Is AI search sending you customers? How to tell, and where analytics gets it wrong
Leadership teams now ask whether AI is sending them customers. The data to answer it exists, but it is split across three places and misfiled in predictable ways.
The question I hear most from leadership teams this year is not "should we be doing AI search?" It is "is AI actually sending us customers, or is this another number on a dashboard?" It is a fair question, and for most businesses the honest answer is that their current reports cannot tell them.
That is not because AI traffic is invisible. It is because it arrives through three different doors, and each one is counted somewhere different.
Three doors, three places to look
1. AI assistants that link to you. When someone clicks a link inside ChatGPT, Gemini, Copilot or a similar assistant, the visit usually arrives with a referrer. Since May 2026, Google Analytics has grouped these visits into a new AI Assistant channel in its default channel reports. The channel definition sets the medium to "ai-assistant" when the referrer matches Google's list of assistants. No setup is needed.
2. Google's own AI features. AI Overviews and AI Mode are not in that channel. Google Analytics counts them as Organic Search, and Search Console counts them inside the normal Performance report under the Web search type, as Google's guidance on AI features explains. In June 2026 Search Console added generative AI performance reports that show how often your pages appear in AI Overviews, AI Mode and Discover's AI features. They report impressions, not clicks, and Google says they have been available to all sites since 31 August 2026.
3. Answers that never send a visit. An assistant recommends you by name, the buyer searches for you a day later or types in your address, and the sale lands as branded search or Direct. No report will ever credit that one to AI.
So the first thing I tell people is that there is no single "AI number". There is an AI Assistant channel that is reliable but incomplete, an organic number that now quietly includes Google's AI features, and an influence you can only measure indirectly.
Where analytics quietly gets it wrong
Even the doors you can see get misfiled. These are the three problems I find most often when I first open a business's analytics.
Visits with no referrer become Direct. The AI Assistant channel only works when the referrer is passed. Links opened from inside some apps, copied into a browser, or sent from a chat on one device and opened on another arrive with no referrer and land in Direct. If your Direct traffic has grown while interest in AI has grown, some of it is probably AI.
Broken sessions become Unassigned. Anything on the site that reloads the page or restarts the session can wipe out where the visitor came from. On one online store I work with, a bundling app reloaded every new visitor's first page. The reload made the site look like its own referrer, so about a third of visits from Google search were landing in Unassigned, where no channel got the credit. Visits from Bing were untouched, which is what gave it away. A few lines of code stopped the reload, and Unassigned sessions fell by about three quarters within a fortnight.
Some orders never reach analytics at all. Express checkouts, orders placed through shopping apps, ad blockers and consent choices all mean some sales have no analytics record. On that same store, around 12% of orders were missing from analytics entirely. Once the tracking was fixed and purchases were also sent from the server, it recorded about 97%.
None of this is specific to AI. But AI traffic is small and new, so it is hit hardest. A few hundred misfiled sessions can halve or double the figure that goes to the board.
The most common mistake I find is not a broken tag, though. It is that nobody has ever lined analytics up against the order book. On that store, analytics revenue looked close enough to reported sales that no one had questioned it. It only looked close because the analytics figure included GST and the sales figure did not. With the tax taken back out, around one sale in six was missing. A healthy-looking total was hiding the problem.
How to sanity-check your number
You do not need a new tool. You need to check the number against something that cannot be wrong.
- Start from sales, not sessions. Take a month of orders and revenue from your store or CRM and compare them with what analytics recorded. If analytics captures less than about nine in ten orders, fix that before you read any channel split. Every channel figure is only as good as that match rate.
- Check the AI Assistant channel picks up what you expect. Look at the sources inside it. If an assistant you know sends you visitors is missing, a custom channel group with your own list of sources will catch it.
- Watch Direct and Unassigned as a share of sales. A sudden jump usually means a tracking break, not a change in your customers.
- Judge AI visits on outcomes, not volume. Compare conversion rate and order value with organic search. In the ecommerce programme I mentioned in my piece on AEO, GEO and AIO, AI visitors have converted at around one and a half times the rate of Google visitors. Small numbers, but the right kind.
- Use Search Console's AI report for visibility, not revenue. Impressions in AI Overviews and AI Mode show whether Google is using your pages. They do not show whether that made you money. The organic revenue line does, because it already includes those visits.
- Measure mentions separately. For the third door, run a fixed set of real buyer questions across the main assistants every month and record whether you are named, whether you are cited with a link and whether you are described accurately. It is the only way to see influence that never clicks.
How I report it to a leadership team
I report AI as part of one line: organic search plus AI, measured against actual sales and compared with the same period last year. Inside that line I show the AI Assistant channel on its own, because leaders want to see it, and I label it as a floor rather than a total. Then I add a short note on what the numbers do not capture: answers that recommend the brand without a click, and orders analytics never saw.
That framing heads off two unhelpful conversations. One is where a small AI number gets dismissed as irrelevant. The other is where a falling organic number gets blamed on AI before anyone has checked the tracking.
The businesses that get this right are not the ones with the most sophisticated AI dashboards. They are the ones that trust their sales data first and read every channel figure, AI included, against it.
So if I could put one question on a CEO's list for Monday, it would be this: what share of last month's orders does our analytics actually record? If nobody can answer it, every channel number in the deck, AI included, is a guess.
Talk to Govind
Founder & Principal, Search & AI Visibility. I take on a limited number of engagements.