How ChatGPT, Perplexity, and Google AI Overviews Actually Choose What to Cite: A Platform-by-Platform GEO Playbook

ChatGPT, Google AI Overviews, and Perplexity each pull citations from almost entirely different pools of websites. Only 11% of the domains ChatGPT cites also show up in Perplexity’s answers, according to a March 2026 analysis by Averi built on a dataset of 680 million AI citations. If you are optimizing for one platform, you are functionally invisible on the others.

That single number breaks a lot of GEO strategies. Most teams still treat getting cited by AI as one problem with one fix: publish authoritative content, mark it up with schema, wait for citations to show up. The data says that approach only works for whichever platform your content style happens to match, and it might not match any of them well. Each engine has a distinct, measurable preference for source type, content structure, and even domain extension, and those preferences barely overlap.

This post breaks down what the research actually shows about how ChatGPT, Google AI Overviews, Perplexity, Gemini, and Claude select and weight sources, then lays out a platform-by-platform playbook instead of a one-size-fits-all checklist. If you want a baseline read on where your brand currently stands before working through this, our GEO audit checks citation visibility across all five platforms at once, individually rather than blended into one score.

Why platform-specific GEO matters now, not eventually

Buyer behavior has moved faster than most marketing teams’ roadmaps. Responsive’s 2025 buyer survey found that 25% of B2B buyers say generative AI has overtaken traditional search as their primary tool for vendor research. G2’s 2025 buyer report puts the shift even higher: 87% of B2B software buyers say tools like ChatGPT, Perplexity, and Gemini are changing how they research software. Forrester’s 2026 B2B buying study named generative AI interactions as the single most cited touchpoint type during the research phase of a purchase, ahead of vendor websites and analyst reports.

The traffic that does arrive from AI platforms behaves differently once it lands, too. Semrush’s 2025 analysis found AI-referred B2B traffic converting at roughly 4.4 times the rate of organic search traffic. Separate portfolio data from DerivateX’s 2026 benchmark study put AI-referred conversion at approximately 17.9% against 2.8% for Google organic, a comparable multiple from an independent dataset. Whatever the exact ratio looks like in your vertical, the pattern holds across every source we found: fewer visitors arrive from AI platforms, but the ones who do arrive with far higher intent, because a chatbot conversation is already closer to a buying decision than a keyword search is.

That’s the case for caring about AI citations at all. The case for treating each platform separately, rather than running one blended GEO checklist, is what the citation research shows next.

The research: 680 million citations, five platforms, almost no overlap

The most detailed public dataset on this comes from Profound’s citation pattern study, which analyzed 680 million individual citations across ChatGPT, Google AI Overviews, and Perplexity between August 2024 and June 2025. Averi’s follow-up analysis in March 2026, working from a comparable dataset, found domain overlap between ChatGPT and Perplexity sitting at just 11%. A separate audit from Passionfruit measured overlap across three platforms at 12%, close enough to the first figure to treat the number as reliable rather than an artifact of one methodology.

In practical terms: getting cited consistently by ChatGPT tells you almost nothing about whether Perplexity will ever mention your brand. They are drawing from different libraries of source material, built on different retrieval logic, and optimized for different kinds of answers. A content strategy tuned for one will, at best, accidentally help with another. It will not reliably transfer.

How ChatGPT actually picks sources

ChatGPT leans hard on established, encyclopedic sources. In the Profound dataset, Wikipedia was ChatGPT’s single most-cited source at 7.8% of all citations, and within ChatGPT’s top 10 most-cited sources specifically, Wikipedia accounted for 47.9% of that top-10 volume. The rest of the top 10 skews toward reference media and long-established editorial outlets rather than blogs, forums, or newer publications.

What that means practically: ChatGPT rewards content that reads like a reference entry rather than a sales page or a hot take. Clear definitions, stable terminology used consistently across your site, a neutral explanatory tone, and structure that mirrors how an encyclopedia article organizes a topic, definition first, then history or context, then mechanics, then comparison, then caveats, tend to perform better here than opinion-driven or narrative content built primarily to persuade. It also means getting your product or category correctly and consistently described on reference-style properties, where that’s earned and appropriate, has outsized value for this one platform specifically, even though it does little for Perplexity’s very different selection logic.

Our GEO glossary is built around exactly this format for that reason: stable, citable definitions written the way a reference source would write them, rather than shifting marketing copy that changes every quarter with a new campaign angle.

How Google AI Overviews actually picks sources

Google AI Overviews behaves the most like traditional search of the three, but with a twist. In the same 680-million-citation dataset, Reddit was the top overall source for AI Overviews at 2.2% of total citations, followed by YouTube at 1.9%. Narrowed to the top 10 most-cited sources, the split becomes more concentrated: Reddit accounts for 21.0% of top-10 citations and YouTube 18.8%, meaning nearly 40% of AI Overviews’ highest-volume citations go to just two platforms built on user-generated content rather than brand-published pages.

That doesn’t mean traditional SEO signals stopped mattering for this platform. It means they stopped being sufficient. Ahrefs’ 2025 research found that only 12% of URLs cited by AI overlap with pages ranking in Google’s top 10, which cuts the other way from what most teams assume: ranking well organically is not enough to get cited in the AI Overview sitting directly above those rankings. AI Overviews increasingly draws on a separate evaluation layer that favors direct, extractable answers, sourced figures, and consistent structured data over general ranking strength alone.

The stakes here are also rising fast. Averi’s 2026 tracking found AI Overviews now appearing on 48% of all Google queries, climbing to roughly 82% specifically within B2B technology search queries. For a B2B SaaS company, that means the majority of your category’s Google searches now show an AI-generated summary before a single organic result, which makes citation inside that summary a bigger prize than the ranking underneath it.

Practically, this means two moves for AI Overviews specifically. First, participate credibly in the Reddit and YouTube ecosystem around your category, since those platforms carry disproportionate weight here that most B2B content strategies ignore entirely in favor of owned-channel publishing. Second, make sure your own pages expose a direct, quotable answer near the top of the content, with real numbers and clear schema markup attached, rather than requiring Google’s system to infer the answer from several paragraphs of surrounding context.

How Perplexity actually picks sources

Perplexity’s pattern looks closer to Google AI Overviews on the surface but with even more concentration. Reddit was Perplexity’s top cited source at 6.6% of total citations, nearly triple Google AI Overviews’ Reddit share, and within Perplexity’s top 10 sources, Reddit represents 46.7% of that volume, close to ChatGPT’s Wikipedia concentration but pointed at a completely different type of source.

The mechanism behind this is Perplexity’s Pro Search behavior, which is built to synthesize multiple angles of a topic rather than surface one authoritative answer. Forum-style discussion threads naturally contain that range of perspectives in one place, arguments, counterarguments, edge cases, and follow-up questions, all in a single thread. Content that acknowledges tradeoffs, alternative approaches, and edge cases, structured the way a well-moderated discussion thread would cover a topic, tends to get pulled into Perplexity’s synthesized answers more than single-angle brand content does, even when that brand content is factually accurate and well written.

Domain type matters here too, and it matters across all three platforms, not just Perplexity. Across the full Profound dataset, .com domains accounted for over 80% of total citations, with .org sites a distant second at 11.29%. That’s a reminder that hosting your GEO content on an unusual subdomain, a separate microsite, or an alternate extension is working against a real, measured preference in the data, not just a stylistic choice with no downside.

Where Gemini and Claude fit in

Public citation-pattern data on Gemini and Claude is thinner than what exists for ChatGPT, AI Overviews, and Perplexity, but brand-mention rate data gives a useful proxy. DerivateX’s 2026 benchmark study, which tracked 50 B2B SaaS companies across roughly 1,400 buyer-intent prompts, found mention rates of 100% for ChatGPT, 100% for Gemini, 90% for Perplexity, and 88% for Claude.

Mention rate isn’t the same thing as citation rate. A platform can name your brand in its answer without linking to your site as a source, which is a weaker but still meaningful form of visibility. The near-universal mention rates for ChatGPT and Gemini specifically suggest both platforms are more willing to surface brand names even from thinner source coverage than Perplexity or Claude are, which tend to hold back on naming a brand unless they can point to something backing it up.

Gemini’s deep integration with Google’s index means many of the same tactics that help with AI Overviews, direct answers, strong schema, credible ranking signals, carry over reasonably well to Gemini too. Claude’s more conservative mention rate in that same study lines up with a general pattern of favoring sourced, well-attributed claims over confident-sounding but thinly supported ones, which argues for the same reference-style rigor that helps with ChatGPT: cite your own numbers, attribute your own claims, and avoid vague superlatives with nothing behind them.

Common mistakes when GEO strategy ignores platform differences

A few patterns show up repeatedly in teams that built a single GEO checklist and applied it everywhere. The first is over-investing in schema markup and structured data as if it were a universal key, when the data shows it is really an AI Overviews and Gemini lever more than a ChatGPT or Perplexity one. Schema helps Google’s retrieval layer parse a direct answer; it does very little to change whether Wikipedia-style reference content outranks you inside ChatGPT.

The second is ignoring Reddit and forum participation entirely because it doesn’t look like a traditional B2B marketing channel. Given that Reddit accounts for 21.0% of Google AI Overviews’ top-10 citation volume and 46.7% of Perplexity’s, treating community platforms as beneath a B2B content strategy means opting out of the single largest citation source across two of the five major AI platforms.

The third mistake is measuring GEO performance with one blended visibility score. A brand that scores well in a combined metric might be entirely absent from Perplexity while overperforming on ChatGPT, and the blended number hides exactly the gap that needs fixing. Given the 11 to 12% overlap figures above, a single score genuinely obscures more than it reveals.

Building a platform-by-platform GEO playbook

The research points to a specific set of moves per platform rather than one universal tactic list.

For ChatGPT: write in a reference tone. Structure content the way a well-built glossary or encyclopedia entry would, definition first, then mechanics, then comparisons, then caveats. Keep terminology consistent across your site instead of varying it for SEO keyword coverage, since consistency is part of what makes a source feel authoritative to this platform’s retrieval logic.

For Google AI Overviews: put a direct, quotable answer in the first 40 to 60 words of every page you want cited. Back every specific number with a visible, named source. Apply Schema.org markup consistently across the site, and don’t assume organic ranking alone will earn the citation, since the Ahrefs overlap data says it usually won’t. Build a genuine, non-spammy presence in the Reddit and YouTube discussions that already exist around your product category, since those two platforms carry roughly 40% of this engine’s top-cited volume between them.

For Perplexity: write content that fairly represents multiple angles of a debate or decision rather than one company’s angle only. Comparison content, tradeoff analysis, and scenario-based “it depends” framing tend to fit this platform’s synthesis behavior better than confident single-answer pages built to close a sale. Where it’s genuinely useful, participate in the relevant discussion threads directly rather than only publishing owned content about the same topic.

For Gemini and Claude: treat them as a blend of the two approaches above. Direct answers and strong schema markup help with Gemini given its overlap with Google’s index. Rigorously sourced, attributed claims help with Claude given its more conservative mention behavior.

Across all of it, host your GEO-focused content on your primary .com domain rather than a disconnected subdomain, given how concentrated citation volume is on standard commercial domains. And measure each platform separately rather than as one number. A blended AI visibility score can hide the fact that you’re strong on one engine and completely absent on another, which is exactly the trap the 11% overlap statistic warns against. For a deeper foundational walkthrough of the discipline this playbook sits inside, see our full 2026 guide to generative engine optimization, and for the service side of implementing it, our AI SEO agency team builds these platform-specific programs directly.

How to track this instead of guessing

Most teams have no reliable way to answer “did we get cited by Perplexity this month” without manually running dozens of prompts by hand and logging the results, which does not scale past a handful of queries. Treat platform-specific citation tracking as a recurring discipline rather than a one-time check: run the same representative set of buyer-intent prompts against each platform on a fixed schedule, log which domains get cited alongside your own, and watch the trend line per platform rather than a single combined average. A single audit gives you the starting map; recurring tracking tells you whether the platform-specific tactics above are actually moving the number on the platform they’re aimed at.

It’s also worth tracking mention rate separately from citation rate, since the DerivateX benchmark data above shows those two numbers can diverge significantly by platform. A brand mentioned in 100% of ChatGPT responses but cited as a linked source in only a fraction of them still has real work to do, even though the headline mention number looks strong.

Where this data comes from

For teams that want to verify these numbers themselves rather than take them secondhand, here is the citation trail behind every figure used above.

None of these numbers stay fixed. Citation patterns shift as each platform updates its retrieval and ranking approach, so treat this as a snapshot from mid-2026 rather than a permanent map. The underlying principle is the more durable takeaway: these platforms are not interchangeable, they are not converging toward one shared standard, and a GEO strategy built around a single blended tactic list will keep leaving real citation volume on the table.

If you want to see where your own domain currently sits on each of these platforms individually rather than guessing from industry averages, our GEO audit runs that comparison directly against your site and your named competitors, platform by platform.

Alex Carter LinkedIn

SEO & Content Strategy, MV3 Marketing

Alex Carter leads SEO and content strategy at MV3 Marketing, specializing in generative engine optimization, technical SEO, and AI-driven content systems for B2B companies.