There is a channel in your business right now that may be converting better than anything else you run. Adobe found that in March 2026, visitors arriving at US retail sites from AI assistants converted 42% better than other traffic and spent 48% longer on site.
It has no line in your budget. Until May 2026 it had no line in Google Analytics either. Even now, a big share of it is filed under direct traffic, alongside the people who typed your URL from a business card.
So a source of visitors that may be your best is mostly unattributed, and you can't invest in what you can't see.
That channel is AI assistants. Someone asks ChatGPT or Perplexity who to buy from. They land on your site with a name in hand, already half sold. You can measure that traffic with four layers: visible referrals, deep-page direct visits, a "how did you hear about us" field, and branded search trends.
Setting up all four takes about an hour.
Why AI visitors arrive ready to buy
Think about how a normal search visit starts. Someone types a few words, gets a page of options and opens four tabs. You're one of the tabs. They're still shopping.
An assistant visit starts further along. The person has already described their problem in full sentences and been given a short list with reasons. Much of the comparison happened inside the conversation. They click through to you to check that what they were told is true.
That would explain Adobe's numbers. A year earlier, in March 2025, the same data had AI traffic converting 38% worse than everyone else. Twelve months later it was 42% better. It is the digital version of a warm referral from a friend, and most businesses have no idea how many they're getting.
We looked at the other side of this coin in why so many AI shoppers still bounce. The traffic is valuable, but only if your pages are built to confirm what the visitor was told.
Why AI referrals hide
Analytics can only credit a source it can see, and this traffic slips out of view in two ways.
Some clicks arrive with no referrer attached, which is common when a link is opened from a mobile app or an in-app browser. Your analytics sees a visitor arriving from nowhere and calls it direct.
A bigger leak never produces a click at all. Somebody reads the recommendation and, an hour later, types your business name into Google. ChatGPT did the selling and Google gets the credit. Your report says organic branded search and you conclude your SEO is working.
The most valuable marketing in your business right now happens in a conversation you were not part of and cannot see. All you can measure is the shadow it casts.
Layer one. Catch the visible referrals
Google Analytics 4 now does part of this for you. In May 2026 Google added an AI Assistant channel to the default channel group, so open your traffic acquisition report and look for that row.
Don't stop there. Google's definition of the channel names ChatGPT, Gemini, Deepseek, Copilot and Grok as example sources, and it leaves out Google's own AI Overviews and AI Mode. So check your referral report by hand for the assistant hostnames:
chatgpt.com and openai.com. perplexity.ai. gemini.google.com. copilot.microsoft.com. claude.ai.
If some of those are still sitting in Referral, or you want one view that also covers the months before the new channel existed, build your own channel group:
- Open Admin, then Channel groups, and create a new group. It starts as a copy of the default one, so nothing else breaks.
- Add a new channel called "AI assistants (all)." Set the condition to source matching a pattern that includes each assistant hostname above.
- Move it above Referral in the order. Traffic goes to the first channel it matches, so if Referral sits first it will swallow your AI traffic.
- Save and use the new group in your acquisition reports. Custom groups can be applied retroactively, so you can see the trend immediately.
- Add new assistants as they appear. Check your referral list monthly for hostnames you don't recognise.
Then compare that segment against everything else on pages per session, conversion rate and average order or job value. Don't assume you will match the retail figures. Adobe measured big US retailers, and your numbers are the ones that matter.
Layer two. Find the hidden ones inside direct
Direct traffic is two very different groups in one bucket. Some people typed your homepage URL. The others clicked a link somewhere, and the referrer got lost on the way.
You can separate them with one observation. Nobody types a deep URL from memory. No human has ever manually keyed in your commercial services page followed by a hyphenated location slug.
So build a segment for direct traffic where the landing page is not the homepage. A good part of it will be lost referrals from assistants and messaging apps. On many sites it will be bigger than the AI traffic you can see by name. If it converts noticeably better than the site average, you have strong evidence of what is happening without needing perfect attribution.
Two refinements. Exclude your own team and known office IP addresses, since staff often land on deep pages directly. Tag the links you put in emails and on printed QR codes with UTM parameters, so they don't pass for hidden AI visits.
Layer three. Just ask
The most accurate attribution tool available to a small business costs nothing and takes ten minutes to install.
Add one optional open text field to your enquiry form. "How did you hear about us."
Make it a text box and not a dropdown. A dropdown gives people the options you already thought of. An open field gives you the sentence "I asked ChatGPT for someone who does this in Gosford and you came up twice."
Give it a month. You will see that assistants are sending people, and what those people were told about you. If three prospects mention the same detail, that detail is what the machine is repeating about your business.
Ask the same question on sales calls and at checkout for phone orders. Write the answers down word for word in your CRM. Over a quarter, you'll have a record of how assistants describe you, in your customers' own words. If you want to go further, run the prompts yourself, which we walk through in what AI is saying when customers ask about you.
Layer four. Watch branded search
Open Search Console and chart impressions and clicks for queries containing your business name, month over month.
Those brand queries are the exhaust of every invisible recommendation. Someone hears about you somewhere you cannot track, then looks you up by name. When more people are hearing about you upstream, this line is usually the first to move.
It is a noisy signal. Word of mouth and the sign on the side of your van push it too. Read it as a trend over quarters, and it will tell you whether your name is being said in rooms you are not in.
A worked example: the accountant who found a hidden channel
Here's a hypothetical. Say you run a small accounting practice in Parramatta. Your analytics shows most enquiries coming from organic search and direct traffic, with a sliver of referrals. You've been debating whether to cut the time you spend answering questions on forums and chasing client reviews, because it doesn't seem to produce anything.
You spend an afternoon on the four layers. The AI assistants channel shows a small but steady stream of visits from chatgpt.com and perplexity.ai. The deep-page direct segment, visitors landing straight on your "small business tax returns Parramatta" page with no referrer, turns out to be much larger, and converts well above the site average. Within six weeks, the open text field on your form has several entries mentioning ChatGPT, and two of them say the assistant described you as "responsive and good with tradies."
The forum answers and reviews were feeding the recommendations that produced your best leads. So you keep that work and add to it. You ask every tradie client for a detailed review and put your response time and trade experience front and centre on the landing page. You only make that call because the channel became visible.
What to do once you can see it
- Fix the landing experience for verification. These visitors arrive to confirm a specific claim. Lead with the fact, answer the obvious follow ups, make the next step tiny.
- Reinforce whatever the machine is saying about you. If the recommendation keeps mentioning your response time, put your response time on every key page and back it with proof.
- Invest upstream on purpose. That means third party mentions and detailed reviews, plus quotable content on your own site. Those are what produce this traffic, and until now you have been running them by accident.
On that last point, two places to start: why AI reads your reviews before it recommends you, and the GEO playbook for getting cited by ChatGPT, Gemini and Perplexity.
The internal argument this wins
Every business has a conversation each year where someone asks what the content and the review chasing are returning. For years the honest answer was a shrug and a story about brand.
With the segment built, you can answer with your own numbers. Picture being able to say: "This segment is 14% of our traffic, converts at nearly twice the site average, and has grown every month since March." A sentence like that protects a budget.
It is an hour of setup, and it makes the channel visible before someone decides to cut the work that feeds it. Go and build the segment.
If you'd rather have it built for you, our digital marketing team sets up the channel groups, segments and form fields, then turns what they reveal into a plan for earning more AI recommendations.
Frequently asked questions
How do I track ChatGPT traffic in Google Analytics 4?
Start with the AI Assistant channel Google added to GA4's default channel group in May 2026. To catch assistants it misses and to look further back, create a custom channel group in Admin, add a channel that matches sources such as chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com and claude.ai, and place it above Referral. Then add a segment for direct traffic landing on deep pages, because some AI visits arrive with no referrer and hide inside direct.
Why does AI referral traffic show up as direct traffic?
Some clicks arrive with no referrer attached, which is common when a link opens from a mobile app, so analytics has nothing to credit. Many people never click at all. They read the recommendation, then type your name or URL later. Analytics can only record what it sees, so those visits are filed as direct or branded search and the assistant gets no credit.
Is traffic from AI assistants worth optimising for?
For most businesses, yes. Adobe's data on US retail sites had AI referred visitors converting 42% better than other traffic in March 2026, and they often arrive with a recommendation already in mind. The volume may be smaller than search. Measuring it properly is the first step, because you cannot justify investing in a channel you cannot see.
How can I find out what AI is telling people about my business?
Ask your customers. An open text "how did you hear about us" field on your enquiry form, plus the same question on sales calls, captures the exact words assistants use about you. You can also run relevant questions through ChatGPT, Perplexity and Gemini yourself. Repeated details show what the machines are saying, so you can reinforce accurate claims and correct wrong ones.