AI search traffic is the visits that reach your site after an assistant answers somebody's question and names your page. Someone asks a chatbot for a recommendation, the answer cites you, and they click through. This article is about finding those visits in your own reports, naming them properly, and judging what they are worth. It is also about the parts you cannot know, because that matters just as much.
Tracking is not the hard part here. The trouble is that no default report knows what to call this. So the visits scatter across referral and direct, and nobody sees them as a channel. If you want the wider picture of how assistants choose their sources in the first place, our guide to generative engine optimization covers that side of the work.
What AI search traffic actually is
Two different things get called by the same name, and mixing them makes a report meaningless.
The first is a click from an assistant's own window. Somebody is inside a chat, the answer links to your page, and they follow the link. That is a visit, so your analytics can see it.
The second is an appearance inside an answer with no click at all. Your page fed the summary and the reader never left. That is visibility rather than traffic, and it belongs in a different row of the report. Counting whether you appear at all is a separate job, which our piece on measuring LLM visibility takes apart properly.
In short, AI search traffic means the clicks. Keep the appearances on their own line, or you will end up reporting one number that answers neither question.
Where AI search traffic shows up in your reports
Referral, under a hostname nobody recognises
A click made inside a browser usually arrives as a referral. The browser tells your site which page sent the visitor, so the assistant's own domain turns up in your referral list.
Start there, sorted by sessions. Assistant domains sit among the forums, the newsletters and the odd scraper. Once you know which names to watch for, you can group them and stop looking them up by hand.
Direct, where the referrer goes missing
Some clicks arrive with no referrer at all. Apps, desktop clients and certain link handlers drop it, so your analytics files the session as direct.
Direct is where AI search traffic hides. It is also where typed addresses, email clicks and untagged links go, so a rise in direct is not evidence of anything on its own. Resist the temptation to claim it.
Search Console, which shows less than people expect
Google's own AI answers are a different case. As things stand, Search Console counts a click from an AI overview alongside every other click on the results page. No filter separates them today. Google does change what that report exposes, so check it yourself before you rule the idea out.
So the tool you already have will not isolate that part of AI search traffic. What you can read is the shape. Impressions holding steady while clicks fall away on informational queries is the pattern to look for, and zero click searches explains why that happens.
Why the number is always an undercount
Every method above leaks, and it helps to know exactly where.
- Referrers vanish when the click comes from an app rather than a browser.
- Privacy settings and blockers stop some sessions reaching your analytics at all.
- Some readers remember your name and search for you later, which files as branded search.
- Assistants sometimes name a business without linking to it, so there is nothing to click.
Therefore treat the figure as a floor rather than a total. It tells you the direction and it names the pages doing the work. It does not tell you the true size of the audience, and nothing can, because the missing sessions never reached a tool at all.
Setting up a channel group for AI search traffic
The fix is a custom channel group, built once and then left alone.
In practice the rule reads like this. If the session source matches an assistant domain on your list, file it under a channel called AI search. Analytics platforms put that setting in different places and use different words for it, so expect to hunt for the screen. The idea underneath does not change.
Three habits keep the group honest.
- Write the domain list somewhere your team can actually see it.
- Review the list each quarter, because new assistants launch and old ones move house.
- Never assume the group rewrites history. Some tools apply a new rule to old data and some do not, so check before comparing this month with last year.
Once the group exists, AI search traffic stops being a manual dig and becomes a row that appears whether or not anyone remembered to look for it.
What to measure once the channel exists
Sessions on their own are a vanity number. The useful questions are about what those sessions did next.
Look at three things. First, the landing pages, because they name the articles assistants actually cite. Second, the behaviour, which means time on page, pages per session and enquiries. Third, the enquiry rate set against your other channels over the same period.
Quality matters more here than volume. A reader who arrives this way has already read an answer about the subject. So they often land with a narrower question than someone clicking straight down a results page. Small numbers of visits like that can still be worth having. If you want the pages that earn the citation in the first place, earning AI citations deals with the writing side.
Reading AI search traffic against your other channels
A new channel always looks disappointing beside an old one, so compare like with like rather than raw totals.
| Channel | What it usually brings | The number to watch |
|---|---|---|
| Organic search | Steady volume across many queries | Clicks against impressions |
| AI search traffic | Readers who arrive having already read an answer | Which pages assistants cite |
| Paid search | Volume you can turn up or down | Cost per enquiry |
| Direct | Returning readers and untracked clicks | Unexplained jumps |
Two habits make that table useful. Report the AI channel on its own line instead of folding it into referral. Then judge it on enquiries rather than on sessions, because a small count swings about and the enquiry is what you wanted anyway.
The crawler side, which is not traffic at all
An assistant has to read a page before it can cite one, and that fetch leaves a trace in your server logs.
Your logs list the agent behind every request. Some of those agents belong to AI companies, and their visits show which parts of the site machines reach easily. No person visited, so this is not AI search traffic. Still, it is an early signal that your pages are in the pipeline at all, and it arrives long before the first click does.
If your hosting exposes raw logs, ask for a monthly summary grouped by agent. Where it does not, a log tool or your delivery network usually reports the same thing.
A worked example of an AI search traffic line
A worked example, with round numbers chosen to make the arithmetic easy rather than to describe a typical month. Suppose the site had 10,000 sessions. 200 of them came through the AI search channel, and another 40 arrived from an assistant with no referrer, which you will never see. Of the 200 you can see, 6 turned into enquiries, so it took roughly 33 visits to produce one enquiry. Run the same sum for organic search and you have a fair comparison. The comparison is the point of the example, not the figures.
Notice what that example refuses to say. It never claims a good number, because a good number depends on your market and your margins. It compares one channel with another inside the same business, over the same weeks, which is the only honest comparison anybody here can make.
What not to promise about AI search traffic
This is where reports go wrong, so it is worth being blunt about it.
Nobody can promise a position inside an AI answer. Assistants publish no ranking you can check. They also build each answer fresh, so the same question can name different sources on two different days. Anyone selling a fixed slot in an answer is selling something they do not control.
Equally, nobody can hand you a complete count of AI search traffic. The gaps described above are structural rather than a fault in your setup. A tool that reports an exact figure is estimating, and a good one says so on the page.
What you can do is honest and still useful. Track the trend, name the pages assistants cite, tie the visits to enquiries, and write down what changed. If the pages themselves need work, writing content for AI search is the next thing to read.
A short monthly routine for AI search traffic
Keep it small enough that it survives a busy month.
- Open the referral report and scan for assistant domains you have not seen before.
- Add anything new to the channel group, and note the date you added it.
- Record sessions, enquiries and the top landing pages for AI search traffic.
- Check Search Console for informational pages where clicks fall while impressions hold.
- Write one line on what changed and why you think it changed.
That last step matters more than the four above it. A number with no note beside it is a number nobody can interpret six months later, and this channel is young enough that the notes are most of the value.
Frequently asked questions
How do I see AI search traffic in Google Analytics?
Start in the referral report and sort by sessions. Assistant domains sit in that list among the forums, newsletters and scrapers you already recognise, so note the names you find. Then build a custom channel group that files those sources into a channel of their own. From that point the split shows up in your standard reports and nobody has to dig for it. One caution: the group only catches clicks that carried a referrer, so read the channel as a floor rather than a full count.
Why does the number look so small in my reports?
Two things push the number down, and only one of them is about your setup. The real volume may simply be low for your market, which no amount of tagging will change. The count also leaks, because clicks from apps often arrive with no referrer and some sessions never reach your analytics at all. So treat what you can see as a floor rather than a total. Watch the direction it moves over months, and watch which pages it lands on, because both stay useful even when the count is incomplete.
Does a click from ChatGPT count as organic search?
Usually not. Analytics tools tend to file it as a referral, because the visit comes from the assistant's own domain rather than from a search engine results page. Some setups then fold that referral into a broad other bucket, which is how it goes unnoticed for months. Default groupings do get updated, so check what your own report is doing rather than trusting a rule of thumb from last year. If you want the split, say so with a channel group rule of your own.
Can I separate AI overview clicks in Search Console?
Not as things stand. Search Console counts a click from an AI overview alongside every other click on the results page, and no filter pulls them apart today. Google does change that report, so look for yourself before ruling the idea out. What you can read meanwhile is the pattern on informational queries: impressions steady, clicks drifting down, average position roughly unchanged. That shape suggests the answer satisfied readers before they reached you. It is an inference rather than a measurement, so label it as one.
Should I use a channel group or a saved filter?
Use a channel group whenever the reporting happens more than once. A saved filter answers a one off question well, but somebody has to remember to apply it every month, and eventually nobody does. A channel group makes the split part of the standard report, so the row appears whether or not anyone thought to look. Keep the domain list written down somewhere your team can read, then review it once a quarter.
What is a good conversion rate for AI search traffic?
There is no honest outside benchmark to quote, and anybody who offers one cannot see your market. Compare AI search traffic against your own other channels instead. Take enquiries divided by sessions for organic search, then run the same sum for the AI channel over the same weeks. That comparison is fair, because the audience and the offer are the same on both sides. Watch it across several months, since a small channel swings wildly week to week.
Can I add UTM tags to track AI search traffic?
No, because you cannot tag a link you do not control. An assistant points people at whatever address it found, which is normally your plain page URL, so there is no campaign parameter for you to add. That leaves the referrer and your own channel group rule as the only handles on AI search traffic. Where you do control a link, in a directory listing or on a partner page, tag that one properly and keep it out of the AI channel so the two do not blur.
Does AI search traffic replace normal SEO reporting?
No, it sits beside it. The work that earns a citation is largely the work that earns a ranking, so this is one more line in the report rather than a new department. Add AI search traffic with sessions, enquiries and the pages assistants cite most often. Then judge it over quarters rather than weeks. Compare the two channels in your own reports before you move any budget, since the split differs from one business to the next.