Experiment
We kept seeing Reddit, forums and review-led content appear prominently in recommendation responses. We wanted to know whether this was a repeatable pattern or simply anecdotal.
Our hypothesis
First-hand sources accounted for 58% of citations across
recommendation-style prompts…
| AI tool | First-hand experience | Brand content | News / editorial | Other |
|---|---|---|---|---|
| ChatGPT | 62% | 18% | 12% | 8% |
| Claude | 54% | 24% | 14% | 8% |
| Perplexity | 59% | 21% | 13% | 7% |
| Gemini | 56% | 26% | 11% | 7% |
A representative result from ChatGPT illustrates the pattern.
TEST PERIOD
QUERY TYPES
AI TOOLS
CATEGORIES
QUERIES TESTED
CITATIONS ANALYSED
The experiment used a fixed prompt set and was conducted during a single seven-day period…
We will repeat the test across additional verticals and isolate whether source preference changes by platform.
Have a question?
We’re always testing new ideas. Tell us what you’d like us to investigate.
Got a hypothesis?
We’re always testing new hypotheses. Tell us what you’d
like to see us explore.
Short updates, no spam. Just new findings from our research.