Key Takeaways
- Agentic commerce shipped on the supply side in under a year. Stripe and OpenAI released the Agentic Commerce Protocol in September 2025, Google announced AP2 with 60+ partners the same month, and Instant Checkout opened to more than a million Shopify merchants in early 2026
- The demand side has not followed. In a corpus of 3,600+ comments about AI product recommendations, roughly 59% of coded responses to a distrusted AI answer are about making it disappear, 33% are about leaving for another engine, and about 2% are about checking it
- Contentsquare found 38% of US consumers have no plans to use AI for shopping or actively avoid it, and 76% still start with traditional search
- The complaint is practical, not philosophical. About 51% of coded complaints are that the answer was stale or wrong, and 20% that it recommended a product that doesn't exist
- None of this means ignore AI. Adobe found AI-referred visits converted 60% better in July 2026. It means treat it as a cheap option, not a channel that has already moved
Edu here. A reader asked me last week whether buyer behaviour is going agentic, and whether that changes what a DTC brand should be researching. It's a good question, and my honest first reaction was that the answer is obviously yes. Everything I read says so.
Then I went back to a corpus I'd already coded, and the answer stopped being obvious. What's going agentic, at real speed, is the plumbing. Card networks, checkout protocols, merchant integrations. What people do when an AI hands them a product recommendation looks almost nothing like that.
So this piece is about the gap between those two things, because I think the gap is where the mistake gets made.
Agentic commerce infrastructure and agentic buyer behaviour are moving at very different speeds. Payment protocols from Stripe, OpenAI, Google, Visa and Mastercard all shipped between April 2025 and February 2026. In a voice-of-customer corpus of more than 3,600 comments, roughly 59% of coded responses to a distrusted AI recommendation were about suppressing it rather than acting on it.
What has actually shipped on the supply side?
Almost the entire stack, and fast. Stripe and OpenAI released the Agentic Commerce Protocol on September 29, 2025, an open standard that lets a merchant sell through AI agents with one integration while keeping control of what's sold and how the brand appears (Stripe newsroom, Stripe powers Instant Checkout in ChatGPT, September 2025).
It didn't happen alone. In September 2025 Google announced the Agent Payments Protocol with more than 60 launch partners. In October 2025 Visa published its Trusted Agent Protocol and Mastercard shipped an acceptance framework for Agent Pay, both built so merchants can recognise a verified agent before it transacts (Digital Commerce 360, Visa and Mastercard both launch agentic AI payments tools, October 16, 2025).
By February 2026, Instant Checkout in ChatGPT was open to over a million Shopify merchants, Glossier and SKIMS and Vuori among them. That is a complete path from question to paid order, and it took roughly eleven months to build.
The traffic data reads the same way. Adobe Analytics, working from more than a trillion visits to US retail sites, found AI-referred traffic up 62% year over year in July 2026 and up 1,219% since October 2024 (Digital Commerce 360, Adobe AI referral traffic data, August 19, 2026).
If you only read that, the conclusion writes itself. Get machine-readable, get into the protocols, and be there when the agents start buying. Which is roughly what every article on this topic says, and it isn't wrong. It's just half the picture.
So why doesn't the demand side look like that?
Because when you code what shoppers say they do with an AI answer they don't trust, the dominant behaviour isn't delegation. It's removal. In a corpus of more than 3,600 comments, roughly 59% of coded responses were about getting the answer off the screen, and about 2% described treating it as a first draft to check (Why Shoppers Block AI Recommendations, Insightios, August 2026).
The methods are specific and they get shared like recipes. A minus-ai operator appended to a query. A URL parameter that returns a plain link-only results page, passed around under the nickname "the disenshittification konami code." An ad blocker that turns out to remove AI Overviews as a side effect.
"you can use https://udm14.com/ , 'the disenshittification konami code', to use google ai-free, or https://noai.duckduckgo.com/"
"For me adblock blocked it. AI overviews are gone if my adblock is active and appear again once i deactivate it"
Another 33% of coded responses were about leaving altogether, moving to Kagi, DuckDuckGo, Startpage, Firefox. Notice the tense in this one. It isn't a complaint about a bad answer, it's a migration already completed.
"After the AI got some really obvious things wrong, I moved my entire family, including my ageing parents, to Kagi. No regrets."
And a recurring note underneath all of it, which is that people resent having to do this at all.
"How sad that we need to actively opt-out of AI and sometimes we have no option."
This isn't only visible in my corpus. Contentsquare surveyed 1,300 US consumers in November 2025 and found 38% have no plans to use AI for shopping or intentionally avoid it, 76% still begin with traditional search, and only 30% say they'd be willing to let an AI agent complete a purchase for them (Contentsquare, AI is reshaping online shopping, December 8, 2025). And that 30% is a stated willingness in a survey, which is the softest kind of evidence there is, for reasons I've written about in why surveys fail DTC brands.
When shoppers distrust an AI product recommendation, suppression beats verification by roughly thirty to one. About 59% of coded responses in a 3,600-comment corpus described removing the AI answer through ad blockers, search operators or URL parameters, 33% described switching search engines entirely, and only about 2% described checking the answer against another source.
What are shoppers actually complaining about?
Not the concept. The output. Of coded complaints about the recommendation itself, about 51% are that it's out of date or plainly wrong, and 20% are that it recommended a product that doesn't exist. That's 71% of the objection sitting in accuracy, which is a fixable category of problem.
The wrongness gets caught fastest by people who know the category, which is exactly the buyer a DTC brand most wants.
"These are wildly inaccurate. A 2ct emerald isn't bigger than a 2ct oval. Ditto the cushion. A 3ct oval won't look nearly that large."
"Multiple times a day, anything even relatively obscure you might search for is responded to with a 'this doesn't exist..' or 'XXXX has never existed, maybe you were thinking of...' it is maddening."
About 18% suspect commercial influence, usually predicting a decline they've already watched happen to search. Worth saying that the corpus argues with itself here, and the people pushing back are often right on the technical facts.
"This might be how they finally enshittify AI enough that people finally stop using it, baking advertisements into the models."
"This is incorrect. The model responses are never influenced by ads. The model training is never influenced by advertisers."
Of course, commercially it doesn't matter who wins that argument. A shopper who believes the answer is bought discounts the answer, and no amount of optimisation reaches someone who's decided the channel is compromised. That's the same dynamic I found across five studies in customers quit a belief, not your product.
The proof layer is failing at the same time
This is the part I'd worry about most if I ran a DTC brand, and it has nothing to do with whether agents start buying. In the review-trust conversation, about 24% of coded mentions report reviews that are openly AI-written, and another 17% are reviewers defending their own use of AI, usually as spelling and structure help.
"I'm of two minds about it. I dislike the obvious AI reviews, which frankly now seem to be pushing 50%+ for some items."
The tells are community knowledge now, and roughly 14% of coded mentions describe spotting the writing itself. One of them stings, given the editorial rules I write under.
"The em dashes give it away immediately. Nobody's doing all that for an Amazon review for a tent."
The sharpest complaint is structural rather than stylistic. An AI review can only restate the product page, so it can never contradict it, which is the one job a review exists to do.
"it just rehashed and rearranged the spiel on the product page, trouble is the info on the product page was wrong and if they had actually used the product they would have found that out."
Another 13% report that the photos are AI too, which matters more than it sounds. A photo used to be the cheap proof that a reviewer actually held the thing. Meanwhile the largest single share, about 32%, makes the fair point that fake and paid reviews long predate any of this. The credibility problem is older than the technology, and AI just made it cheaper.
The assets brands rely on to close a confidence gap are weakening at the same time AI recommendations are. In a coded review-trust corpus, about 24% of mentions reported openly AI-written reviews, 14% described identifying them by writing style, and 13% reported AI-generated review photos, removing the cheapest available proof that a reviewer used the product.
Does this mean a DTC brand should ignore AI shopping?
No, and I want to be careful here, because the contrarian version of this argument is as wrong as the hype version. The visits that do arrive from AI are unusually good. Adobe found AI-referred traffic converted about 60% better than non-AI traffic in July 2026, the eleventh straight month of outperformance, generating 53% more revenue per visit with a 33% lower bounce rate.
So the channel is small and excellent, which is a nice problem. The right response is to treat it as a cheap option rather than a bet. Adobe also found 39% of retail homepages weren't machine-readable to language models by July 2026, ranging from 76% readable in apparel down to 59% in grocery. Most of that is technical hygiene a developer can fix in a sprint, not a budget reallocation.
What I'd avoid is the third move, the one where a brand rebuilds its positioning around a buyer who mostly doesn't exist yet. Optimising to be picked by an agent is a real project. Assuming your customer has handed the decision to one is a forecast, and right now the evidence says the forecast is early. For the practical side of this, I wrote up how customer language drives AI visibility.
| What the data says | What it doesn't say |
|---|---|
| AI-referred traffic grew 62% year over year in July 2026 | That it's a large share of visits. Contentsquare measured AI referrals near 0.2% of Q4 traffic |
| Those visits convert about 60% better | That the average shopper is arriving this way |
| 39% of retail homepages aren't machine-readable | That fixing it changes your revenue mix this quarter |
| Agentic checkout is live across a million merchants | That buyers are delegating purchases. Only 30% say they'd even be willing |
Where is the buying decision actually happening?
In the same place it was before, which is other people, except now shoppers are routing there on purpose. The corpus is full of people describing the workaround explicitly, as a way around both the AI answer and the search results underneath it.
"the reddit trick works because you're basically skipping past the seo and AI slop to get actual people, but you can push that same idea further by going to primary sources instead of anything summarising them."
And when the purchase is expensive enough, the fallback goes all the way back to the oldest one there is.
"Either way, the comments make it clear that I should simply go to shops and ask them. The deal seems amazing but I don't think I want to blindly buy again."
Which is the practical takeaway, and it's slightly boring. The conversations that decide purchases in your category are still happening between customers, in threads and comment sections, and they're now carrying more of the weight than they did before because the layers above them lost credibility. That's where I'd spend research time. The method is in the complete guide to VOC research for DTC brands, and the platform-level version is in how to use Reddit for DTC research.
What do buyers in your category say when no algorithm is in the room?
Insightios collects the threads where your market talks to itself, codes what repeats, and hands you the objections and language ranked by how common they are, with the real quotes attached. Flat fee, fixed turnaround.
Frequently asked questions
Is buyer behaviour actually becoming agentic?
The infrastructure is, faster than almost anything in payments. Buyer behaviour is not, at least not yet. In a study of more than 3,600 comments about AI product recommendations, roughly 59% of coded responses to a distrusted AI answer were about suppressing it, and only about 2% described treating it as a draft to check. Contentsquare found 38% of US consumers have no plans to use AI for shopping at all.
What is the Agentic Commerce Protocol?
An open standard co-developed by Stripe and OpenAI, announced on September 29, 2025, that lets AI agents transact with merchants through a single integration while the merchant keeps control of what is sold and how the brand appears. It powers Instant Checkout in ChatGPT, which opened to more than 1 million Shopify merchants in early 2026.
How much ecommerce traffic actually comes from AI assistants?
Very little in absolute terms, and a lot in growth terms. Adobe Analytics found AI-referred traffic to US retail sites up 62% year over year in July 2026 and up 1,219% since October 2024. Contentsquare measured AI referrals at roughly 0.2% of fourth-quarter traffic. The channel is compounding from a base near zero, which is why growth rates and share tell opposite stories.
Should DTC brands still optimise for AI shopping if adoption is low?
Yes, but as a cheap option rather than a channel bet. The visits that do arrive from AI converted about 60% better than non-AI traffic and generated 53% more revenue per visit, per Adobe Analytics in July 2026. Adobe also found 39% of retail homepages were not machine-readable to language models, so most of the work is technical hygiene rather than budget reallocation.
Why do shoppers block AI recommendations instead of checking them?
Because checking costs attention every single time and blocking costs one setting once. Of coded complaints about the recommendation itself, about 51% were that it was out of date or simply wrong, and 20% were that it recommended something that does not exist. Once someone decides a channel is unreliable, removing it permanently is the cheaper move.
Where do shoppers go instead of AI recommendations?
To other people, and often deliberately around both AI and search. Shoppers describe appending reddit to queries to reach human answers, switching to engines like Kagi or DuckDuckGo, and on higher-stakes purchases walking into a shop to ask. The review layer is losing its role at the same time, with about 24% of coded review-trust mentions reporting openly AI-written reviews.
What to do next
Do the cheap technical work, because it's cheap. Make your product and category pages readable to a model, get into the checkout protocols if your platform makes that a toggle, and take the well-converting visits that show up. None of that requires believing anything about the future.
Then go and find out where your own buyers are actually deciding. Not by asking them, since that gets you the 30% who say they'd be willing rather than the behaviour underneath. Read the threads in your category and count what repeats. If your buyers turn out to be delegating to agents already, you'll see it there first, in the way they talk about it.
All in all, I think the agentic thesis is right and the timeline is wrong, and being early is expensive in a way that being late usually isn't. I don't know how long the gap lasts. I do know it's still open, and that it's measurable, which is more than most predictions offer.
Sources
- Insightios. (2026). Why Shoppers Block AI Recommendations Instead of Verifying Them (3,600+ comments), August 10, 2026. Link
- Stripe. (2025). Stripe powers Instant Checkout in ChatGPT and releases Agentic Commerce Protocol codeveloped with OpenAI, September 29, 2025. Link Retrieved August 25, 2026.
- Digital Commerce 360. (2025). Visa and Mastercard both launch new agentic AI payments tools, October 16, 2025. Link Retrieved August 25, 2026.
- Digital Commerce 360. (2026). Adobe: AI-referral traffic spending, converting more than counterparts (Adobe Analytics, July 2026 data), August 19, 2026. Link Retrieved August 25, 2026.
- Digital Commerce 360. (2026). Adobe: AI-referred traffic to retail sites doubles in a year (Adobe Analytics, May 2026 data), June 17, 2026. Link Retrieved August 25, 2026.
- Contentsquare. (2025). AI is revolutionizing online shopping (survey of 1,300 US consumers via Pollfish, November 2025), December 8, 2025. Link Retrieved August 25, 2026.