Flash Original research · September 2026

Only One AI Assistant Shows Its Sources

We analysed 8,371 Perplexity answers and 362,179 source citations across 61,571 distinct websites. Then we checked whether ChatGPT, Claude and Gemini cite anything at all. Mostly, they do not.

95.1%

of Perplexity answers carry numbered citations

0.0%

of ChatGPT answers do

43.3

sources cited per Perplexity answer

45.8%

of them cite Reddit

Star
Flash Methodology

How We Measured This

Ranklytics sends real prompts to each AI assistant and captures the full response. This report covers 8,371 Perplexity answers run between February and September 2026 across 619 websites and 7,366 keywords.

Perplexity returns a structured list of sources alongside its answer. We verified these are genuinely sources rather than brands named in passing: only 6.7% of the cited URLs appear anywhere in the answer text itself. They are supporting references, not prose mentions.

For ChatGPT, Claude and Gemini there is no such list to read. That absence is the first finding, and it is also why this report does not rank those three against Perplexity on citation counts — there would be nothing on the other side of the comparison. Their brand-naming behaviour is covered in a separate report.

What we counted

  • Every source domain in Perplexity's citation list
  • Deduplicated within a single answer
  • Presence in an answer, and share of total citation volume, reported separately
  • Google's vertexaisearch redirect wrapper excluded as an artifact
  • Numbered [n] markers in response text used to test whether the other three cite at all

What this is not

  • Not a like-for-like ranking of all four assistants — only Perplexity publishes sources
  • Not a random sample; keywords are customer-selected and skew commercial and SMB
  • 94.6% of checks ran in August 2026, so no trend over time is claimed
  • Google AI Overviews are not included
  • Answers vary run to run; this is a snapshot
Star
Flash Finding 01

Three of the Four Cite Nothing

We looked for numbered citation markers — the [1], [2] footnotes that indicate an answer is grounded in retrieved sources — across all 30,510 answers we captured: 8,930 from Gemini, 8,371 from Perplexity, 7,194 from Claude and 6,015 from ChatGPT. This is the one comparison in this report that spans all four assistants, because it asks whether a citation surface exists at all rather than what is in it.

Perplexity95.1%
Gemini6.5%
Claude0.1%
ChatGPT0.0%

Perplexity marks its claims in 95.1% of answers. ChatGPT does so in 0.0% — two answers out of 6,015.

This is an architectural difference, not a quality judgement. Perplexity is built as a search product that retrieves and attributes. The others answer from the model and attribute nothing, which means there is no citation surface for a publisher to appear in at all.

The practical consequence is uncomfortable: for three of the four major assistants, there is currently no mechanism by which being a trusted source earns you a visible credit.

Star
Flash Finding 02

What Perplexity Actually Cites

Share of Perplexity answers containing at least one citation to each domain.

#SourceShare of answers
1reddit.com45.8%
2en.wikipedia.org39.4%
3linkedin.com22.9%
4dictionary.cambridge.org12.8%
5merriam-webster.com11.9%
6play.google.com10.5%
7britannica.com10.2%
8medium.com9.9%
9de.wikipedia.org9.7%
10yelp.com9.3%
11collinsdictionary.com8.9%
12dictionary.com8.6%
13pmc.ncbi.nlm.nih.gov8.4%
14amazon.com7.9%
15en.wiktionary.org7.5%

Reddit is the single most-cited source, appearing in 45.8% of answers — ahead of Wikipedia at 39.4% and LinkedIn at 22.9%.

Dictionary and encyclopedia sites occupy four of the top twelve places, which says more about the definitional questions in our keyword sample than about Perplexity's preferences. We flag it rather than bury it.

Star
Flash Finding 03

Reddit Is in Half the Answers — and 1% of the Citations

These two facts are both true and are routinely confused. Reddit appears in nearly half of Perplexity's answers. It also accounts for roughly one percent of Perplexity's total citation volume.

The reconciliation is that Perplexity cites 43.3 sources per answer on average (median 40, interquartile range 19–64). Reddit is usually one of them. The rest are overwhelmingly independent sites, which is where almost all the volume sits.

That figure is roughly double the ~22 citations per response reported elsewhere. We are not claiming the other number is wrong: our prompts are commercial and SMB-weighted, which pulls longer source lists, and Perplexity's citation behaviour varies by model and has changed over time. We report the distribution above so the difference can be judged rather than guessed at.

Independent sites92.5%
Reference & encyclopedic3.6%
Professional & social1.7%
Community & forums1.1%
App stores & big tech1.1%

So the widely repeated claim that AI answers are dominated by Reddit is half right. Reddit has extraordinary reach across answers and a small share of citations. If you are deciding whether to invest in Reddit presence, reach is probably the number you care about — but quoting the two interchangeably produces nonsense.

Star
Flash So What

What Follows From This

Three consequences, in rough order of confidence.

01

Citation strategy is Perplexity strategy

For now, it is the only assistant where being a cited source is visible at all. Optimising for citation across “AI” generally means optimising for one product.

One surface
02

Reach and share are different numbers

Reddit is in 45.8% of answers and about 1% of citations. Decide which question you are asking before quoting either.

Presence Volume
03

The long tail carries the volume

Independent sites take 92.5% of all citations across 61,571 distinct domains. Being the specific right answer still works.

Specificity
Star
Flash FAQ

Common Questions

Because three of them do not publish citations. ChatGPT carried numbered markers in two answers out of 6,015. Ranking them against Perplexity would produce a chart where three bars are near zero for a structural reason, not a behavioural one, and readers would draw the wrong conclusion.
They are deliberately separate. That report measures which brands ChatGPT, Claude and Gemini name inside their answers, read from the answer text. This one measures which sources Perplexity cites, read from its citation list. Mixing the two would compare a citation list against in-prose mentions, and we do not.
Yes, with a link to this page. If you need a specific cut of the data for something you are writing, get in touch and we will run it.
Sometimes, and more than the other assistants, because the citation is a visible link. But many answers resolve the question without a click. Treat it as attributed reach rather than a traffic channel.