AI assistants sent about one percent of our traffic

There is a widely repeated claim that traffic is shifting from search results to AI assistants, and a correspondingly loud market in services that promise to capture the shift. We had a live property, fifty six posts and an analytics instrument, so the question was answerable rather than arguable.

The measurement

Sessions by source and medium, filtered to the blog, over twenty six days. Not a sample, not an estimate, not a third-party panel. Every session the property recorded in that window.

Twenty nine sessions in twenty six days across fifty six posts. About one a day, for the entire blog.

The counting error, which is the reusable part

The first pass filtered on the traffic medium, looking for sessions tagged as coming from an AI assistant, and returned twenty five. The correct figure is twenty nine. Four sessions were sitting under source values that name an assistant while carrying a medium that does not.

Count this channel by source, never by medium. A medium filter undercounted by sixteen percent here, and it undercounts in the direction that flatters whoever is reporting a small number and understates whoever is reporting a growing one.

What it is worth

At roughly one session a day, valued at the best page rate this property achieves, the entire assistant channel is worth on the order of a rupee a month. That is not a rounding error in the revenue. It is smaller than the rounding error.

The consequence was specific and it killed a plan. A new site on a new domain starts with no search authority, and the argument for launching one rested on assistants being a viable route to readers without that authority. At one percent, they are not. The channel is real, it is measurable, and it cannot carry a cold domain.

What the lane does respond to

The signal here is too small to lean on, and it is offered as a direction rather than a result. The sessions clustered on timely, event-bound, question-shaped pages: dates, transits, a change to a tax form, whether a given thing is safe. Not on methodology pages.

Worth stating plainly because it cuts against our own preference: three pages on this property were built deliberately to the citability specification, with a checkable number, a stated method, a sample size and full parameter disclosure. Over the same window all three together drew twenty four sessions, of which exactly one came from an assistant. Building for citation did not produce citation traffic.

The limit of this measurement

Analytics counts clicks from assistants, not citations by them. An engine can name a source and send nobody. So this number is a floor on influence and an exact figure for traffic, and those are different things.

For a revenue question the click is what counts, because a citation that sends no session produces no reader. For a brand question it is not, and we have not measured the brand question. Asking engines directly whether they cite the property is the other half of this and it remains unrun.

How to check your own

Pull sessions by source for your last month. List every source value containing an assistant name rather than filtering on medium. Divide by total sessions. If anyone is selling you work on the strength of this channel, that percentage is what they are selling against, and on most sites it is small enough to read before you commit a budget.