The Backlash Was Never About Gradients. It's About Trust.
Slop was an aesthetics problem the market could roll its eyes at. Trust is a behavior problem it can't.
At Cannes Lions this June, the loudest signal wasn't a campaign. It was fatigue. A Harris Poll doing the rounds on the Croisette found consumers are tired of hearing about AI, less likely to trust an ad they suspect was generated, and quicker to call a brand "cringey" for over-using it. A week earlier, Fortune ran a piece on businesses "declaring war on AI slop" — Publicis, Hachette, the Publishers Association — and argued they're mostly losing. Same month, Microsoft's aggressive AI push across Windows and Office earned a nickname that tells you everything: Microslop.
I wrote last month about the Slop Rebellion — the purple gradients, the pillow buttons, the sameness that AI-generated products wear like a uniform. That post treated slop as an economics problem with an aesthetic surface. I still think that's right. But the last four weeks have moved the story somewhere sharper, and I want to name it before it gets filed under "design taste" again.
The problem is no longer that AI output looks the same. It's that people have started to actively distrust anything they think it touched.
That's a different order of problem. Aesthetics you can ignore until the quarter turns. Distrust shows up in behaviour — in the click that doesn't happen, the renewal that lapses, the "made by humans" badge a competitor is now using to take your customer.
Phase One Was Recognition. Phase Two Is Avoidance.
The slop discourse of the spring was about recognition. Everyone learned to spot the tells. That was annoying but survivable, because recognizing slop and avoiding it are two different consumer behaviors, and only one of them costs you money.
We're now in the second one. Search Engine Land reported this quarter that AI-search adoption is rising while consumer trust in it falls — helpfulness sentiment dropped sharply year over year even as usage climbed. Read that gap the way I read the Figma quality gap last month: usage going one way while belief goes the other is not a stable position. It's a countdown.
The receipts are piling up in commerce. Aerie's "no AI" pledge reportedly drove a double-digit sales lift last quarter. Brands are now running "no AI" disclaimers as a selling point — the digital equivalent of "hand-stitched." When the absence of your most-hyped capability becomes a competitor's marketing hook, the market has told you something about trust that no adoption dashboard will.
Brian Solis put the frame better than I will: the AI backlash, he wrote in June, is really a crisis of trust and agency. People don't resent the technology. They resent being handed outputs they didn't ask for, from systems they can't interrogate, on decisions they used to control.
Trust Is a Design Problem, Not a Trust-and-Safety Ticket.
Here's where my discipline usually gets it wrong. We treat trust as governance — a policy, a red-team, a compliance sign-off that happens next to the product. Nielsen Norman Group's State of UX 2026 argues the opposite, and I think they're correct: trust is now the central design problem, and it grows every time a company ships an agent before it's ready.
The reason is structural. UI quality is commoditizing — if your value was making interfaces look correct, the floor just rose to meet you. What doesn't commoditize is the user's confidence that the thing in front of them will do what it says, tell them when it can't, and let them take the wheel back. That confidence is built from unglamorous fundamentals: transparency about what the system did, control when it's wrong, and a graceful failure state for the moment it breaks in front of someone who matters.
Most AI features shipped this year got none of that design attention, because the demo looked plausible and the question never got asked. Sound familiar? It's the 80% trap again, wearing a different coat. The model produced something confident, the org shipped it, and the 20% that earns trust — the "here's why I did that," the "you can undo this," the honest "I'm not sure" — got left on the floor because nobody was paid to defend it.
Somebody Owns the Trust Debt. Right Now, Nobody's Name Is On It.
For those of us in program and design ops, this is the operational heart of it. Every agent you deploy accrues a liability the way a codebase accrues tech debt. The Cloud Security Alliance has a phrase I've started borrowing: accountability lineage — the ability to say, after the fact, why an agent acted, what it read, and which human is answerable for it. Most organisations shipping agents this year cannot produce that sentence for a single one of them.
That's the gap I'd close first, and it doesn't require winning the philosophical argument. Keep an agent accountability register the way you keep a risk log: every agent gets a business owner, a technical owner, and a named human who answers when it's wrong in front of a customer. It's boring. It's also the difference between a trust incident you can explain and one that becomes a Fortune headline with your logo on it.
The instinct will be to route this to legal or security and move on. Resist it. Trust is experienced at the interface, not in the policy PDF — which makes it a design and delivery responsibility, owned by the people who decide what ships and in what state. That's us.
The Ceiling Moved Again.
Last month I argued the floor of competence was rising and the real question was whether anyone in your org was paid to care about the ceiling. The ceiling just moved, and it's worth saying plainly where it went. For a decade the design ceiling was is this good. This year it quietly became do people believe you.
That's a harder bar, and a more durable one, because it can't be generated. A model can produce a clean flow, a plausible summary, a competent first pass. It cannot produce the user's belief that someone trustworthy is on the other side of the screen. That belief is earned through the parts of the work AI is worst at — admitting uncertainty, handing back control, failing honestly — which happen to be the parts most orgs are cutting to hit the new floor faster.
The brands that win the next two years won't be the ones with the most agents. They'll be the ones whose customers still open the email, still take the recommendation, still assume good faith when something goes sideways. That's not a governance metric. It's the whole business. Make sure somebody senior is paid to protect it, and make sure they know it's a design job — not a disclaimer.
Further reading: Businesses are declaring war on AI slop. They are fighting a losing battle — Fortune, June 5 2026 · The AI Backlash Is Really a Crisis of Trust and Agency — Brian Solis, June 2026 · Consumers are sick of hearing about AI (Harris Poll, Cannes Lions) — Marketing Brew · AI search adoption rises as consumer trust declines — Search Engine Land · State of UX 2026: Design Deeper to Differentiate — Nielsen Norman Group