Google added a new row to every GA4 property this spring, and most owners still have not looked at it. The company confirmed an AI Assistant channel in the default channel group on May 13, and by the first week of June it was populating in most accounts. No setup, no toggle, no announcement email. Sessions from ChatGPT, Perplexity, Claude, and the Gemini app now get their own line instead of hiding inside Referral. For a local operator, this is the first native answer to a question that got expensive this year: is AI actually sending you customers?
The answer matters because the old scoreboard broke. AI Overviews now show on roughly 68% of local searches, and when one appears, average click-through drops 58%. Meanwhile ChatGPT has climbed to the third most popular source of local business recommendations, behind only Google and Facebook. You are getting fewer blue-link clicks and more AI-referred visitors, and until now your reports mashed them together. GA4 finally separates them. It is also wrong in two specific ways, and you should understand both before you trust the number.
What the New Channel Counts, and What It Quietly Skips
The mechanics are simple. When someone clicks through to your site from ChatGPT, Perplexity, Claude, Copilot, or the Gemini app, GA4 now files that session under AI Assistant in the default channel group. You will find it in Reports, then Acquisition, then Traffic acquisition. Nothing to configure. The classification runs on Google’s side, and it updates as new assistants earn enough volume to matter.
Here is the skip: Google excluded its own AI surfaces. Clicks from AI Overviews and AI Mode count as plain Organic Search, not AI Assistant. For most local websites, Google’s AI surfaces drive more AI-shaped traffic than every chatbot combined, and that traffic stays invisible in this report. So the AI Assistant row understates reality by design. Treat it as a floor, not a total.
One more catch. GA4 processes data forward only. The channel started filling in when it reached your property, and history will not be restated. If you compare July against February, February’s AI visits are still buried in Referral and Direct, and any year-over-year comparison will overstate your AI growth. Write down the date the channel first shows data in your property, annotate it in your reporting notes, and only trend from there. Clean baselines beat long ones.
The Two Blind Spots That Shrink Your Number
Blind spot one is missing referrers. Analysts digging into the new channel estimate that 60 to 70 percent of real AI sessions arrive with no referrer at all, because mobile apps and privacy layers strip the source before the browser opens. Those sessions land in Direct. If your Direct traffic has crept up over the past year while nothing about your brand changed, some slice of that creep is AI traffic wearing a blank name tag.
Blind spot two is the zero-click answer. When an assistant reads your site, quotes your prices, and answers the customer without a click, no session fires and GA4 records nothing. The customer may still call or walk in. This is the same lesson Google Business Profile taught in June when its stats froze: the surfaces deciding your revenue do not always report to you. Your phone log and your intake question, “how did you find us,” are now real analytics tools. Have staff log the answer for two weeks and compare it against what GA4 claims.
Here is the before and after you should expect. Before: Traffic acquisition shows Organic Search at 45%, Direct at 30%, AI Assistant at maybe 1%, and you conclude AI does not matter. After the fix below, plus an intake tally at the counter: AI-touched demand lands somewhere between 8% and 15% for most service businesses, once chatbot referrals, blank-referrer sessions, and “I asked ChatGPT” walk-ins get counted against the same week. The 1% was never the truth. It was the reporting gap.
The One-Afternoon Fix: Build Your Own AI Channel
The default channel is Google’s definition, and Google has already changed its mind once about what belongs in it. Build your own custom channel group so you control what counts, and so a future reshuffle cannot rewrite your trend line the way the GBP changes rewrote everyone’s June. In GA4, open Admin, then Data display, then Channel groups. Click Create new channel group, add a channel named AI Referral, and set the condition to Source matches this regular expression:
chatgpt\.com|openai\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com|meta\.aiDrag your new channel above Referral so it wins the tiebreak, save, and then select your custom group from the dropdown in Traffic acquisition. Next, open Engagement, then Landing page, and add a filter for the same sources. This report is the payoff. It shows which pages assistants actually cite and send people to. For most local sites it is not the homepage. It is a service page with clear pricing or an FAQ page that answers one question well. Whatever page wins, that is the format to repeat.
Be honest about the limits. A custom channel group also applies forward only, so build it this week, not after you need it. The regex needs a quarterly review because new assistants keep appearing. And no regex rescues the blank-referrer sessions sitting in Direct. Your proxy there: watch new-user share on deep landing pages inside Direct. Nobody types your pricing page URL from memory. If Direct new users keep landing three levels deep, an assistant sent them.
Read Your Number Before You Buy the Next Ad
The reason to get this right in July is that Google started selling against the confusion. SE Ranking published a study this month, built on 50,000 keywords pulled June 30, showing text ads now appear on 29.45% of commercial AI Mode queries. Ad presence climbs with keyword price, from 24% on cheap terms to 54% above $10. When ads show, 71% of the time there are two of them. The auction inside the AI answer is already crowded, and lead-generation categories, which is what every local service business is, surface ads the heaviest.
Now the number that should stop you: in 88% of those ad keywords, the advertiser’s domain was not among the sources the AI answer cited. Buying the ad does not buy the citation. Those are separate games, and the study is blunt about it: spend guarantees a slot, not credibility. The ad rents a slot inside the answer box. The citation, the part that makes an assistant recommend you by name, still comes from the unpaid work: structured pages, consistent listings, reviews, and content an assistant can quote.
So sequence it. First, spend the afternoon building the custom channel and the landing-page view. Second, run the two-week counter tally. Third, only then decide whether AI Mode ads deserve a test budget, and judge them on booked jobs, not impressions, since the citation data says the ad will not improve how often AI mentions you organically. The signal to watch over the next 30 days: whether Google breaks AI Mode clicks out of Organic Search the way it broke out chatbots. The moment that row appears, the operators who built their own measurement this summer will know exactly what changed. Everyone else will be starting from zero again.
Atlas Unchained exists to keep working operators ahead of shifts exactly like this one, a week at a time. If this saved you a wrong ad buy, subscribe and get the next play in your inbox every weekday morning.
