Last week I logged every source ChatGPT used to answer five "which PR agency should I hire?" questions. The obvious follow-up questions: is it just ChatGPT, and is it just one industry?
So I widened the test. Fifteen buyer questions across three B2B categories (PR agencies, cybersecurity software, CRM software) went to ChatGPT, Google AI Overviews, Perplexity and Gemini. After every answer I asked for the full URL of every source, then sorted each source by type.
Setup
- Five question types per category: broad ("best X"), industry, specific (a named guarantee or integration), head-to-head ("X vs Y") and name check ("is X any good?").
- 55 answers and 523 sources. Gemini answered 10 of the 15; Google's bot check stopped the CRM set.
- Source types were classified by URL pattern, so treat the percentages as approximate.
1. The question type decides what gets read
| Question type | What the tools read most |
|---|---|
| Broad | List articles (~48%) and vendor pages (~46%) |
| Specific | Vendors' own pages answering that exact question (~78%) |
| Comparison | Comparison and list articles (~59%) |
| Name check | Review sites and review articles (~35%), plus forums |
2. Each tool has its own diet
| Tool | Lists | Vendor pages | Reviews | Forums/video |
|---|---|---|---|---|
| Google AI Overviews | 30% | 49% | 8% | 12% |
| ChatGPT | 16% | 76%* | 7% | 2% |
| Perplexity | 58% | 36% | 6% | 1% |
| Gemini | 35% | 48% | 17% | 0% |
*ChatGPT adds a homepage link for every company it names, which inflates its vendor share.
3. Specific pages beat broad ambitions
I tracked my own agency as a test case. One specific page answers one specific question: which agency can guarantee a Business Insider placement. Google, ChatGPT and Perplexity all named us for that question, citing that page. A week earlier ChatGPT hadn't. For the broad "best agency" questions, every tool still picked the firms that appear across many independent lists.
What I'd do with this
- Broad and comparison questions: get onto the lists. This matters most on Perplexity.
- Specific questions: write one page per real buyer question, with terms and proof.
- Name checks: reviews and independent coverage. Gemini leans on these hardest.
- Google: Reddit and YouTube threads matter there in a way they don't elsewhere.
Full tables and method: jamcopr.com/ai-pr-agency-study
