Learn
How much of your traffic comes from AI agents?
There is no industry figure that answers this for your store, because the number depends on your category, your customers and whether you are permitting agent traffic at all. What you can do is measure your own, from server logs for agent requests and from analytics for the people assistants refer, while accepting that both undercount and that neither can be backfilled.
There is no published figure that answers this for your store. The share of your traffic coming from AI agents depends on your category, your customers, and whether you are permitting agent traffic in the first place, and that last one alone can take the number to zero without anything appearing wrong.
What the published numbers actually describe
Third party figures do exist and they are about aggregate populations. Digital Commerce 360 reported on 19 August 2026, citing Adobe Analytics, that AI-referral traffic to US retail sites was up 62 percent year over year in July 2026, converted 60 percent higher than non-AI traffic, and produced 53 percent more revenue per visit.
Growth rates and conversion ratios travel better between stores than absolute shares do. A number describing US retail in aggregate tells you the direction of the thing. It does not tell you your share, and using it as though it did is how a forecast becomes fiction.
Ottom publishes no traffic estimate of its own, and will not until its own benchmark dataset exists.
Measure two things, and keep them apart
There are two populations and they answer different questions.
- Agent requests. Machines reading your catalogue, visible only in server logs. This tells you whether you are being considered at all.
- Referred people. Shoppers an assistant sent to your store, visible in analytics. This tells you whether being considered is turning into anything.
Adding them together produces a bigger number and destroys the only useful thing about either. Reported separately, the ratio between them is genuinely informative: plenty of agent reads and almost no referred shoppers is a comparison problem, and it is a fixable one.
Both numbers are floors
Log based counts miss agents that do not identify themselves. Referral counts miss every session whose referrer was stripped by a redirect, a consent wall or a referrer policy. Every source of error points the same way, so what you measure is a lower bound rather than an estimate.
The number that matters more
The more useful question is not what share arrives, but what share of arrivals get anywhere. An agent that reads your catalogue and stops is not traffic in any sense you can bank, and that failure will not show up as a number at all.
If the share turns out to be small, the decision about whether to act is covered in do I need to do anything for AI shopping agents.
Questions
What is a normal share to expect?
There is no normal to compare against yet, and anybody offering you a single number for all of retail is describing a population you are not in. Your category, your price point and your bot rules move this more than any industry average could account for.
Is it too small to bother with?
That is a legitimate conclusion, and it is only legitimate once you have measured. The failure mode is deciding it is small because you cannot see it, when you cannot see it because you never instrumented it, or because you are blocking it.
Should I count agent requests or referred people?
Both, separately, and never added together. They measure different things: one is the machine reading your catalogue, the other is the shopper it sent. Combining them produces a number that means nothing and flatters itself.