On 10 September, Amazon Ads opened a pilot with OpenAI that lets selected American advertisers extend campaigns already running through Amazon Ads into ChatGPT. Delta Vacations is among the first brands testing it. The integration puts paid placement inside an interface where customers are already exploring options, comparing choices, and making decisions.
Paid inventory normally arrives in a channel some years after the industry has agreed how to measure the unpaid version. Here the order has reversed. Money can now be placed inside an assistant’s answer before anybody has established what the free version of that answer was worth, and the budget lines written this month will be defended with the only numbers available.
Focus On: The Premium Measures the Customer, Not the Channel
Those numbers are strong. Adobe Analytics, drawing on retail data covering more than a trillion visits, reported AI-referred shoppers generating 53 per cent more revenue per visit and converting at a rate 60 per cent higher than non-AI traffic in July. It was the eleventh consecutive month of conversion outperformance, a reversal from the position a year earlier. Shopify supplies the mechanism behind that reversal, and it is the most useful disclosure of the year: half of all AI-referred sessions land directly on a product page, against roughly a fifth for traditional search. The assistant has read the catalogue, narrowed the field, and matched an intent before anyone clicks. What lands on the site is closer to a decision than a query.
That is not a measure of the channel. It is a measure of the customer, arriving at the end of a process the brand did not run and cannot see.
Which is why Contentsquare’s 2026 benchmark, comparing the fourth quarters of 2024 and 2025, sits so oddly beside it. Across 99 billion sessions, AI referrals accounted for 0.2 per cent of total visits while growing 632 per cent, with an absolute conversion rate of 1.3 per cent. Both datasets are sound. One reports a relative premium on pre-qualified traffic, the other an absolute share of a very large denominator, and a marketing organisation can quote either to justify what it had already decided to do.
The case for genuine incrementality is nonetheless better than sceptics allow. Seventy-five per cent of Shopify’s AI-attributed orders came from outside its top hundred categories, the sort of request keyword search handles badly and an agent querying structured attributes handles well, and new buyers arrived through AI channels at nearly twice the rate of other channels. That is consistent with demand creation. It is equally consistent with demand redistributed from search, marketplaces, and direct navigation towards whoever happened to be machine-readable when the agent asked. Nothing published this year separates the two, because none of it comes from a controlled comparison.
Two numbers on the same earnings call
Target reported both kinds on 19 August, which makes it the clearest case available.
The first is external. Digital traffic from AI platforms is growing at more than three and a half times the industry rate, a figure the chief executive immediately qualified as still small in total today. The denominator was not disclosed. In June the company had published a fact sheet claiming a 2,000 per cent rise in AI-driven traffic in the first quarter against roughly 400 per cent for retail generally, alongside the claim to be the first mass retailer live across ChatGPT, Google’s AI Mode, and Copilot. Impressive growth rates, unknown base, no counterfactual.
The second is internal and barely got reported. Inside Target’s own app, AI-powered wish-list creation increased by more than half and the number of items added more than doubled. Across the relevant back-to-school pages, conversion improved by nearly 20 per cent, although Target did not isolate how much of that improvement the AI feature caused. That evidence comes from a surface the company owns and can instrument. It is smaller, duller, and potentially worth considerably more, because Target could subject the experience to a controlled test.
The pattern generalises. The AI work you can prove is the work on your own property. The AI channel you cannot yet prove is the one your competitors are about to buy inventory in.
Before the quarter turns
Machine discoverability should be funded now and is largely uncontroversial. Adobe found that 39 per cent of retail homepages in its expanded July cohort were still not machine-readable. That is concerning enough on its own, and more so when Shopify reports that half of AI-referred sessions land directly on product pages. Structured attributes, availability, pricing, and corroborating third-party evidence are becoming distribution infrastructure, and belong to the machine customer rather than to search.
Commercial proof is harder and needs designing before the money moves. Hold something back: a market or product group where you deliberately do not improve machine-readable presence, and measure incremental orders rather than referral sessions. Treat referral counts as incomplete, because some AI-influenced visits lose their referral information and arrive looking like direct traffic. Then ask for the only number that settles anything, which is the gap between the exposed group and the held-back one.
Most organisations will skip this, as they skipped it with paid search, because a conversion premium arrives pre-built and flatters the team reporting it. By the time anybody asks for the counterfactual there is no clean population left to supply it.
My prediction is specific. Within eighteen months a serious brand will publish an incrementality test showing that a meaningful share of its AI-attributed revenue was demand it already owned, and the number will be uncomfortable enough that the industry treats the study as an outlier rather than as the first honest measurement. The brands that ran their own holdout before the paid inventory arrived will know which of the two it was. Everybody else will be arguing about attribution windows with a vendor.
Decide this quarter which market you are willing to leave alone. After Q4 you will not have one.
Can your analytics team currently distinguish an AI-referred order from an order that would have arrived anyway, and if not, what would it cost you to find out before the next budget round?
Follow me
That’s all for this week. To keep up with the latest in generative AI and its relevance to your digital transformation programs, follow me on LinkedIn or subscribe to this newsletter.
Disclaimer: The views and opinions expressed in Chronicles of Change and on my social media accounts are my own and do not necessarily reflect the official policy or position of S&P Global.