Aqxle follows AI recommendations to the cart and shows where the sale leaks.
Independent measurement of the AI-to-cart path. Every route observed, not modeled.
The Commerce Graph maps the route from an AI recommendation, through publishers, affiliate links, retailers, marketplaces, sellers and carts. At each hop, attribution, demand, control or revenue can leak. We measure where.
Per 100 times ChatGPT names a beauty brand, only 46 send shoppers somewhere they can buy it.
168 non-branded U.S. beauty prompts × 5 captures, 19–24 Sep 2026. Per-run average. Observed, not modeled.
A shopper asks ChatGPT what to buy.
non-branded beauty prompts, 5 captures each.
ChatGPT cites a publisher’s article.
of links in the AI answer carry an AI source tag (4,082 of 4,084).
The article links out to retailers.
of shopping links on the cited publisher page are rerouted through an affiliate network.
Followed by script, redirect by redirect.
shopping links credit the publisher, not the AI.
The retailer page where the shopper lands.
still show an AI origin. Sephora, Ulta and Nordstrom: 0%.
Whether the product can actually be bought.
of Amazon links with a seller result go to a reseller or a listing no one can buy from.
On Amazon, prestige leaks more than mass.
Every link ChatGPT cited, followed to the page where the shopper lands. ChatGPT tags nearly every link it sends. By the time the shopper reaches the retailer, almost none of that signal survives.
Source: Aqxle, AI Commerce Leakage Index, Beauty, Edition One (September 2026). ChatGPT, U.S., 19–24 Sep 2026. Links followed by script to the final page; no live clicks or purchases. Link shares are observed; nothing here is a revenue estimate. Each figure is measured at its own hop, on its own set of links, so this is not a funnel.
Links are only half the route. We asked Meta's shopping agent, Muse, to buy 17 best-selling beauty products at the best delivered price, then checked every store by hand.
times Muse picked the cheapest store it could buy from
12 of 17 runs
picks went to Meta partners Sephora and Walmart
Target, not a partner, got the most
misses broke the shopper's rules
A pickup-only listing and an unauthorized reseller
No sign that Muse favors Meta's partners. Its problem is finishing the job.
“When an agent decides for the shopper, a bad route doesn't cost you a click. It costs you the sale, and you never see it.”
17 best-selling beauty products, one test account, Oct. 21, 2026. Delivered prices (item, shipping and tax) were checked by hand at every store. Nothing was bought. Observed, not modeled. 17 runs shows no sign of partner favoritism; it doesn't prove there is none.
We'll run your top 10 products through Muse and show you which store it picks, what it quotes and where it goes wrong.
“In AI Commerce, the money doesn't go to whoever the AI recommends. It goes to whoever controls the route from answer to checkout.”
Proving that media works has been the through-line of my career.
For 25 years I have measured the commercial impact of advertising and media for major brands:
Winner of the WPP Atticus Award for Thought Leadership.
Former head of the 4As Measurement Committee. Former Media Ratings Council board member. Current member of the ARF Analytics Council.
AI Search is the newest channel that needs the same rigor. Most tools stop at mentions. I follow the recommendation all the way to the cart so brands can see what is actually captured—and what leaks.
Readers are affiliated with 225+ U.S. organizations, including Amazon, Google, PepsiCo, Procter & Gamble, JPMorganChase, Bloomberg, Nielsen and WPP Media. Reader affiliations are shown for context and do not imply endorsement, sponsorship or affiliation with Aqxle.
Advisor-led measurement, powered by the Commerce Graph.
A conversation to explore mutual fit. No pitch.
A fixed-scope, advisor-led engagement. We follow 25–30 high-intent shopping prompts to the cart across four AI assistants, benchmark two competitors, estimate what the leakage is worth, and hand you a prioritized fix list in a 60-minute readout.
Monthly re-measurement plus advisory as routes change.
No pitch. Nothing to prepare. Brands, retailers and publishers welcome.
Prefer email? Write to hello@aqxle.ai.
Choose a slot on my calendar. It’s free and takes 15 minutes.
Choose a timeOpens my Calendly page in a new tab.
Or send a few details first and we’ll find a time.
A conversation, not a pitch. You tell me about your brand and category, I share what the research shows for brands like yours, and we decide which path fits, or whether none does.
We explore mutual fit in the free 15-minute conversation. The AI Search Revenue Diagnostic maps the route from AI recommendation to cart in a fixed-scope, advisor-led engagement. Ongoing measurement adds monthly re-measurement and advisory as routes change.
Nothing for the conversation. For an AI Search Revenue Diagnostic, your brand and one product or category is enough to start.
Brands, retailers and publishers. The booking form asks your role so I can prepare.
No: observed, not modeled. Every figure is a per-run average of five captures, with the range across captures. Link shares are observed by script, with no panels, no revenue estimates and no live clicks.
Edition One, September 2026. A standing measurement of what happens between the moment an AI names a product and the moment someone gets paid for it.
Weekly analysis at the intersection of AI search and commerce, plus the monthly deep dive before anyone else sees it. Written for the people who own the number.
No. SEO still matters, but showing up is no longer the finish line. When ChatGPT names a brand, the shopper usually gets there through a publisher's article and an affiliate link, not the brand's own page. In our Beauty study, only 46 of every 100 ChatGPT brand mentions led anywhere the shopper could buy that brand. The job now is turning a recommendation into a sale.
For most brands, nobody can say yet, because the sales aren't being tracked. AI search ROI is the money from purchases that started with AI, minus what you spend to earn AI mentions, divided by that spend. The hard part is the money. In our Beauty study, only about 3 in 100 ChatGPT links still showed an AI origin at the retailer. So most AI-driven sales get logged as publisher, affiliate or direct traffic. Until the full route is measured, any AI search ROI number is a guess.
Because the AI tag usually disappears before the shopper arrives. In our Beauty study, nearly every link in a ChatGPT answer carried a ChatGPT tag (4,082 of 4,084). Only about 3% still showed an AI origin at the retailer, and none did at Sephora, Ulta or Nordstrom. Publishers and affiliate networks rewrite the link along the way, so the sale gets logged as publisher or affiliate traffic.
Often not as directly as brands assume. In our Beauty study, 54 of every 100 ChatGPT brand mentions sent shoppers nowhere they could buy that brand. Among Amazon links that showed a seller, 39% went to a reseller or a listing nobody could buy from. A recommendation creates demand, but the route to the cart decides who gets the sale.
Usually the publisher, not the AI and not you. In our Beauty study, 7 in 10 ChatGPT shopping links credited the publisher through affiliate tracking. That means AI-driven demand shows up as publisher commission, and your analytics give the AI channel little or no credit.
In our test of 17 beauty products, Meta partners Sephora and Walmart got 4 of 17 picks, and Target, which isn't a partner, got the most. That's no sign of favoritism, but 17 runs is too few to prove there's none.
Muse picked the cheapest store it could buy from 7 times in 10. Its misses came from guessed prices, an unauthorized reseller, and a listing that couldn't be delivered.
You have to test your products through the agent and check every store by hand. Aqxle runs this as an agent shopping audit.
