Merriam-Webster picked “slop” as its 2025 word of the year, and marketers have spent the months since arguing about how to spot it. We think that is the wrong argument. The better question is who set the standard the AI was meant to meet, and whether anyone gave that standard to the AI.
We build an AI marketer, so we have a stake in this. Read what follows with that in mind.
The word of the year is a symptom
Merriam-Webster defines slop as low-quality digital content, produced in bulk with AI. Nothing in that definition blames the model. It describes volume that nobody checked against anything.
Marketing content has always gone wrong in one place: the gap between the person who owns the brand standard and the person doing the work. A new freelancer, a regional team and an agency junior all work from a brief and a brand book PDF that nobody reads to the end. AI is the newest person in that gap, and the least informed. It is fast, it never gets tired, and it has not read your brand book.
Give it a vague brief and it writes for everyone. Writing for everyone is how you end up sounding like everyone.
Audiences react to the quality, not the label
A 2026 DoubleVerify report on EMEA found that 49% of UK consumers would feel negatively toward a brand whose AI-made ads look low-quality or strange. When the AI-made ads looked polished and professional, 29% still reacted negatively.
DoubleVerify’s own conclusion is that people respond less to the use of AI and more to the quality and context of how it is applied. The fall from 49% to 29% is the quality effect. The 29% that remains is a reason not to treat polish as a free pass.
Kantar’s 2025 survey, as reported by Digiday, found that 68% of consumers feel positive about what generative AI can do, against 75% of marketers. That measures optimism, not trust. It does not describe audiences who have turned their backs on the technology. It describes audiences who will judge what they are shown.
Detection and disclosure arrive too late
Labels have a place. Digiday reports that more than 1.3 billion videos on TikTok now carry an AI label. A label tells an audience what was made, after it was made.
It does nothing for the person deciding what to make. Whether a post sounds like your brand is settled at the brief, long before anyone applies a tag or runs a detector.
So the useful work is upstream. Taste lives in the brand owner’s head, and the job is making it readable to every person and every tool that writes for the brand. Until that is done, detection is only a faster way to find the slop.
What a written standard looks like
A standard an AI can use is the same kind a new hire can use: specific enough to disagree with. “Friendly but professional” fits every brand, so it tells a writer nothing.
In Mavic, the standard lives in Brand DNA. It holds the brand’s personality and positioning, its writing style, a ban list, customer profiles and the product catalogue. Each part answers a question a writer would otherwise guess at: how the brand sounds, who a post is for, and what is true about the product.
Our help centre teaches a simple test. Open a fresh chat, ask for three Instagram captions for your most popular product, and read what comes back. If they sound generic, it points to writing style, personality and customer profiles as the three biggest levers.
Our rule of thumb is that generic output is usually a setup problem, not a model problem. The help centre puts a full setup at under two hours, and we would rather say plainly that the result depends on how well it is done.
One more limit, stated plainly. A ban list is a strong nudge, not a hard filter, and a banned phrase can still slip into a long document.
Agents raise the stakes
One model writing one caption from a vague brief gives you one off-brand caption. That is annoying, but you can see it.
Several agents working in parallel from a vague brief is a different problem, because each one fills the gap with its own guess. The guesses do not match, and the mismatch is what an audience reads as a brand with no voice.
That is why a shared standard matters more as AI gets more capable. In Mavic Tandem, one brief is split across specialist AI agents, and every agent works from the brand’s saved Brand DNA. The principle is the one in this piece: fast workers need one shared reference.
Keep a person at the gate
A standard tells the writer what good looks like. An approval step tells you whether the work got there. In a Bynder survey of 1,800 business leaders, nine in 10 said human oversight remains essential when using AI to protect brand identity and compliance.
In Mavic, nothing publishes until a person approves it. We say that plainly because it is the control we would want ourselves.
An approval step has its own limit. A tired approver waves things through, and an approver who has never seen the standard cannot hold it. Which brings us back to writing the standard down.
A test you can run on Monday
None of this needs Mavic. You can check where your own gap is with five small steps.
- Pull your last five published posts and hide the logo. Ask someone outside the team which brand wrote them.
- Write down three things your brand always says and three things it never would.
- Name who the post is for, as a person, not as “our audience”.
- Give those notes to every person and every tool that writes for you.
- Decide who approves, and what they check against.
Slop is a standards problem before it is a technology problem. Write the standard down, hand it to everyone who writes for you, and keep a person at the gate. Mavic is one way to do that. A shared document and a firm editor is another.
