What the Slop Backlash Actually Shows

In 2025, two dictionary committees on opposite sides of the world named a version of the same word. Merriam-Webster chose "slop." Macquarie chose "AI slop," and its committee and its public vote landed on the same term: the fourth time in the prize's history those two have agreed. That is the strongest single fact in this discussion, and it is a fact about vocabulary. It tells you the phenomenon became common enough to need a name. It does not tell you anyone stopped buying.

Most of what has been written for marketing leaders since then runs ahead of the evidence. The evidence is worth separating into what happened, what people say will happen, and what remains untested, because the three carry different weight and only one of them should move a budget.

What happened

The junior execution rung thinned out. Reporting on mid-size marketing teams describes them now carrying "far fewer, sometimes zero" purely execution-focused junior roles, with AI absorbing most junior-level execution. Note what that dataset covers: content throughput. It says nothing about strategy roles, brand roles, or positioning work. The layer that got absorbed is the layer that was already the most standardized inside the org, the person assigned to fill the calendar.

Trust in machine-mediated answers is low and was measured. The Reuters Institute put trust in AI chatbot answers at 20% globally, 6% in the UK, against 37% trust in news overall, which is the lowest figure they have recorded since 2015. Read together, those numbers describe a channel where trust was already falling before AI content arrived in volume. The slop did not open the hole. It arrived into one.

This pattern has run before, and each time the same thing happened to the people doing the work. Blogging made publishing free, and editors got more valuable. Streaming made recorded music effectively infinite, and playlist curation became a business. Content farms made articles cheap, and the byline started carrying weight it had not needed before. Those cycles played out over years. This one is playing out over months. That is an impression from watching it happen, not a timed measurement, and it changes how the shift feels more than it changes what the shift does.

What people say will happen

A widely-cited survey found 72% of Gen Z respondents reported taking action against a brand after identifying AI-generated marketing, and 43% reported stopping purchase altogether. That is stated intent collected in a survey, and stated intent about brand punishment is notorious for outrunning what actually shows up in card statements. It is the standard finding across decades of attitude-behavior research, not a number specific to this survey.

Use that number to size a risk. Do not use it to justify a line item. If a CMO builds a case on 72% and someone in the room knows how stated-preference data behaves, the case falls apart and takes the rest of the argument with it.

What is untested

Categories where the customer hands over money or data on the strength of believing you, payments, lending, insurance, health, should absorb more damage from generic copy than categories selling taste or convenience. I have not seen a study isolating that. It follows from what trust does in those categories, and it is inference.

The same holds, harder, for markets with a high fraud baseline. In the Philippines a large share of people met digital money through cloned wallet pages and marketplace sellers who took the deposit and went quiet. The comment sections under unfamiliar brands are full of one question: legit ba 'to? And the texture of an actual scam page there is not sloppy. It is flat, confident, slightly-off-register English that reads generic. A brand publishing in that register does not read as lazy to that reader. It reads as one of those. I have no conversion data isolating the penalty, and nobody I know does. It is the thing I would instrument first if my revenue depended on it.

What to do about it

Four things, narrow enough to hold.

1. Give one named person the authority to kill work, and make it someone senior. Not a review committee, not a checklist. A person whose name is attached to what shipped and what didn't. The rejection is the part of the process that AI cannot supply.
2. Keep hiring juniors, into a different job. The junior role that got absorbed was production. The one that didn't is relational: creator partnerships, community, PR, anything where a named person has to be known by another named person over months. Agents do not hold relationships, and senior staff cannot hold one on a junior's behalf.
3. Stop reporting volume. Pieces published, posts per week, and content velocity all measure the input that stopped being scarce. Report content-attributed signups or content-attributed pipeline, and report a kill rate alongside it. A quality gate that rejected nothing last month did not operate.
4. Put names on things. Author pages, real bylines, named customers on the page and on camera. This is cheap, and it survives a trust collapse that first-party claims about your own reliability do not.

We're in a fragile state

Model output keeps improving and detection keeps getting harder, and if "human-made" becomes unverifiable, it stops being worth anything as a signal. Anyone who spent this year buying certification marks and made-by-a-person badges may find they bought decoration.

The editorial work does not have that exposure, because it changes what gets published rather than what gets stamped on it. That is the reason to build the second thing and treat the first as optional.

One test tells you whether an organization has actually changed: ask what it killed last month and who made the call. A team that ships forty pieces and rejected none of them is running the 2023 operating model with better tools, and the 20% trust figure is what that model now runs into.

Sources: Merriam-Webster / Macquarie "slop," via NYT · Gen Z brand-action survey · Reuters Institute Digital News Report 2026 · CMO.works, junior marketer restaffing