ChatGPT vs Gemini vs Perplexity: Where AI Referral Traffic Is Going

AI referral traffic is heavily concentrated in ChatGPT today, but the field is moving fast: Gemini just overtook Perplexity for second place, and the whole pie is growing. So the right move isn’t picking a winner — it’s tracking and optimizing across every engine your buyers use.

The current shares, in one table

StatCounter’s March 2026 press release put the split of AI-chatbot referrals to websites like this:

Engine	Referral share (StatCounter, March 2026)	Notable move
ChatGPT	78.16%	Still dominant, but no longer the whole story
Gemini	8.65%	Quadrupled from 2.31% (April 2025); now #2
Perplexity	7.07%	Down from a 12.07% peak (April 2025), a fall of more than 40%

The headline is Gemini’s climb. Per StatCounter (March 2026 data), Gemini jumped from 4.74% in December 2025 to 7.20% in January 2026, overtaking Perplexity as the second-largest source of AI-chatbot referrals for the first time. Perplexity, meanwhile, has slid more than 40% off its April 2025 peak.

The nuance that changes how you read this

Here is the single most important caveat: these are referral shares, not volumes.

The overall AI-referral pie is growing. That means an engine’s share can fall while the actual number of visits it sends you keeps rising — the pie is simply expanding faster than that engine’s slice. Perplexity losing share does not automatically mean Perplexity is sending you fewer visitors. It might be sending more, just a smaller cut of a much bigger total.

So treat StatCounter’s percentages as a map of where attention is concentrating, not as a verdict on any one engine’s traffic. To know what’s actually happening to your site, you have to look at your own analytics — absolute sessions per engine, over time — not just the market-wide share chart.

B2B looks different from the whole market

Market-wide averages also hide segment-level shifts. In Goodie’s B2B panel, ChatGPT’s share of AI referral traffic fell from 89.1% (May–August 2025) to 62.6% (March–April 2026) — a much faster erosion of dominance than the whole-market StatCounter figure shows.

Read that carefully, though. That’s a single GEO vendor’s brand-averaged panel of B2B properties in Goodie’s B2B panel, not a whole-market measurement. It’s a useful signal that concentration can look very different inside a specific segment — but it’s one panel, not the market. The lesson isn’t “the numbers disagree”; it’s that your mix depends on your audience, which is exactly why you measure your own.

The crawl-to-referral gap

There’s a second reason not to over-index on referral counts alone: engines read far more than they send back.

Cloudflare’s June 2025 network snapshot measured crawl-to-referral ratios of roughly:

• 14:1 for Google — crawls about 14 pages for every referral it sends
• 1,700:1 for OpenAI
• 73,000:1 for Anthropic

Present these as a June 2025 snapshot; the ratios have tightened since. But the shape of the problem is durable: an engine can be ingesting your content heavily and still send comparatively little click-through. That’s why “referral traffic” is a lagging, partial view of AI visibility — being cited and mentioned inside an answer matters even when the user never clicks. Referrals count only the clicks, not the influence.

What to actually do about it

The strategic conclusion writes itself: don’t build your AI-visibility strategy around one engine, and don’t confuse a share chart for your own results.

1. Instrument every engine separately. Segment your analytics so ChatGPT, Gemini, Perplexity, and Google AI surfaces each get their own line. Watch absolute volume and share side by side — you need both to interpret either.
2. Track presence, not just clicks. Because of the crawl-to-referral gap, referral traffic understates AI influence. Pair it with citation and mention tracking so you see where you’re being used in answers, not only where you’re being clicked. Our roundup of AI citation tracking tools covers the options.
3. Run the same buyer prompts across engines on a cadence. Use one prompt set and one scorecard per engine; our guide to tracking your brand in Perplexity, Gemini, and AI Overviews walks through the method engine by engine.
4. Roll it up into one number you can trend. Individual engine stats are noisy; your share of model aggregates them into a directional metric you can actually manage against.
5. Diversify your optimization, not just your measurement. Each engine sources answers differently, so a page tuned only for one surface leaves the others on the table. For the wider context on how the ground shifted this year, see what actually changed in AI SEO in 2026.

The takeaway

ChatGPT leads AI referrals by a wide margin, Gemini has overtaken Perplexity for second, and the whole pie is expanding — which means falling share and rising volume can happen at the same time. The engines will keep trading places; model versions, product launches, and search integrations move these numbers quarter to quarter. Betting on one is a bet you don’t need to make. Measure all of them, in your own analytics, and optimize for the surfaces your buyers actually use.

Want to know exactly where your brand stands across ChatGPT, Gemini, Perplexity, and Google’s AI surfaces — in real numbers, not market averages? Get an AI visibility audit and we’ll map your presence engine by engine and show you where to win more of it.