TL;DR: Less than search does, and the comparison is misleading. AI answers send fewer clicks per impression and the clicks they do send convert better, because the engine has already filtered. More importantly, being named in an answer has value whether or not anybody clicks, and treating a click as the only outcome will make you underinvest in the channel until it is too late to catch up.
The honest answer on volume
AI assistants send meaningfully less referral traffic than traditional search, for an obvious reason: the answer is on the page. Somebody who asks ChatGPT which invoicing tool to use and receives three names with a sentence each may never click anything. They may open a new tab and search your brand directly, which arrives in your analytics as direct or branded search rather than as AI referral.
That last part is why most measurements of this understate it. A meaningful share of AI-influenced traffic is invisible as AI traffic.
What the numbers look like
For most businesses we see, AI referral traffic is currently a low single-digit percentage of total sessions. Growing, and small.
Two things complicate the comparison:
Attribution is poor. Some assistants pass a referrer, some do not, and brand searches triggered by an AI recommendation are indistinguishable from any other brand search.
Quality differs sharply. The clicks that do arrive tend to convert better than average, because the person has already been filtered: they asked a question, received a recommendation, and clicked deliberately rather than scanning ten results.
Why traffic is the wrong primary metric
Because the recommendation is the product, not the click.
If a buyer asks an engine which tool to use and your name is one of three, you are in the consideration set. If your name is absent, you are not — and no amount of traffic from elsewhere fixes being excluded from the shortlist at the moment of decision.
This is closer to how word of mouth or an analyst mention works than to how a search result works. You would not measure a Gartner mention by the clicks it sent.
What to measure instead
Presence rate. On your buying questions, how often are you named. This is the number that maps to being considered.
Share against named competitors. Who is named when you are not. This tells you who you are losing to, which is more actionable than a traffic figure.
Sentiment and description. Being named badly is its own problem.
Branded search volume. The best available proxy for AI-influenced demand that does not show up as referral traffic. If AI recommendations are working, branded search tends to rise before referral traffic does.
And yes, referral traffic, as a secondary metric, acknowledging it undercounts.
The case for investing before the traffic justifies it
The uncomfortable argument, stated plainly.
Positions in this channel are currently cheap. Our research across fourteen markets found entry bars — how small a cited site can be and still get quoted — ranging from 12 to 66 out of 100, and in several markets the typical cited site scored under 30. Those are positions a small company can take now.
They will not stay cheap. Every market we measured where the bar was high was high because incumbents had accumulated presence, and accumulation is exactly what is happening now in the markets where it is low.
So the honest framing is not "this sends traffic today". It is "the cost of entry is unusually low today and will not be later", which is a different kind of decision and should be made deliberately rather than by waiting for a traffic report to make it for you.
When it is reasonable to wait
If your buyers genuinely do not use AI assistants — some markets, some demographics — then measure that and act accordingly. The way to find out is to ask customers rather than to assume in either direction.
If you have urgent problems in channels that currently produce all your revenue, fix those first. This is not an emergency.
Frequently asked questions
What percentage of my traffic should be from AI?
There is no benchmark worth quoting yet, and anybody offering one is guessing. Low single digits is typical today.
Will AI traffic replace search traffic?
Not in any timeframe worth planning around. It is a second surface, not a replacement, and traditional search still sends the majority of clicks for almost everybody.
How do I track AI referral traffic?
Google Analytics will show some of it by referrer. Search Console shows AI Overview impressions. Both undercount, and the gap is the part that arrives as branded search.
Is it worth investing if the traffic is tiny?
That is the real question and it depends on your market. The measurable argument is that entry costs are low now; whether that is worth a budget line is a judgement call, and the measurement at least makes it an informed one.
How do I know if my buyers use AI assistants?
Ask them. A question in your next ten sales calls will tell you more than any industry statistic.