Is AI traffic worth anything? Measuring revenue per AI session

The number you want

Revenue per session for the AI channel, compared with revenue per session for your other channels. It is a single division and it is more informative than any visibility figure you will ever buy.

In GA4, build an exploration with Session source as the row dimension, filtered to your AI hosts, with Sessions and Total revenue as values. Divide one by the other. Then repeat without the filter for a site-wide baseline, and repeat again for organic search as the most meaningful comparison.

Three numbers, one afternoon, no purchase required. Whatever you decide afterwards, this should come first.

How to read the result

If AI revenue per session is close to your organic figure, the channel behaves like search and can be treated the same way in planning.

If it is materially higher, you have something worth investment. Assistants often send readers who are further along in deciding, and when that shows up it shows up clearly.

If it is near zero on a meaningful number of sessions, that is the most valuable of the three outcomes and the one nobody wants. It means the traffic arrives and does not convert, and no amount of improving your visibility fixes a conversion problem. Acting on that finding saves more money than the other two make.

Sample size, before you believe any of it

AI channels are small on most sites. A revenue-per-session figure computed on forty sessions is noise wearing a decimal point, and dividing by a small number produces impressively precise nonsense.

A rough working floor is a few hundred sessions and at least a couple of dozen conversions before the ratio means anything. Below that, report the raw counts and refuse to divide.

If you are below the floor, extend the window rather than shortening the conclusion. Three months of data with a stated window beats one month with an unstated one.

The systematic bias you cannot remove

Everything above covers only identified AI sessions. Assistants that strip the referring page contribute conversions that are credited to Direct, so your AI channel is missing revenue that genuinely belongs to it.

This biases the figure downward, and unlike most measurement error it is one-directional. Whatever revenue per session you compute for AI is a floor, and the true value is higher by an unknown amount.

This is worth saying out loud when reporting, because it changes what a low number means. A low figure might mean AI traffic converts badly. It might also mean most of it is invisible and what you can see is unrepresentative. Those two readings imply opposite responses, and nothing in the data distinguishes them.

What to do with the answer

If the channel is small and converts poorly, stop investing in visibility and look at the landing pages. Traffic that arrives and leaves is a page problem, not a distribution problem.

If the channel is small and converts well, the constraint is volume, and content that assistants cite is the lever. That is the only case where buying a visibility tool is the obvious next purchase.

If the channel is invisible but your logs show heavy crawler activity, you are being read and the readers are not arriving, or they are arriving as Direct. Either way, the next move is measurement rather than more content.

Comparing against the right baseline

Comparing AI revenue per session against your site-wide average is the obvious move and usually the wrong one. The site-wide figure includes branded search and direct traffic from people who already decided, which sets an unrealistically high bar.

Organic search is the fairer comparison, because it is the same kind of discovery motion. Paid search is fairer still if you run it, since it has a known cost attached and makes the value question concrete.

Segment by landing page before concluding anything. AI traffic concentrates on a handful of pages, and if those pages are informational rather than commercial, a low revenue per session says more about which pages get cited than about the channel's worth.

The most useful version of this analysis compares AI traffic on a given page against all traffic on that same page. If AI visitors convert worse on the identical page, that is a genuine finding about the visitors. If the gap disappears, the earlier gap was about page mix.

What this number should and should not decide

It should decide how much effort the channel deserves relative to others, which is the question most teams are actually trying to answer when they start looking at AI traffic.

It should not decide whether AI search matters in general, because your figure describes your site, your pages and your buyers. A neighbouring business in the same category can reasonably get a different answer.

And it should be recomputed rather than remembered. This is a young channel whose behaviour is changing, and a figure from two quarters ago is history rather than a current fact.

One habit makes this analysis worth repeating rather than a one-off: write the three numbers and the window into the same place every time. AI revenue per session, organic revenue per session, and the session counts behind both.

A short series of those triples, built over a few quarters, answers questions that no single reading can. Whether the channel is growing in value or only in volume, whether its conversion behaviour is converging with search or diverging from it, and whether a change you made had any effect at all.

Recording the window alongside the figures matters as much as the figures themselves, because a quarter measured over eight weeks and one measured over thirteen are not comparable however similar the numbers look.