Engine Behavior

Perplexity Cites Sources 3x More Often Than ChatGPT

Across 1,690 answers about four B2B brands, Perplexity named at least one source in 93% and ChatGPT in 28%. The ranking held for every brand we could test.

An AI answer that cites its sources is an answer you can compete for. One that cites nothing is a black box: the model is drawing on what it absorbed in training, and there is no link to earn. So before asking how to get cited, it is worth asking a blunter question — how often do these engines cite anything at all?

We measured it across every enterprise answer in our own scan data. The engines are not close.

Perplexity cited a source in roughly nine answers out of ten. ChatGPT did so in fewer than three. The two engines in the middle are indistinguishable from each other — 50.8% and 50.6% is not a difference, it is the same number twice.

The gap widens when you count citations rather than answers. Perplexity produced 64% of every citation in the dataset from 31% of the answers, because it both cites more often and cites more deeply: 11.3 sources per citing answer against ChatGPT's 4.7.

How this was measured

Every figure comes from answers our platform collected while running routine scans for live accounts — not from a benchmark built to be published.

• 1,690 answers across 112 distinct prompts and 22 scan sessions, between 14 April and 4 October 2026.
• Four B2B brands: two in industrial refractories, one in enterprise networking, one in IT trade media.
• A "citation" is a source URL attached to the answer and captured by our extraction. An engine that names a company without linking to it is not counted here — that is a brand mention, which we track separately.
• Excluded: our own brand, because research that includes the publisher as a data point is marketing. Also excluded are Local-product scans, which in this period were 97% Perplexity and would have inflated exactly the engine already leading.

The engine samples are uneven — 529 answers for Perplexity against 299 for ChatGPT — because accounts configure different engine mixes. That affects the precision of each rate, not the ordering, which is separated by far more than the sampling error at these sizes.

Does it hold per brand, or is one account carrying it?

A single dominant account can manufacture any average, and one brand does supply over half our answers. So the finding is only worth anything if it survives being split apart.

It holds. In all three brands with coverage on all four engines, Perplexity cites most and ChatGPT cites least — unanimous, with no exceptions to explain away. The middle two swap places twice each, which is what you would expect of two engines that are genuinely tied.

The networking brand is the instructive case. Three of four engines cited almost nothing — ChatGPT literally zero times in 24 answers — while Perplexity still cited in 70%. Questions about a large, famous company appear to be answered from what the model already knows, with no need to go and look anything up.

The engines that cite most are hardest to be cited by

A citation slot is not the same as a citation slot you can win. We also recorded how often each engine's citations pointed at the tracked brand's own website, and the ordering inverts.

Perplexity hands out the most citations and is the least likely to hand one to you. Claude is the opposite: it cites in half its answers, but when it does, better than one citation in nine goes to the brand's own site. The two behaviours describe different jobs. On Perplexity you are competing against the open web for a slot in a long list. On Claude and ChatGPT there are few slots, and your own site is a realistic candidate for one of them.

What actually gets cited: your competitors

Ranking the cited domains makes the competitive picture uncomfortably concrete. Setting aside the tracked brands' own sites — which were two of the five most-cited domains — the list is dominated by their direct competitors' websites.

• Of the ten most-cited third-party domains, seven were competing manufacturers' own websites. When an engine answers a question about this category, it is largely reading the competition's marketing.
• The single most-cited domain was a market-data aggregator with 542 citations — and every one came from Perplexity. One engine's view of an entire category leaned heavily on one source no brand in it controls.
• LinkedIn and a business directory also appeared in the top twelve, again Perplexity-only. Third-party profiles are doing work that brand sites are not.

The practical reading: a competitor's product page can be the source an engine uses to answer a question about you. Publishing more pages on your own domain does not change that by itself, because your domain was never the contested surface — theirs was.

What this does not prove

• Four brands is a small sample, two of them in the same narrow industry. The engine ordering was unanimous across them, but four is four.
• Our prompt set is B2B and commercial. Consumer questions, news, or anything time-sensitive could behave completely differently — Perplexity's lead may simply be larger on questions that reward looking things up.
• This is a six-month window ending 4 October 2026. Engines change their retrieval behaviour without announcing it, and any of these numbers could move.
• Citation rate says nothing about accuracy, sentiment, or whether the answer recommended anyone. An engine citing six sources can still get you wrong.
• We measure citations our extraction captured. An engine that consulted a source without surfacing a link is invisible to this method.

The finding we would stand behind is narrow and specific: across these brands and this period, the four engines differ enormously in how often they cite anything, the ordering was consistent in every brand we could test, and the engine with the most citation slots is the one where those slots are least likely to be yours.

Questions