Proving AI Search Is Worth It: Attribution Past the Session Count

The channel that moves revenue while showing up in your reports as direct traffic.

Sessions Are Not The Answer

The usual way to measure AI search is to build a GA4 segment for referrals from chatgpt.com and its peers and report the session count. That is a reasonable first step and a bad final answer.

The number will be small. For most small businesses it lands somewhere between a fraction of a percent and three percent of sessions. Reported on its own, it makes the channel look irrelevant, and it will lose the budget argument against paid search every time.

The measurement is fine as far as it goes; sessions are just the wrong unit for a channel whose main effect happens before anyone visits your site.

Why AI Traffic Under Reports

Several things compound to hide it, and it is worth knowing all of them before you present a number to anyone.

Some assistants strip the referrer, so the visit lands in your reports as direct. Some answers name your business without linking at all, and the person then searches your name, which books as organic or direct. Desktop app traffic frequently arrives without referrer data. And a person who reads an AI answer on Monday and calls you on Thursday is not attributable by any tag.

The result is a channel that is systematically undercounted by every tool you have. The gap between measured and actual is large enough that treating your session count as the truth will mislead you.

The Assisted Conversion Problem

AI search mostly does what a good referral does. It creates awareness and trust, then the conversion happens through a channel that takes the credit.

The same dynamic makes display and organic social look weak in last-click reporting. The difference is that AI search is newer, so it does not yet have the benefit of the doubt that older channels have accumulated.

If you run last-click attribution and nothing else, AI search will look worthless indefinitely, no matter how well it is performing.

Three Measurements That Hold Up

There are three things you can measure that survive scrutiny, and together they are enough to make a decision.

First, branded search volume. Pull it from Search Console monthly. When an assistant starts naming you, people search your business name. A rising branded trend with flat spend is the clearest quantitative fingerprint of the channel.

Second, citation share. Once a month, run your top ten customer questions through the major assistants and record how often you are named versus your competitors. It is a share-of-voice number rather than a revenue number, and it moves before revenue does.

Third, direct traffic to deep pages. Direct traffic to your homepage is ambiguous. Direct traffic to an interior service page that nobody would type from memory is usually an unattributed referral, and AI answers are now a substantial share of those.

Ask The Customer

The most reliable attribution method available for this channel is also the least technical. Add one question to your intake form or your phone script. "How did you hear about us?"

Leave it open text rather than a dropdown, because the dropdown will not have the option you need and people will pick "Google" by default. Read the answers monthly.

When "ChatGPT recommended you" starts appearing, you have attribution that no analytics configuration would have produced. For most small businesses this shows up in intake conversations six to twelve months before it shows up in a report.

What A Reasonable Target Looks Like

Some honest numbers, so you can calibrate rather than chase a figure someone invented for a sales deck.

AI referral sessions in the range of one to three percent of total sessions is normal right now for a small business doing this reasonably well. Being named in three or four of your ten priority questions is a solid position. Branded search up meaningfully year over year with no brand spend is a strong signal.

What is not realistic is AI search replacing organic traffic, or any promise of a specific citation rate. Anyone quoting you a guaranteed number for this is guessing, because the engines vary their answers run to run.

Reporting It Without Overclaiming

Report the three measurements together with the intake answers, and be explicit that the session count undercounts. That framing is both accurate and more persuasive than a single inflated number, because the first person to check will find the undercount anyway.

Set the expectation that this is a leading indicator. Citation share moves first, branded search follows, revenue follows that. Judging it on this month's session count is judging it on the one metric guaranteed to look worst.

This is the sort of thing askotter reports on monthly, alongside the search work, because the whole point of a transparent platform is that you can see what moved rather than taking a number on faith.