How to convert AI traffic: capture shoppers arriving from ChatGPT and Gemini

What AI-referred traffic actually looks like

When a shopper asks ChatGPT for the best lightweight running shoe for wide feet under $120 and clicks the store it recommends, they arrive with the research already done. They described their need in their own words, compared the options inside the conversation, and clicked through to verify a decision, not to start one. That is a different visitor from the one your store was designed for: the Google-era browser who lands on a category page and works their way in.

The behavior shows up in the aggregate numbers. Adobe's retail panel has AI-referred visitors converting better, staying longer, and reading more pages than any other referral class it tracks:

One caution against overexcitement: the assistant that sent the shopper stays in the loop. The person keeps deciding; the AI does the legwork of searching, comparing, and answering questions along the way. Converting this traffic means serving that pattern, a well-informed human mid-decision, not building for some imagined robot buyer.

Find it in your analytics first

Before optimizing anything, make the traffic visible. In GA4 (or any analytics tool that records referrers), build one segment that collects sessions whose source matches the AI assistant domains. The list that covers the traffic that matters in 2026:

• chatgpt.com (and the older chat.openai.com) for ChatGPT. Links clicked inside ChatGPT answers also commonly carry a utm_source=openai parameter, which gives you a second, independent way to catch them.
• gemini.google.com for the Gemini app.
• perplexity.ai for Perplexity.
• copilot.microsoft.com for Microsoft Copilot.
• claude.ai for Claude.

Two honest caveats about what this segment can and cannot see. First, clicks from Google's AI Mode and AI Overviews arrive with a regular google.com referrer, so they blend into your organic search bucket; the gemini.google.com slice is only the separable part of Google's AI traffic. Second, referral tracking only catches the direct click. A shopper who gets your brand name from ChatGPT and then Googles you shows up as organic or direct, so whatever your segment reports is a floor, not the full contribution. Which engines send how much, and where each one gets its answers, is covered in AI traffic sources.

Expect the absolute numbers to look small at first. That is normal, and it is not a reason to ignore the channel:

A small share growing triple digits with the highest conversion rate of any referral class is exactly the profile a channel has right before everyone starts paying for it. Right now it is still free.

Why the click dies on a static landing page

Here is the frustrating part: stores are already winning these clicks and losing them on arrival. The shopper was mid-conversation. They asked something specific, and the assistant sent them to your site to finish the job. The page they land on was built for a different entry point: a hero banner, a size chart, an add-to-cart button, and none of the context of the question they just asked.

In the scans we run, the failed handoff takes two recognizable shapes. Sometimes the AI cites the wrong page outright: a years-old blog post, a B2B portal, a discontinued product. At least as often it cites a perfectly reasonable page that simply cannot answer the shopper's actual question: whether the fabric runs sheer, whether the part fits a 2019 model, whether the shade works for cool undertones. The visitor bounces not because your product was wrong but because the page could not hold up its end of a conversation the shopper thought they were still in.

That gap is the whole economics of this channel. Being recommended gets you the click; what the visitor finds on arrival decides whether the 54% conversion advantage lands in your revenue or evaporates. This is the handoff problem that sits at the center of agentic commerce: the conversation that understood the shopper stays behind in the chat window, and your store inherits a visitor whose question it never heard.

Capture it without rebuilding your store

Start with the wrong answers, because they are the expensive ones. Re-platforming to a headless stack is months of engineering and six-figure budgets to solve a problem that lives at the landing-page layer. Waiting for your ecommerce platform to ship a native fix means accepting their timeline and their lock-in. And bolting a chat widget onto the existing site solves a different problem: a widget engages people already browsing, but it does not change what the AI engines can read about your store, and it is not built around the question the AI-referred visitor arrived with.

What actually moves the number is making the arrival answerable, in escalating order of effort:

• Let the AI crawlers in and be readable. Most AI crawlers do not execute JavaScript, so your product content has to exist as static HTML, and the bots have to be allowed: the ChatGPT shopping user agent guide lists every crawler and what blocking each one costs.
• Answer the questions on the pages that get cited. Materials, fit, compatibility, care, returns: written in plain prose on the product and guide pages themselves, not buried in a PDF or a reviews widget. This is the cheapest fix and it compounds, because answerable pages also earn more citations in the first place.
• Keep your structured data honest. Product and Offer schema with real-time price and availability, so what the assistant tells the shopper matches what the landing page says when they arrive. A mismatch kills the trust the conversation built.
• Give the arriving shopper a way to continue the conversation. This is the step a static page cannot do alone: an agent-ready layer on your own domain that can answer the specific question the visitor came with, in the moment they arrive, the way an AI storefront is designed to.

Full disclosure: that last step is the work Aigency does. We host a parallel, agent-readable version of your store on your own subdomain, so AI-referred visitors land on a page that holds the conversation the assistant started, and checkout stays on your existing site, untouched. It deploys in days, not a re-platform. If you would rather build in-house, the checklist above is the complete map, and every step of it pays for itself independently.

Treat it like a channel, not a novelty

The mistake merchants made with every previous channel shift was treating the early trickle as noise until it was a flood with established winners. The playbook that compounds is boring and weekly: watch the AI segment's sessions and conversion rate against site average, note which products and questions the assistants send people to, and fix the worst landing mismatch each week. The AI segment converting below your site average is the clearest possible signal, because this traffic should convert above it.

• If the segment is small but converting above average: your problem is visibility, not conversion. Work on getting cited more, which is the GEO vs SEO discipline.
• If the segment is growing but converting at or below average: you are winning clicks and losing the handoff, which is exactly the landing-page problem this guide exists for.
• If you cannot tell: build the segment from the analytics section above before spending anything else on this channel.

The merchants who figure out the AI-referral handoff in 2026 get a compounding advantage: better conversion feeds better signals back to the engines, which cite them more, which sends more of the best-converting traffic on the web. It is easier to own a channel while your category still treats it as a curiosity.