There is a traffic source growing faster than anything in your analytics, and there is a very good chance you have never opened the report.
AI referral traffic to United States retail sites grew 393% year over year in the first quarter of 2026. March alone was up 269%. It is one of the fastest growing commerce traffic sources on the internet.
Here is the frustrating part. Most of that traffic lands on a page designed for a completely different kind of visitor, gets confused, and leaves.
Which is an odd thing to be annoyed about, because these are the best visitors you will get all year.
Who these people actually are
Someone arriving from a Google ad is early. They clicked a promise. They are still deciding whether the problem is worth solving.
Someone arriving from an AI assistant has already had a conversation. They described their situation, got options, asked follow ups, narrowed the field and then chose to come to you specifically.
The numbers reflect it. AI referred visitors convert up to 42% better and spend around 45% longer on site than the average visitor.
They arrive knowing what you do, roughly what you charge and why they were sent. They are not browsing. They are verifying.
An AI referral is not a click. It is a warm introduction from a source your prospect already trusts. Most websites answer it with a hero banner and a slideshow.
Why they bounce anyway
The mismatch is simple. The assistant told them something specific. Your page opens with something general.
Say the assistant recommended you because you do same week commercial callouts across the Central Coast. Your visitor arrives looking for one thing, confirmation of that claim, and lands on a homepage that says "Excellence in service since 2011" above a photo of a handshake.
They now have to hunt for the fact that brought them. Roughly half will not bother, because the assistant offered them two other options in the same breath and one of those might confirm faster.
The failure is not traffic quality. It is a verification page that never got built.
Building the page that closes them
Three changes handle almost all of it.
Lead with the specific claim, in plain text, above the fold. Not a headline about your values. The actual fact. Coverage area, turnaround time, price range, guarantee, credential. Whatever an assistant would use to justify recommending you, put it in the first thing a human reads.
Answer the follow up questions on the page. The prospect has already asked the assistant three or four clarifying questions. They arrive with a fifth. Put your genuine frequently asked questions on the page as real answers, not an accordion of one liners written for search engines in 2018.
Make the next step tiny and obvious. These visitors are close. They do not need nurturing. They need a phone number, a form with three fields and a clear statement of what happens after they submit it.
How to actually see this traffic
This is where most businesses give up, because by default the report looks empty.
A large share of AI referrals arrive with no referrer at all and land in direct traffic. Some assistants pass a referrer, some strip it, and some route through link wrappers. So the fastest growing channel in your business hides inside the least informative bucket in your analytics.
Three fixes, in order of effort.
- Filter your referrals for assistant hostnames. Look for chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com and openai.com in your referral report. Whatever shows up there is the visible tip of a much bigger number.
- Watch direct traffic behaviour, not volume. Segment direct visitors who land on a deep page rather than the homepage. Nobody types a service page URL from memory. That segment is almost entirely assistant referrals and dark social.
- Ask. Add one optional field to your enquiry form. "How did you hear about us." The number of people typing some version of "ChatGPT said" into that box will tell you more than any attribution tool you can buy.
The compounding part
Here is what makes this worth acting on quickly rather than eventually.
Assistants are conservative recommenders. Once a model has a confident, well supported answer about who to hire in your category, it repeats that answer. It does not go looking for challengers every week.
So the businesses that get named now, and then confirm the recommendation with a page that closes cleanly, get named again. And the loop tightens, because satisfied buyers write reviews, which become more source material, which reinforces the recommendation.
The businesses that are ambiguous today do not get a second look next quarter. They get skipped by default.
What these visitors do differently on site
Watch a few session recordings from this segment and the pattern is unmistakable.
They scan rather than read. They ignore your navigation almost entirely. They hunt for one or two specific facts, usually price, coverage or timeframe, and they either find them in the first fifteen seconds or leave.
They also skip your carefully sequenced journey. The story you built across four scroll sections is irrelevant to someone who already had the story explained to them by a machine.
Design for scanning. Facts near the top, in plain text, in the order an assistant would have mentioned them. Save the persuasion for people who arrived without a briefing.
What to do this week
Open your analytics. Find every session from an assistant hostname, and every direct session that landed on a deep page. Look at what those visitors did.
Then take your top three landing pages and rewrite the first fifty words of each so they confirm, in plain language, the specific fact that would have caused an assistant to send someone there.
That is an afternoon of work against the fastest growing traffic source in your category, arriving pre sold, converting 42% better, and currently bouncing off a slideshow.
There are not many fixes left in marketing with that ratio. Take this one.