Which AI Engines Cite Which Sources: A Per-Platform Citation Behavior Comparison

Roughly 11 percent of the source domains ChatGPT cites are also cited by Perplexity. That figure, drawn from a March 2026 audit of 680 million citations, reframes the practical question of which sources do AI engines cite. There is no shared list to optimize against. A content team that tracks one engine and assumes the finding generalizes is reading a small fraction of the landscape.

The four surfaces most marketing organizations care about, ChatGPT, Perplexity, Gemini, and Google AI Overviews, each run a distinct citation diet. The same brand can be prominent on one and absent from the next for reasons that have little to do with content quality and a great deal to do with which platforms each engine trusts. This comparison sets out what each engine favors, where the four converge, and how a team decides which surface to work first.

The short answer: four engines, four citation diets

The four engines do not cite the same sources. Pairwise domain overlap between them runs roughly 16 to 59 percent depending on the pair and the methodology, with the ChatGPT and Perplexity pairing at the bottom of that range. Each engine also has a recognizable character:

• ChatGPT is the encyclopedic-and-press engine. Wikipedia and mainstream American trade and tech press (Forbes, TechRadar, Reuters, PCMag, LinkedIn, PRNewswire) dominate its citations. It links out in only about 31 percent of responses.
• Perplexity is the community-and-primary-source engine. Reddit anchors it, but NIH and PubMed, government sites, Britannica, Gartner Peer Reviews, G2, and Yelp all appear in its top lists. It returns external links in more than 77 percent of responses.
• Gemini is the diversified Google-ecosystem engine. YouTube, Reddit, Wikipedia, Medium, PCMag, and Forbes share the spotlight, alongside Google properties such as Maps, Scholar, and the Knowledge Graph. It cites the least news media of the four.
• Google AI Overviews is the social-platform engine. YouTube, Reddit, Facebook, LinkedIn, Quora, Wikipedia, and Google.com itself account for roughly 70 percent or more of its citations.

The practical consequence is that the same brand can be cited on ChatGPT through a Wikipedia article, on Perplexity through a Reddit thread, on Gemini through a YouTube transcript, and on AI Overviews through a Quora answer. The visibility story is engine-specific even when the brand is the same.

Per-engine citation comparison at a glance

Two reading notes. "Most diversified" refers to the lowest single-source concentration across categories, not the largest citation volume. Where a source does not appear in an engine's top list, the entry reflects absence from that list rather than a measured zero.

ChatGPT vs Perplexity sources: the widest documented gap

The ChatGPT vs Perplexity sources comparison is the sharpest contrast in the current research, and it is worth treating as the reference case for how far two engines can diverge.

ChatGPT's top domains run Wikipedia, Reddit, Forbes, TechRadar, Reuters, NYPost, LinkedIn, PRNewswire, Medium, G2, and Business Insider. Wikipedia is about 7.8 percent of all ChatGPT citations but roughly 47.9 percent of its top-10 cited sources, which means Wikipedia carries far more weight in ChatGPT's upper tier than its aggregate share suggests. The press ChatGPT favors is trade and technology rather than prestige: PCMag, PRNewswire, Mashable, Tom's Guide, and TechTarget appear more often than the Wall Street Journal or the New York Times. ChatGPT cites Reuters approximately fifty times more often than Claude does on the same prompts.

Perplexity's top domains run Reddit, LinkedIn, YouTube, Wikipedia, NIH and PubMed, Gartner, Yelp, NerdWallet, TripAdvisor, Britannica, and PCMag. NIH, PubMed, and Britannica appear in Perplexity's top lists and not in ChatGPT's or Gemini's, which marks a genuine tilt toward primary and reference sources. Perplexity also cites approximately 1,430 unique news outlets, against roughly 881 for Google AI Overviews and 707 for OpenAI, so its news footprint is the most diffuse of the group. Academic analysis of Perplexity's news citations finds a pronounced left-liberal concentration in the outlets it surfaces.

The link behavior gap matters as much as the domain gap. Perplexity returns external links in more than 77 percent of responses; ChatGPT does so in about 31 percent, meaning roughly two-thirds of ChatGPT answers contain no clickable source at all. For a team that measures success in referral clicks rather than influence, Perplexity is structurally the easier surface. For a team that measures brand presence inside the answer text, the distinction between a citation and a mention becomes the thing to track.

ChatGPT: the encyclopedic-and-press engine

ChatGPT rewards two asset classes above all others. The first is Wikipedia, whose share of ChatGPT's top-tier citations is higher than on any other engine. The second is American trade and technology press, whose concentration inside ChatGPT has grown roughly fivefold between 2024 and 2026.

The implication for a content program is uncomfortable but clear. A strategy built entirely on owned blog production will under-index on ChatGPT, because the surfaces ChatGPT leans on are ones the brand does not control. Wikipedia notability, trade-press placements, and press-release distribution move ChatGPT citations more reliably than another well-optimized blog post does.

ChatGPT is also the most volatile of the four. A SEMrush study covering 13 weeks and 230,000 prompts recorded ChatGPT's Reddit citation share falling from approximately 60 percent to approximately 10 percent over roughly two weeks in September 2025, with Wikipedia dropping from approximately 55 percent to under 20 percent and Forbes roughly doubling in the same window. Any claim about ChatGPT's source mix that lacks a date is unreliable within a quarter.

Perplexity: the community-and-primary-source engine

Reddit is Perplexity's center of gravity, sitting at approximately 6.6 percent of citations globally and rising to approximately 46.7 percent in the product, software, and comparison categories where B2B buyers ask most of their questions. For a software brand, presence in relevant subreddit discussion is not a supplementary tactic on Perplexity; it is the main channel.

Beyond Reddit, Perplexity's profile favors sources with institutional or primary authority. Health, science, finance, and technical queries pull NIH, PubMed, government sites, and Britannica far more often here than on the other three engines. Review and comparison platforms carry real weight as well, with G2, Gartner Peer Reviews, NerdWallet, and TripAdvisor appearing in top lists. Yelp sits at roughly 0.8 percent of Perplexity citations globally but matters disproportionately for local and product intent.

Recency compounds all of this. Perplexity and AI Overviews reward upvoted, recent Reddit threads more strongly than the other engines do, which makes thread freshness part of the visibility calculation rather than a background variable.

Gemini: the diversified Google-ecosystem engine

Gemini's standalone chat draws only about 5.7 percent of its citations from news or trade press, the smallest share of any major engine. The remaining share comes from community platforms, reference sites, Google properties, video, and reviews in roughly balanced proportions. Gemini is the engine where no single category dominates.

Two structural advantages define it. YouTube is the first: video sits at or near the top of Gemini's source list, larger than Reddit or Wikipedia by most measures, and because YouTube citations are transcript-derived, chapter structure, captions, and descriptions become extraction surfaces rather than metadata. The second is Google's own properties. Gemini surfaces Google Maps, Google Scholar, and Knowledge Graph data heavily, and it is the only engine of the four with that pull. A well-maintained Google Business Profile or Scholar record does more for Gemini visibility than for any other surface.

The practical read is that a diversified content mix, spanning video, forum participation, reference pages, and reviews, outperforms single-format strategies on Gemini, and that owned Google surfaces deserve treatment as citation assets rather than local-SEO housekeeping.

Google AI Overviews: the social-platform engine

AI Overviews source types cluster more tightly than any other engine's. Reddit, YouTube, Facebook, LinkedIn, Quora, Wikipedia, and Google.com together account for roughly 70 percent or more of its citations. A brand missing from at least two of those platforms is functionally invisible on the surface.

Three details separate AI Overviews from the rest. YouTube overtook Reddit as its most-cited social source by mid-2026, at approximately 21.1 percent against approximately 18.5 percent, based on an Ahrefs analysis of more than three million United States queries. Quora is meaningful here and close to invisible elsewhere, ranking around fourth in Google AI Mode at roughly 7.25 percent of responses, which makes it a Google-surface lever specifically. Wikipedia's share has trended down from approximately 6 percent in 2025 to approximately 4.8 percent by mid-2026 even as its absolute citation count has grown.

The strongest signal for planning is the relationship with organic search. AI Overviews shares roughly 38 percent of its top sources with Google's organic top-10 results, while ChatGPT's overlap with those same results is only about 6 to 8 percent. Traditional ranking work therefore translates into AI Overview presence at roughly five times the rate it translates into ChatGPT presence. Brands already winning Google organic tend to inherit AI Overview citations; ChatGPT visibility remains a separate program.

Where AI engines pull sources by category

Viewing the same data by source type rather than by engine clarifies where AI engines pull sources and which assets serve more than one surface.

Wikipedia and reference. Highest on ChatGPT at roughly 7.8 percent overall and roughly 47.9 percent of top-10 citations. Middle tier on Perplexity at 5 to 6 percent, supporting on Gemini at 4 to 5 percent, and declining on AI Overviews at roughly 4.8 percent. Wikipedia's human pageviews are down approximately 8 percent year over year, with a median decline of 33 percent across a 31-article sample, so engines are citing Wikipedia more while people visit it less.

Reddit and forums. Dominant on Perplexity and second only to YouTube on AI Overviews at approximately 18.5 percent, which is why AI Overviews source types skew social rather than editorial. Moderate on Gemini. Reduced to approximately 3 percent on ChatGPT after September 2025.

YouTube and video. The top source on AI Overviews at approximately 21.1 percent and at or near the top on Gemini. Rising on Perplexity from a small base of roughly 2 percent. Minimal on ChatGPT.

News, journalism, and trade press. Widest outlet spread on Perplexity at roughly 3.1 percent of citations across approximately 1,430 unique outlets. Roughly 4.3 percent globally on ChatGPT, reaching approximately 15.4 percent trade press for specific queries. Roughly 4.6 percent and most diversified on Gemini. Lowest on AI Overviews at approximately 2.75 percent.

Review and community platforms. G2, Gartner Peer Reviews, Yelp, TripAdvisor, and NerdWallet feature on Perplexity. G2, PCMag, and Tom's Guide feature on ChatGPT. PCMag, Medium, and Forbes feature on Gemini. Quora substitutes for the review and forum role on AI Overviews, where Yelp does not appear meaningfully.

Owned versus earned. Approximately 82 to 89 percent of all AI citations come from earned media: press, forums, Wikipedia, reviews, and third-party listicles. Owned brand websites are a minority share on every engine without exception, which is the most consequential pattern in the entire dataset.

The patterns that hold across all four engines

Three findings survive the differences between engines and should shape planning regardless of which surface a team prioritizes.

Earned media carries the load. At 82 to 89 percent of citations across editions of the research, third-party coverage is the largest source category everywhere. Owned production still matters, because it is what press, forums, and reviewers reference, but a program that treats the company blog as the primary citation asset is optimizing the smallest slice.

Concentration is extreme. The top 15 domains absorb approximately 68 percent of all AI citations, and that concentration has held since 2024. A relatively small number of platforms determines the majority of citation volume, which makes platform selection a higher-leverage decision than page-level optimization for most teams.

Brand-level overlap exceeds source-level overlap. Two engines can share almost no citation URLs while both citing the same brand through different pages. A brand can therefore be present across all four engines with no single page shared between them, which is why per-engine measurement matters more than a single aggregate visibility score.

Volatility: why any AI engine citation comparison needs a date stamp

An AI engine citation comparison describes a moving system. Citation volatility is now measured in weeks rather than years, and three documented shifts illustrate the pace.

The September 2025 ChatGPT shift is the largest on record: Reddit from approximately 60 percent to approximately 10 percent of prompt responses and Wikipedia from approximately 55 percent to under 20 percent, both inside roughly two weeks. YouTube overtaking Reddit on AI Overviews by mid-2026 is the second. The fivefold growth in trade-press concentration inside ChatGPT between 2024 and 2026 is the third.

Two definitional cautions belong alongside those numbers. Google AI Overviews and Google AI Mode are related but not identical surfaces, and most published research measures AI Overviews. Gemini's standalone chat also differs from Gemini as it operates inside Google Search, where rankings correlate more strongly with Google organic results. Comparisons that blur either pair will produce conclusions the data does not support.

Tracking the per-engine difference in practice

A cross-engine comparison is only actionable if a team can see its own numbers rather than the aggregate market's. The measurement problem has three parts: a fixed set of buyer prompts, per-engine collection on a schedule, and a record of which specific sources each engine returned. Screenshots do not scale to that, and manual spot-checks cannot detect a shift the size of September 2025 while it happens.

DeepSmith is an AI search analytics and content production platform built around that loop. Its AEO module tracks mention rate, citation rate, and share of voice with a per-platform breakdown, reports the sources AI cites most for a tracked prompt set, shows which of a brand's own pages earn citations, and maps which competitor pages win citations on which platform. Engine coverage rises by plan: Pro at $99 per month tracks ChatGPT, Grow at $199 adds Perplexity, Scale at $399 adds Gemini, and Enterprise covers all five named engines, which are ChatGPT, Gemini, Perplexity, Claude, and Google AI Mode. Annual billing lowers those rates to $80, $160, and $299 respectively.

Two honest constraints follow from the data above. Single-engine tracking on the entry plan sees ChatGPT only, and with domain overlap as low as 11 percent between engines, one engine is not a proxy for the others; ChatGPT is nonetheless the largest press-and-Wikipedia footprint and the surface where brand-driven buyer research concentrates, and Grow and Scale extend coverage to the engines whose source diets differ most. DeepSmith also covers Google AI Mode rather than Google AI Overviews, which are related Google surfaces but not the same one; teams whose priority is the AI Overviews box specifically should weigh that difference against the value of tracking four other engines in the same workspace.

The second half of the loop is what distinguishes the platform from a monitoring tool. Because the same system holds the brand's product context, personas, brand voice, and content types, the gaps the tracking surfaces feed directly into production, and the articles that come out are publish-ready rather than first drafts waiting on a rewrite.

Which engine to prioritize, by situation

Full four-engine coverage is the correct answer for a well-resourced brand team. For everyone else, the sequence depends on the business.

• Solo founder or small team with a consumer product. Reddit first, which serves Perplexity and residually ChatGPT, then YouTube for AI Overviews and Gemini, then a Wikipedia presence. Yelp and G2 follow where the category warrants them.
• Mid-market B2B SaaS. Reddit for Perplexity, trade-press placements for ChatGPT, and YouTube for Gemini. G2 and Gartner Peer Reviews compound slowly and are worth starting early.
• Enterprise content and brand team. Cover all four engines and treat earned media as the primary channel, with owned-channel citations as a deliberate second.
• Local or multi-location business. Yelp and Reddit for Perplexity, YouTube for AI Overviews, and a maintained Google Business Profile for Gemini.
• Health, medical, or science publisher. NIH and PubMed visibility for Perplexity, trade and consumer health press for ChatGPT, and explainer video for AI Overviews and Gemini.

If only one engine can be tracked, the choice follows the goal. ChatGPT suits brand and public-relations-led visibility. Perplexity suits click-through and earned-media conversion. Gemini suits Google-ecosystem reach. AI Overviews suits organic-search-led programs already winning on Google.

Start measuring the difference

The cross-engine comparison in this article describes the market. A brand's own numbers will differ, sometimes substantially, and the only way to know is to track a fixed prompt set on each engine and read the sources it returns. Start a 7-day DeepSmith free trial to see the per-platform breakdown for a real prompt set, with no long-term contract.