By Rahul Saini, Author at Search Counsel Co. Last updated August 2026.

1) The big picture: the engines don’t agree

Most people picture “AI search” as one thing. It isn’t. An analysis of 680 million citations found that only about 11% of domains are cited by both ChatGPT and Perplexity, and even Google’s own AI Overviews and AI Mode share the same cited URLs only around 14% of the time. A page that wins in one engine can be invisible in another. If you’re new to the mechanics behind this, start with how AI search works.

There is a shared foundation. Every engine needs your content to be crawlable, clearly structured, credible, and reasonably fresh. But on top of that, each one has its own retrieval method, its own favorite sources, and its own citation habits. One more thing worth knowing up front: these preferences shift fast. ChatGPT’s share of Reddit citations reportedly fell from roughly 60% to 10% in about six weeks after a single Google parameter change. Optimize for the durable signals, not this month’s quirk.

2) How ChatGPT chooses sources

ChatGPT starts with its training data, a large body of text frozen at a cutoff date. It only runs a live web search when you ask it to or when it senses it needs recent information. That’s why ChatGPT sometimes answers with no citations at all: if it didn’t retrieve, there’s nothing to cite.

When it does search, it uses a “query fan-out.” A single prompt triggers several sub-queries, reportedly 5 to 15, sent to Bing’s index, and ChatGPT then pulls the passages that best answer each one. It’s selective, citing fewer but more authoritative sources, and it heavily favors pages that put the key facts up top, often in the first few hundred words. Its most-cited source is Wikipedia by a wide margin, with Reddit, major publishers, and third-party review sites close behind.

3) How Perplexity chooses sources

Perplexity works the opposite way from ChatGPT. It runs a real-time web search on every query and never answers “from memory,” and it always attaches clickable citations. It’s the most citation-dense of the big engines, often showing 20 or more sources in a single answer.

Two things stand out about what Perplexity favors. First, recency: its retrieval leans hard on recently published or updated content, more than the other engines, which makes a disciplined content refresh cycle unusually valuable here. Second, community sources: Reddit alone accounts for close to half of Perplexity’s top citations, well ahead of Wikipedia, because it treats community-validated, real-world experience as a strong trust signal. Perplexity also runs a Publisher Program that shares revenue with cited publishers, the only major engine that pays for citations.

4) How Google AI Overviews and Gemini choose sources

Google’s AI Overviews are powered by Gemini and draw from Google’s existing organic search index through grounded retrieval. For years the shortcut advice was “rank in the top 10 and you’ll get cited,” and there’s truth to it: in mid-2025, around 76% of AI Overview citations also ranked in Google’s top 10 for that query.

That link is weakening, though. By early 2026, separate analyses put the overlap far lower, roughly 38% in one study and around 17% in another. Strong organic ranking still helps, but it’s becoming a weaker guarantee, not a stronger one. It’s also worth separating two Google surfaces: AI Overviews and the newer AI Mode cite different pools. AI Mode pulls from a wider range, including more community content like Quora, and the two share only a small fraction of the exact URLs they cite even when their answers agree. Video matters here too, with YouTube heavily favored in Google’s AI results. To watch this surface in your own reporting, see tracking SERP features and AI Overviews.

5) How Claude chooses sources

Claude uses its training data and adds web search when the feature is enabled, citing sources when it retrieves them. Its clearest tendency is a preference for depth and structure. Analyses suggest Claude is noticeably more likely to cite pages that use clear headings and bullet-pointed, well-organized content it can parse cleanly. If your page is a dense wall of text, Claude is the engine most likely to skip it.

6) The mention-source divide (the part most people miss)

Here’s a pattern that surprises brands. When an AI engine recommends you, it often doesn’t cite your website. It cites a third party talking about you. Your competitor gets the recommendation, and a Reddit thread, a G2 review page, or an industry roundup gets the citation link.

This is because AI engines read agreement across independent sources as a trust signal, the effect we cover in multi-source consensus. Your own site saying you’re the best is one data point. A dozen forum posts, review pages, and articles saying it is a consensus. For B2B especially, the sources engines cite most often are places like Reddit, review platforms, and analyst sites, not company homepages, which is why brand mentions matter differently from backlinks here.

7) How to get cited across all four

You can’t run four separate strategies forever, so start with the shared foundation, then add platform-specific moves.

The shared foundation, which helps everywhere:

• Be reachable. Let AI crawlers access your site, keep pages indexable, and don’t hide content behind heavy JavaScript. An engine can’t cite what it can’t read.
• Front-load answers. Lead each section with the claim, in a self-contained sentence a model can lift without the surrounding paragraphs.
• Add evidence. Specific facts, numbers, and cited sources beat vague, hedged prose. Engines reward evidence density.
• Stay fresh. Keep a visible last-updated date and refresh key facts on a schedule. Perplexity and ChatGPT both weight recency.
• Build topical depth. Publish deeply on one topic rather than thinly across ten. Engines build source-authority around focused expertise, so topical authority compounds here.

Then tune by platform: structure cleanly with headings and lists for Claude, keep content genuinely current for Perplexity, earn and maintain strong organic rankings for Google’s Overviews, and put your key facts in the first few hundred words for ChatGPT. And across all of them, invest in third-party mentions, because that’s where the citations often live. To see which engines already name you, start with tracking your AI visibility.

Sources used for this guide

FAQ: how AI engines choose sources

Conclusion: optimize for the engine, not just “AI”

Treating AI search as one channel is like running the same campaign on every social network and expecting it to land. ChatGPT, Perplexity, Gemini, and Claude each find and cite sources differently, and the pages that win across all of them share a foundation: reachable, clearly structured, evidence-backed, current content, plus a presence on the third-party sources these engines trust.

Get the foundation right first, then tune per platform. For the groundwork, start with what GEO is and how it works, and for the strategy behind these tactics, see the SEO, GEO, and AEO taxonomy and our full AI search optimization guide.