Learn / Strategy
What agentic commerce does to conversion rate
Agent traffic distorts conversion rate in both directions: requests that never run your analytics are invisible, and referred sessions arrive with intent far higher than your baseline. Reporting them together produces a number that describes no real population, so the practical response is to segment assistant referred sessions and judge them against themselves rather than against site average.
Agent traffic distorts conversion rate in two directions at once, which is why the aggregate number becomes less useful rather than simply moving.
The invisible half
Most agent requests never execute your analytics script, so they are served by your infrastructure and absent from your reports. They are not counted as sessions and they do not dilute anything.
This is why the expected flood of non-converting traffic mostly does not appear. See how to find AI agent traffic in Google Analytics.
The high intent half
When an assistant sends a shopper to your checkout, that session arrives after the comparison has already been won. The evaluation, the shortlisting and the choice all happened elsewhere, so what lands is much closer to a purchase than your baseline traffic.
Mixed into a site average, that cohort is invisible in both directions: too small to move the number and too different to be described by it.
Where these sessions actually fail
They fail at checkout, in the environment you do not test: cold, signed out, narrow, inside an in-app browser. A referred session that reaches you having already decided, and then loses its cart across a redirect, is the most expensive failure your store has.
See how to test checkout in an in-app browser.
What to measure
- Segment assistant referred sessions and report them on their own.
- Use revenue per session rather than a raw rate, since the cohort is small and the rate is noisy.
- Track checkout step completion within the cohort, because that is where the loss is.
- Keep agent request counts entirely separate. They measure whether you are being considered, not whether you are converting.
And treat every one of these as a floor: referrers get stripped, and unidentified agents are missed. See how much of your traffic comes from AI agents.
Questions
Will agent traffic tank my conversion rate?
Usually not, because most agent requests never execute the analytics script and so are not counted as sessions at all. The distortion is subtler than a flood of non-converting visits.
Should assistant referred sessions convert better?
They arrive with much higher intent, since the comparison already happened. Whether that shows up depends on your checkout surviving a cold session in a narrow in-app browser, which is exactly where these sessions fail.
What should I report instead?
Segment them, report them separately, and use revenue per session rather than a raw rate. Comparing an assistant referred cohort against site average tells you very little about either.