Google's results and Overviews leaned on the same directories plus Reddit and Quora.
Ecommerce: ChatGPT cited product testing publishers (RTINGS, WIRED, Tom's Guide). Google's Overviews linked to Google's own product listings in 21 of 25 answers, so for product questions Google's shopping data is part of the answer itself.
03Communities and Video
Reddit and YouTube: big in Google, absent in ChatGPT
04Method
How the data was collected
Questions: 25 per industry, written in the words a customer would use: "best" questions, how-to-choose questions and cost questions. Fintech covers consumer and small business money products, healthcare covers patients choosing care, legal covers people choosing a lawyer or legal service, ecommerce covers consumer product purchases. The full list is in the dataset.
Google: desktop results, United States, English, via the DataForSEO SERP API, collected on September 26, 2026 (timestamps in the files are UTC). Some AI Overviews load after the page; for those 37 questions we re-ran the same search the same evening with the Overview loaded and took its sources and top ten from that run. For 7 questions the Overview did not appear in the re-run, so their sources are missing; they still count as "Overview shown".
ChatGPT: web search on, logged out, United States, English, one run per question, via the DataForSEO ChatGPT scraper, same evening. One healthcare question returned no answer on the first run and was run once more.
Limits: 25 questions per industry is a snapshot, not a census. One run, one country, desktop only. Local questions ("best divorce lawyer") were asked without a city, so results reflect national rather than local answers. Treat the figures as directions, not constants.
05Data
Check it, do not believe it
- queries.csv (100 questions)
- per-question.csv (Google top ten, Overview sources and ChatGPT cited domains for each question)
- chatgpt-citations.csv (every ChatGPT citation with its URL)
Released under CC BY 4.0. Industry pages: fintech, healthcare, law firms, ecommerce.
06Next Step
Find out which sources shape your market.
Every market draws on a different mix of sources. The Category Salience Brief maps yours across Google and the AI surfaces, with competitors side by side.