Which Engine Cites What: Perplexity Barely Shares a Source Diet With ChatGPT
Published October 16, 2026 · Adam Rock
The short version
We logged 17,223 source citations across 7 engines and 5 brands between 30 September and 15 October 2026. There is no single AI source diet. ChatGPT and Gemini lean on Reddit and YouTube; Perplexity cites neither among its top sources. So "get on Reddit, make YouTube videos" is a ChatGPT and Gemini play, and does little on Perplexity.
What we measured
- Universe: 5 brands, one per category (CRM, project management, ecommerce, AI-visibility tools, running shoes), 6 buying questions each - 30 questions - run daily over 16 days (30 Sep - 15 Oct 2026).
- Engines: ChatGPT, Gemini, Google AI Overviews, Perplexity, Claude, Grok and Microsoft Copilot. Three of them (ChatGPT, Gemini, Perplexity) ran all 30 questions; the other four ran a single CRM question.
- Window: 30 September 2026 to 15 October 2026 (UTC).
- Measure: citations the engines made to a publication (classified as info_site), counted per engine.
- Sample: 17,223 source citations out of 18,943 total records.
- Coming next: DeepSeek joins the set next week, and its cut will be published once it has real answers behind it.
The finding
Source citations per engine, 30 Sep - 15 Oct 2026.
| Engine | Source citations | Distinct URLs | Brands covered |
|---|---|---|---|
| Perplexity | 7,693 | 1,616 | 5 |
| Gemini | 4,679 | 1,034 | 5 |
| ChatGPT | 3,917 | 2,355 | 5 |
| Grok | 382 | 139 | 1 |
| Google AI Overviews | 215 | 36 | 1 |
| Microsoft Copilot | 184 | 20 | 1 |
| Claude | 153 | 24 | 1 |
| Total | 17,223 | - | - |
Top sources per engine (top four each).
| Engine | #1 | #2 | #3 | #4 |
|---|---|---|---|---|
| ChatGPT | reddit.com (521) | techradar.com (420) | en.wikipedia.org (113) | youtube.com (92) |
| Gemini | youtube.com (498) | reddit.com (383) | technologyadvice.com (105) | runrepeat.com (103) |
| Google AI Overviews | youtube.com (56) | reddit.com (31) | forbes.com (20) | - |
| Perplexity | runnersworld.com (324) | forbes.com (256) | clickup.com (236) | rtings.com (211) |
| Microsoft Copilot | forbes.com (17) | fitsmallbusiness.com (17) | crm.org (17) | capterra.com (16) |
| Grok | itechguides.com (19) | zdnet.com (17) | pcmag.com (15) | forbes.com (13) |
| Claude | techradar.com (33) | stackshare.io (20) | fayedigital.com (17) | fitsmallbusiness.com (10) |
Three things stood out
- There are engine "families", and the usual advice only serves one. Among the three engines that ran the identical 30 questions, ChatGPT and Gemini share a Reddit plus YouTube plus publisher diet while Perplexity leans on specialist review and benchmark sites (runnersworld, rtings, runrepeat). So "get on Reddit, make YouTube videos" is a ChatGPT and Gemini play and does little for Perplexity. The other four engines ran a single CRM question each, so I would not read their source lists as characteristic of them.
- Being cited on one engine says little about the next. reddit.com appears across five engines, while en.wikipedia.org drew 113 citations on exactly one (ChatGPT). forbes.com reached all seven engines; runnersworld.com reached two.
- The risk is absence, not negative sentiment. Of the records we scored, roughly 93% were neutral, 6.6% positive and 0.1% negative. Read that carefully: our sentiment classifier is keyword-based with a deliberately wide neutral band, so neutral is the default bucket rather than a precision measure. The defensible reading is that AI rarely says anything strongly polarised about a brand. The failure mode that actually costs you is not being in the answer at all.
What to do about it
- Segment your source strategy by engine. A Reddit and YouTube play is an OpenAI, Gemini and AI Overviews play. For Perplexity and Claude, the equivalent is getting into the independent review and benchmark sites that already rank.
- Buy the evidence layer, not the assertion. Engines reuse sources that corroborate a claim in public. A review, a comparison table or a test result is quotable; a homepage claim is not.
- Report per engine. A single blended visibility score hides the engine where you are actually missing.
The limits
- Coverage is uneven. Three engines (ChatGPT, Gemini, Perplexity) ran all 30 questions across all 5 brands, so those three are directly comparable to each other. The other four ran one CRM question, so their rows are a small sample and are not comparable to the first three.
- Rows, not pages. Counts are raw citation rows; distinct URLs are reported separately, which is why ChatGPT shows 2,355 distinct URLs against 20 for Microsoft Copilot.
- Sentiment is coarse. A keyword model with a wide neutral band. Do not read the ~93% as a precision measure.
- Claude was under-extracted. An extraction bug dropped the Claude sources for the whole window; 153 rows were recovered from the stored raw responses. Treat single-engine numbers as provisional.
- Engine set. This window covers the seven engines that ran. DeepSeek joins the set next week.
- No paid placements were involved, and we make no claim that any of this causes citations.
The dataset
The numbers above, the method box, the limitations and the SQL we used are published as a first-party dataset: /data/ai-citation-sources-per-engine-2026-10/.
Frequently Asked Questions
Do all AI engines cite the same sources?
No. ChatGPT, Gemini and Google AI Overviews lean on Reddit and YouTube, while Perplexity leans on specialist review and benchmark sites and Claude on editorial reviews and vendor blogs.
How many citations did you measure?
17,223 source citations out of 18,943 total records, across 7 engines and 5 brands from 30 September to 15 October 2026.
What is coming next?
DeepSeek joins the monitored set next week; its per-engine cut will be published once it has real data behind it.