Claude just passed Perplexity in AI referral traffic

The pecking order of AI referral traffic is not settled - it is moving. In one closely-watched snapshot, Claude pulled ahead of Perplexity, and Gemini sits far higher than most people assume. The numbers matter less than the lesson: do not pin your strategy to one engine or one month.

It is tempting to treat the AI-assistant market as ChatGPT and everyone else. The referral-traffic data says the "everyone else" is both bigger and more fluid than that framing allows - and that has a practical consequence.

The snapshot

According to Similarweb data (all-device, worldwide, through 26 May 2026), the shares of AI referral traffic broke down roughly as: ChatGPT 52.7%, Gemini 27.3%, Claude 8.9%, Perplexity 1.3%. Two things stand out. First, Gemini is a strong second at over a quarter - far more than its share of conversation would suggest. Second, Claude is now ahead of Perplexity in this data, a reshuffle from where many assumed the order sat.

Curious how AI engines describe your brand right now? Get a free visibility audit and see where you stand across ChatGPT, Gemini and Perplexity.

The honest caveat: studies disagree

Here is the part most write-ups skip. These numbers vary by study. A different analysis, using a different method and window, shows ChatGPT far higher (in the 70s), Gemini and Perplexity lower, and Perplexity still ahead of Claude. So "Claude passed Perplexity" is true in the Similarweb snapshot and not true in another. That is not a contradiction to resolve - it is the nature of measuring a fast-moving market with different panels.

The takeaway from the disagreement is itself useful: never quote an AI-share number without its source and date, and never build a strategy on a single chart.

"Claude passed Perplexity in one study and not in another. That is not a contradiction - it is a warning."

Why Claude is climbing

Claude's referral growth is interesting because it is not really a search story. A large part of it comes from Claude's coding and work-agent products - people using Claude to do research, write documents, build things. Those working sessions generate referral traffic even though Claude is not positioned as a search engine. It is a sign of where AI referrals increasingly come from: assistants doing tasks, not just answering questions.

What to do

1. Do not optimise for one engine. The order changes, and the studies disagree about the order. A ChatGPT-only strategy is fragile; so is a strategy built on any single month's leaderboard.
2. Respect Gemini. A quarter-plus of AI referral traffic in this snapshot is not a rounding error. If you have been treating Gemini as an afterthought, the data says reconsider.
3. Measure your own mix. Market averages across many sites tell you nothing about your buyers. Track where your referrals and citations actually come from.
4. Watch the work-agents. As assistants do more tasks, referral traffic will come from flows you are not thinking of as "search." Keep an eye on which ones send you visitors.

The takeaway

The leaderboard reshuffled, Gemini is bigger than its reputation, and the studies cannot even agree on the order below ChatGPT. All of that points to one discipline: stop chasing whoever is winning this month, and measure your own visibility across every engine your buyers use. The market is too fluid for shortcuts.

How to measure your own mix

If the market average is the wrong number to plan around, the right one is your own. The good news is that most of it already sits in tools you have. Open your analytics and look at referral sources: the AI assistants show up as their own domains, so you can see which of them actually send you clicks and how that trend moves month on month. Segment by landing page too, because the pages that get cited in an answer are rarely your homepage - they are the specific, well-structured pieces that resolve a specific question.

Referral clicks are only half the story, though. A lot of AI visibility never shows up as a visit at all. Someone asks Claude or Gemini a question, your brand gets named in the answer, and the user acts on it without ever clicking through. That is influence you cannot see in a referral report. To catch it, you have to ask the engines directly - run the questions your buyers ask, across each assistant, and record whether you are named, how you are described, and who is named alongside you.

• Referral share, per engine. What fraction of your AI referrals comes from ChatGPT, Gemini, Claude, Perplexity and Google's AI surfaces - not the market's, yours.
• Citation presence. For the questions that matter to your category, are you named in the answer at all, and are you named first or fifth?
• Share of voice against rivals. When you are absent, who is present in your place? That gap is the most actionable number you will find.

What this does not mean

A reshuffled leaderboard invites two overreactions, and both are worth naming. The first is to conclude that Perplexity no longer matters. A small slice of aggregate referral traffic does not mean a small slice of your buyers. Perplexity skews towards research-heavy, high-intent questions, and for some categories the people asking those questions are exactly the ones you want. A market average flattens that away. If your own data shows Perplexity sending qualified visitors, the aggregate chart is simply the wrong lens for you.

The second overreaction is to treat Claude's rise as a signal to chase Claude. It is not. The lesson of the snapshot is that the order below ChatGPT is unstable and measured differently by everyone, so betting on this month's number two is the same mistake as betting on last month's. The durable move is not to pick a winner - it is to be present, accurately described, and citable across all of them, so that whichever engine your buyer happens to open, you are already in the answer.

"Being named in every engine beats winning any one of them."

A concrete example

Picture a mid-market analytics vendor whose team obsessed over ChatGPT because it was the biggest single source. They tuned their pages, they got cited there, and they stopped looking. When they finally ran the same buyer questions through Gemini and Claude, the picture changed. In Gemini, a competitor was named first and they were absent entirely. In Claude, they surfaced only inside coding-adjacent answers, because Claude's work-agent flows were the ones pulling their documentation - a channel they had never thought of as marketing.

Nothing about their product had changed. Their visibility simply lived in places their ChatGPT-only view could not see. Once they measured all four engines, the fix was obvious: strengthen the pages Gemini kept skipping, and make the technical docs Claude was already surfacing clearer and more quotable. The point is not that any one engine is the answer. It is that you cannot fix a gap you have never looked at, and a market-average chart will never show you your own.