AI Slop Just Became a Line Item.

Written By Sébastien Simpson

Six months ago I'd have called this a taste problem. Someone ships AI-flattened copy, a client notices, everyone winces, the agency quietly redoes the work. Nobody files a budget line for wincing.

That's changed. The wincing now has a dollar figure attached, a source you can name, and someone in the org who's supposed to be reducing it.


The AI slop problem stopped being about taste the day it started showing up in a budget line.

$186 per employee, per month.

Jeff Hancock, who directs Stanford's Social Media Lab, and Kate Niederhoffer, VP of BetterUp Labs, surveyed 1,150 U.S. desk workers and found that 40% had run into what they call "workslop" — AI output passed off as finished work — in the previous month alone. Cleaning it up took an average of nearly two hours per instance. They put the cost at $186 per employee per month, north of $9 million a year for a large company. Published in Harvard Business Review.

That's a program cost. It's hours nobody scoped, sitting inside a budget that already didn't have slack in it.

Let’s get real.

Fortune's Sam Birchall reported that 53% of consumers now distrust AI-generated search results, and 70% are uncomfortable with AI-generated media generally. Some businesses have seen organic search traffic drop 5% to 35% as AI answer engines reroute how people find things. In the same piece, Niel Bornman, who runs Publicis Groupe's Connected Media UK, put it plainly: "a significant portion of people now operate under the assumption that everything online is fake." Source: Fortune.


Read that as a trust problem and you're only half right. It's also a traffic problem, a pipeline problem, and eventually a revenue problem. The consumer doesn't distinguish between your AI slop and someone else's. They just stop trusting the category.

Governance hasn't caught up.

Here's the part that should worry program leads more than the workslop number itself: most organizations still don't have anyone formally watching this. A Brafton survey of 132 companies using AI in marketing found only 42.4% have a formal AI policy in place. The other 57.6% are shipping AI-assisted work with no governing document at all. Where policies do exist, IT departments own them most often, with legal and the C-suite behind. Source: Brafton.

That's an improvement on the year before, when nearly three-quarters of companies had nothing. It's also nowhere close to where a function generating measurable cost and measurable brand risk needs to be. You wouldn't run a media budget with no one checking spend against plan. Most teams are running their AI output exactly that way.

One newsroom drew a harder line.

The BBC's approach to generative AI, reported by Broadcast ahead of its formal publication, is worth studying regardless of what kind of organization you run. The rule is narrow: no generative AI for factual research or news reporting, period. Approved uses are specific — subtitle generation on BBC Sounds, live sports text pages, translation, voice recreation only with an editorial justification on file. AI use is now a checkbox on every commissioning form, which forces a conversation between producer and commissioner before anything ships, backed by an AI steering group and AI representatives embedded in individual teams. Rachel Jupp, the BBC's editorial executive for the gen-AI programme, summed up the operating principle: "the key editorial drive is not misleading audiences." Peter Archer, programme director for gen AI, co-announced the approach. Source: Broadcast.


I want to be precise about what this example does and doesn't prove. The BBC is a public broadcaster under license-fee funding and editorial-standards regulation, not an agency chasing a client's KPI. You can't lift its structure and drop it into a marketing org unchanged. What it does prove is that "AI governance" doesn't have to stay an abstraction. A checkbox, a named accountable group, and a short list of approved use cases is not a hard system to build. Most organizations simply haven't built it yet.


I wrote a while back, in a piece about brand and taste in the AI era, that homogenized output was mostly a problem of what it said about the people making it. That argument still holds. But the numbers above say something sharper: it's also a problem of what it costs, measured in hours, traffic, and trust, and cost is the language that gets governance funded.


The gap between AI adoption and AI governance isn't a technology gap. It's a program design gap — nobody owns the checkpoint between "AI produced this" and "this shipped." Fixing that isn't a creative decision or a legal one. It's the kind of unglamorous, structural work programs exist to do, and right now it's mostly not getting done.

Further Reading.

Sébastien Simpson
Next

Botsitting Is a Full-Time Job. You're Just Not Paying Anyone for It.