Search Beyond Google: Where Perplexity, ChatGPT, and Bing Traffic Actually Lands

The interesting question about AI search traffic is no longer how much of it there is. It is which page it arrives on — and on that measure it behaves nothing like the traffic you are used to.

Most coverage of this topic is a list of engines. Here are the twelve AI search tools, here is what each one does, optimise for all of them. That list is easy to write and almost useless, because it tells you nothing about volume, nothing about which of those tools actually sends anyone to your site, and nothing about what those visitors do when they arrive.

So let's do it the other way round. Start with the volume, narrow to the platforms that measurably matter, and then spend most of the time on the finding that changes what you'd actually do — where these visits land.

First, the number that puts everything else in proportion

SE Ranking's analysis of aggregated Google Analytics data from more than 101,000 websites across 250 countries put AI referral traffic at roughly 0.32% of all website traffic in 2026. Organic search, over the same period, accounted for 42.75%.

That is a ratio of about 134 to 1. Roughly one visit in every 312 arrives from an AI search engine.

If you have read a dozen posts implying that AI search has displaced Google, that number should be jarring. It should also be held alongside the other half of the finding: that same figure was 0.02% in 2024 and 0.24% in 2025. A sixteenfold increase in under two years is not nothing, and the ratio against organic has compressed from roughly 300 to 1 at the start of 2025 to 134 to 1 now.

Both things are true. The channel is small and the channel is compounding. Which of those you emphasise should depend on your planning horizon, not on which one makes a better headline. We made the same argument from the Google side in our mid-year state of SEO reality check.

The list of engines worth tracking got shorter, not longer

This is where most roundups mislead. They lengthen the list every quarter as new tools launch. The referral data has moved in the opposite direction.

Nine platforms produced measurable referral traffic in 2025. By 2026 only these five stayed consistently above a tenth of a percent of AI referrals. DeepSeek held about 0.37% of AI traffic in early 2025 and then fell to essentially zero from September onward. Grok spiked briefly around a product launch in mid-2025 and did the same.

The practical read: usage experiments and referral traffic are different things. A tool can have real users and send you nobody, because whether it links out at all is a product decision, not a popularity contest. Track five platforms properly rather than fifteen badly.

Featured Recommendation AD · AFFILIATE

4.8 / 5.0

Rankmath

Most popular WordPress SEO plugin with powerful on-page optimization features built in

Best for: WordPress SEO Plugin

Visit Website

View details Compare

The correction most articles haven't made yet: Perplexity

Nearly every listicle still ranks Perplexity as the number two AI search engine. For referral traffic, that stopped being true at the start of this year.

Gemini overtook Perplexity globally, in the US and in the UK in January 2026, and in the EU in February. And Perplexity has not simply been outgrown — its absolute traffic softened slightly while the overall AI referral pool grew by around 36%.

The sharpest fall is in what used to be its strongest market. Perplexity held 19.73% of US AI referral traffic in early 2025, 11.42% across the full year, and 6.85% in 2026. In the UK its share went from 5.64% to 3.06%. The EU is the one place it still runs close to Gemini, at roughly 9.02% against 9.8%.

None of that means Perplexity is irrelevant. It means that if you are allocating limited attention across AI platforms, the ranking you inherited from a 2025 article will point you at the wrong one.

Where the traffic actually lands — and why it matters more than the volume

Here is the finding that should change something in your week.

Around 60% of AI-referred visits land on the homepage, compared with roughly 17% from organic search. That is close to four times the homepage concentration, and it is a structural difference rather than a quirk.

The difference in one line Search sends people to a page. AI sends people to a company. Those two visitors need completely different things from the screen they land on.

It follows from what people ask assistants. Nobody asks ChatGPT for a specific blog post. They ask for tool recommendations, comparisons, alternatives to something they already use — and the natural destination for an answer like that is the brand's front door, not its best-optimised article.

Which creates a mismatch. Most homepages are built for people who already know who you are and arrived deliberately. Someone landing from an AI recommendation is often hearing your name for the first time, in a list alongside two or three competitors, and has clicked to find out whether the recommendation was any good.

That visitor needs your positioning stated plainly in the first screen, quick proof you do what was claimed, an honest comparison against the alternatives they were just shown, and an obvious next step that isn't a demo request. Most homepages provide none of those, because they were never designed for a cold, comparison-stage arrival. If yours is due a rethink, the principles in our practical guide to landing page design transfer directly, and the exercise in positioning your brand as the obvious choice is what the first screen has to communicate. If the rebuild is bigger than a copy change, that is a website design and development job rather than a content one.

There is a broader point here about how sites are structured for machine readers as well as human ones, which we covered in designing websites for AI search and in the push to be cited by machines. But the homepage finding is the one that needs no new tooling to act on.

What changed in May 2026

If your ChatGPT referrals jumped this spring and you never worked out why, this is the explanation.

ChatGPT's worldwide referral share rose from 0.23% in April to 0.32% in May — a 36.7% increase in a single month, and about 8% above its previous peak from October 2025. For context, the whole of January to April had produced roughly 1.5% growth.

The jump appeared in every region tracked, with the EU up 42.7%, the UK up 38.7% and the US up 23.2%. That synchronisation matters: those three regions had been on different trajectories beforehand, so a simultaneous turn points at a platform change rather than market noise.

The most credible explanation is a link-format change around May 7 that made brand names clickable directly inside answers. Before, being mentioned created awareness and required the reader to go and search for you separately. After, the mention became a click. Two other things landed in the same window — GPT-5.5 Instant became the default model on May 5, and OpenAI opened up self-serve advertising tools — though the researchers describe the connection as a working hypothesis, not a proven cause.

The strategic implication is worth sitting with. Being named in an AI answer used to be a brand-awareness event you couldn't measure. It is now, at least partly, an acquisition event you can. That shifts generative engine optimisation from a speculative investment toward one with a visible return line — and it also explains why the gains went disproportionately to homepages.

Bing and Copilot are two separate questions

Worth untangling, because they get merged constantly.

Copilot as an AI referral source is small but real: 3.51% of AI referrals globally, up about 31% year on year, and notably stronger in the US at 5.12%. It first showed up as a measurable referrer in mid-2025 and has been stable rather than explosive since. It is the fourth platform on the list, not a rounding error, but it will not move your numbers on its own.

Bing as a traditional search engine is a different matter entirely, and it sits inside that 42.75% organic figure rather than the 0.32% AI one. If Bing is a meaningful share of your organic traffic, that is a conventional SEO question. Research into how these engines pick sources has found Copilot's citation behaviour to be the least like everyone else's — it cites the fewest sources per answer, overlaps least with the other tools, and is markedly more willing to cite domains under five years old than Google's AI Overviews are. For a newer site, that makes it the friendliest of the four to get cited by, even if the traffic that follows is modest.

Lean traffic, but a different kind of visitor

The volume argument cuts one way and the behaviour argument cuts the other.

Visitors arriving from AI platforms spent about 68% longer on site than those from organic search — averages of roughly nine minutes nineteen seconds against five minutes thirty-three. Medians, which strip out the outliers, tell the same story more modestly: two minutes twenty-four against one minute fifty-three.

The likely reason is that the filtering happened before the click. Someone who has spent three exchanges narrowing their options with an assistant arrives already qualified in a way that a first-time search visitor usually isn't. Fewer visitors, further along.

Two cautions on reading too much into this. The regional spread is wide — the US advantage was only about 29%, the EU nearly double — and averages on the smaller platforms are badly skewed by tiny samples with a handful of very long sessions. Treat the direction as reliable and any specific figure as indicative.

What to actually do about it

In rough order of return on effort.

• Fix the homepage first. It is where most of this traffic lands and where the mismatch is worst. Nothing else on this list beats it.
• Separate AI referrals in your analytics. If these visits are sitting in an undifferentiated referral bucket you cannot see any of this in your own data. Track the five platforms individually — the shares differ enough that an aggregate number hides what's actually moving.
• Track mentions, not just clicks. Being cited without being clicked still happens constantly, and it is the leading indicator for the clicks. This is a variant of the measurement problem we've written about in the zero-click rethink and in why attribution keeps getting harder.
• Watch the surfaces that feed the answers. Community and video sources carry disproportionate weight in what these systems cite, which is the subject of where Google is sending clicks now.
• Don't rebuild your SEO programme around 0.32%. The work that gets you cited by assistants overlaps heavily with the work that gets you ranked, as what's working in AI Overviews and our look at AI Mode both found. Treat it as an additional surface for existing work, not a separate programme.

One more thing worth saying plainly: this data moves fast enough that a single monthly reading is close to meaningless. May's jump would have looked like an anomaly in isolation and only became a story because sixteen months of context sat behind it. If you are going to report this internally, report the trend line — the same argument we made about dashboards leadership will actually read.

The short version

AI search sends roughly 0.32% of web traffic against organic's 42.75%, but that gap has closed from 300 to 1 down to 134 to 1 in eighteen months. Five platforms account for essentially all of it, and the list is consolidating rather than expanding. Gemini has passed Perplexity almost everywhere. About 60% of these visits land on your homepage against 17% from search, which makes your homepage — not your content — the highest-leverage thing to fix. And the visitors who arrive are fewer, but they stay considerably longer.

The volume doesn't justify a strategy yet. The landing page data justifies an afternoon.