What ChatGPT & Perplexity Actually Cite (6,500 Answers)

Reddit is the single most-cited website in Perplexity's answers about B2B software. It barely registers in ChatGPT's. We found that split after pulling every source link from about 6,500 ChatGPT and Perplexity answers we track for 68 B2B software brands, and it held up across almost every product category we looked at.

That one gap tells you most of what you need to know about optimizing for AI search in 2026: the two biggest answer engines read almost opposite parts of the web. Optimize for one and you can be invisible in the other.

Here's what our own data shows, where the two engines agree, where they diverge, and why "getting cited" turns out to be a weaker goal than most teams assume.

What we measured

We run fixed sets of buyer-style prompts against ChatGPT, Perplexity, and Google AI Overviews on a daily cadence for the brands using RankSurf. For this analysis we looked at roughly 7,700 completed answers from 68 business software brands across about 1,000 tracked prompts, current as of July 2026. For every answer we record whether the brand was named in the response, whether one of its own URLs showed up as a source, and every source link the engine returned.

Two honest caveats before the numbers. First, the 68 brands cluster in a handful of B2B software niches, including meeting assistants, social scheduling tools, and product-feedback platforms, so read this as a deep sample rather than a census of all B2B SaaS. Second, Google AI Overviews doesn't expose per-answer source links the way the other two do, so the citation breakdowns below cover ChatGPT and Perplexity, about 6,500 answers between them. Where Google AI Overviews has usable data, on whether a brand gets named at all, we've included it.

Which sources does ChatGPT actually cite?

ChatGPT leans on editorial and encyclopedic sources. Its three most-cited domains in our sample were TechRadar (548 citations), Wikipedia (539), and arXiv (108). Established tech publishers and reference sites, in other words, plus the occasional research paper.

The pattern is consistent: ChatGPT prefers a small set of high-authority publications it seems to trust, and it returns to them often. Its top ten domains account for 17% of all its citations, drawn from about 3,290 distinct domains overall. Compared with Perplexity, that's a concentrated, predictable source diet. If you know which five publications cover your category, you already know where most of ChatGPT's citations come from.

What ChatGPT mostly ignores is worth naming too. Reddit showed up just 73 times in ChatGPT answers, against Perplexity's 760. Community forums, social profiles, and user-generated posts sit near the bottom of ChatGPT's list.

Which sources does Perplexity cite?

Perplexity reads the social web. Its most-cited domain by a wide margin was Reddit (760 citations), followed by LinkedIn (365), G2 (238), Facebook (211), SourceForge (209), and GitHub (161). Forums, review sites, professional networks, and developer communities dominate.

It also cites far more, and far more widely. Perplexity named a brand's own site as a source in 30.1% of answers, nearly double ChatGPT's 15.9%. And it pulls from a broader pool: 4,743 distinct domains, with its top ten accounting for only 11.2% of citations. Where ChatGPT concentrates, Perplexity spreads.

There's a straightforward reason Reddit sits at the top. Both Google and OpenAI signed content-licensing deals with Reddit in 2024, and Perplexity's answers lean heavily on the same community discussions those deals made accessible. When a real person asks "which tool is actually good," Perplexity tends to surface where other real people already answered that question, and a lot of the time that's a Reddit thread or a G2 review page.

The two engines barely overlap

Put the two side by side and the divergence is hard to miss.

Reddit is 10x more common in Perplexity. Wikipedia is 60x more common in ChatGPT. These aren't two dials on the same machine set slightly differently. They're two different reading habits.

For a marketing team, that kills the idea of a single "AI SEO" checklist. A wave of Reddit threads and G2 reviews might lift your Perplexity presence and do almost nothing for ChatGPT, where a mention in TechRadar or a solid Wikipedia entry carries more weight. You have to work both source diets, and you have to measure them separately.

Being cited isn't the same as being recommended

Here's the finding that surprised us most, and it's the reason "get cited by AI" is an incomplete goal.

Citation and mention move independently. ChatGPT named a brand in 35.1% of relevant answers but cited that brand's own site in only 15.9%. So more than half the time ChatGPT recommends a brand, it does so from what it already knows, with no live link to that brand's site at all. Perplexity named brands in 39.3% of answers and cited them in 30.1%, a tighter gap, but still a gap.

The two outcomes even come apart in the same answer. In our sample, a brand's page was used as a source without the brand being named anywhere in the response in 5.3% of ChatGPT's cited answers and 8.0% of Perplexity's. Your content did the work of informing the answer, and a competitor got the recommendation. We call that state cited but not recommended, and it's a specific failure mode you can't see if you only track citations.

The takeaway isn't subtle. A citation is a signal, not the prize. The prize is being the brand the model names when someone asks what to buy. Those correlate, but loosely, and optimizing only for citations can leave you feeding answers that recommend someone else. This is a big enough pattern that we built tracking for it into the product, and it's the subject of a companion piece on the four states of AI visibility. Our earlier look at 1,100 answers covers the citation-earning mechanics in more depth.

What this means for B2B SaaS teams

Three things follow directly from the data.

Treat ChatGPT and Perplexity as separate channels. The content that earns a Perplexity citation, an active Reddit presence, current G2 reviews, useful answers on forums where your buyers already are, is not the content that earns a ChatGPT citation. For ChatGPT, aim at the editorial and reference sources it trusts: coverage in established publications, an accurate Wikipedia entry where you qualify, and appearances in the "best X tools" roundups those publishers run. Our guide to ranking in Perplexity goes deeper on the community-source side.

Don't outsource your whole strategy to one engine's habits. Perplexity's love of Reddit is real, but Reddit rewards genuine participation and punishes spam, and its weighting can shift the moment a licensing deal changes. Build presence on the sources you'd want cited even if the engines re-weighted tomorrow: your own well-structured pages, third-party reviews, and credible editorial coverage.

Measure mentions and citations as two different metrics. If you only count citations, you'll miss the answers where you're recommended without a link, and the answers where you're cited while a rival gets recommended. Both matter, and they tell you different things about where you stand.

The short version

The AI search story in 2026 isn't one engine's rules. It's at least two engines with opposite tastes in sources, and a gap between being read and being recommended that most tracking misses entirely. Win in AI search by working both source diets and by measuring the outcome that actually drives revenue, which is the recommendation, not just the link.