Kevin Smith
Ask a marketing team how many hours AI saves them each week and you'll probably get a confident number. Then ask how many of those hours go to fixing what AI produced and the room will get a little quieter.
It's that second number that matters more than most teams are willing to admit.
In a Workday survey, 85% of respondents said that AI saves them somewhere between one and seven hours a week. The same research found that about 37% of that saved time goes right back into rewriting and correcting weak AI output. For every 10 hours AI saves, nearly four go to cleanup.
The Cleanup Lands on Someone Else
The person who drafts a marketing strategy with AI in 20 minutes sees that as a win. On the other hand, the colleague who spends the afternoon trying to figure out what that strategy is trying to say and what they should do with it may see things differently.
Researchers at Stanford's Social Media Lab gave this a name: workslop, meaning AI-generated work that looks finished but lacks the substance to move a task forward.
According to their research, about 40% said they had received workslop in the past month. Each instance took an average of about 2 hours to deal with.
What's worse, there's a real risk of damaging your reputation and straining your professional relationships. About half of recipients said they thought less of the sender's ability, and 42% said they trusted that person less.
For an agency or an in-house team, that's the part that should sting. An AI-generated deliverable, with no real strategic perspective or relevant recommendations in it not only costs the reader valuable time, but it costs the sender credibility.
Teams Feel Faster Than They Are
I spend a lot of time writing about human behavior in marketing, so in researching this topic, one study in particular caught my attention. METR discovered that experienced software developers took 19% longer to finish tasks when they were allowed to use AI tools. Yet afterward, those same developers estimated that AI had made them about 20% faster.
I think it comes down to how most of us judge productivity, which is by feel (or vibes, if you'd rather). Seeing a draft appear in seconds feels like progress, so we tend to overlook the increased amount of reading and rewriting work that often follows. Or worse, skip it, ship it, and pray for the best.
Part of the rework caused by AI-generated deliverables is fact-checking. AI models still make things up, a problem the industry calls hallucination, and they do it in the same confident tone they use for everything else.
There are plenty of viral clips of lawyers and politicians reading AI-generated briefs and speeches containing made up quotes and case law. Chances are you've caught something like this in your own AI use only to have the AI model reply, "You're absolutely right. Good catch!"
In McKinsey's 2025 State of AI survey, inaccuracy was the risk organizations most often said had actually caused them problems, with nearly a third of respondents reporting that it came with consequences.
The second problem is less obvious. The approach many take to try to get better AI output is to feed the model more context, so teams paste in brand guidelines, past campaigns, research decks, meeting transcripts, long chat histories, etc.
Researchers at Chroma discovered that may make things worse. They tested 18 leading models and found their performance grew less reliable as the input got longer, even on simple tasks. They called it context rot.
This could look like a long chat session where the copy slowly drifts from the voice guidelines you set at the start, or a constraint buried on page 30 of the brand book that never makes it into the draft.
Your Competitors Are Also Using AI
Like it or not AI is how work will increasingly get done. And don't get me wrong, it's extremely useful and definitely has a role in a marketing workflow, but it shouldn't be trusted to lead your marketing strategy, serve as your brand voice, or be the 'source of truth' on your customer.
Remember that these AI-platforms are trained to predict what text most plausibly comes next, based on patterns learned from huge amounts of writing. That makes them great at producing fluent, familiar-sounding copy, and not so great at knowing anything about your customers or your strategy. Left on its own, the model gives you the average of your category, not your point of view.
If you tell it to "emphasize personal service," you'll get a smooth paragraph about personal service that sounds like every competitor in your market, and someone will have to rewrite it.
Generic inputs absent of strategic thought and a unique brand voice or point-of-view will produce generic drafts, and turning something generic into something specific and meaningful for your brand is slow, skilled work.
AI has moved that work to the end of the process, where it's harder to capture and easier to miss. So, next time someone brags about how many hours AI saved them, ask what happened after the initial draft.