What AI traffic is worth

The question

What purpose does the traffic arriving from AI serve for us, and how much are we willing to pay for it?

The “what purpose” is not decoration: it is the part that makes this a decision and not a measurement.

How it differs from “Staying in AI answers”

That card decides whether to be there. This one decides what it is for.

They are separate because they can be got wrong independently: you can tend the channel with conviction and judge it on the wrong metric, or judge it correctly and not tend it at all.

Why it is a decision

Because the metric the tools return by default is volume, and volume in this channel always goes down. If you do not decide what the channel is for, the tool gives the answer: it is going badly.

What the literature says

• Volume. CTR from −15% to −89%; 93% zero-click in AI Mode; Reuters at −33% and −38%; Search Console giving impressions only since 3 June 2026.
• Quality. 0.5% of sessions produce 12.1% of signups: roughly 23 times the average rate.
• Composition. Direct at 19.9% against 38.8%, branded at 55.9% against 40.4%.
• Volatility. Presence of AI Overviews on e-commerce queries from 29% to 3.2%.
• Non-observability. The informed non-opener: an informed user who opens nothing. It is not an absence of interest, it is an absence of trace.
• Downstream behaviour. 58% of shoppers use AI to inform themselves, 37% start a purchase journey with it, only 17% are comfortable completing it; and in chat the measured conversion rate is a third of that of click-outs, while generating twice as many new customers. The pattern is clear: you discover in AI, you buy on the site. If the channel is there to acquire, the number to watch is not in-chat conversion.
• Sources to cross-reference, because none is enough on its own: long-tail from Search Console, keywords from Search Ads, AI queries from Bing, internal site-search queries.

The cost of either branch

Rewriting the metric: you lose the time series and comparability with the other channels, exactly where management asks for it.

Keeping it: you make budget decisions on a number that measures the part of the phenomenon destined to disappear by construction.

How it is measured

Here the card has to say something uncomfortable: the measure exists, but it is not comparable with the previous one. Moving to conversion, qualification and traffic composition means accepting a declared discontinuity, not finding an equivalent metric. Anyone promising equivalence is reconstructing the old number with worse data.

Who decides

Digital / Web Analyst together with whoever owns the channel budget. The decision is about allocation, not measurement — and it has to be argued to the second, not the first.

Where it falls

The customer journey shortens is the parent decision. Then Intent-based interaction and Making a traditional product usable by an agent, because agentic traffic pollutes A/B tests and has to be segmented before measuring anything at all.