I knew within the first two minutes of glancing through the deck.

A strategy workshop for our senior stakeholders had just wrapped. The session itself was good: the partner facilitating it followed our brief, the debate was sharp, and an AI tool captured the notes. When the deck arrived, I opened it and felt that familiar dip. Clean layouts. Confident headings. Icons beside every bullet. And almost nothing that sounded like the room we had just sat in.

I can’t prove how it was made, and I won’t pretend to. But the output looked like what happens when notes go in one end of a tool and a deck comes out the other, and nobody with judgement reads it as the audience would.

What the AI got right

Capture was mostly there based on what could be heard visibly. Almost every comment was transcribed, and the notes were more complete than anything a human note-taker would have produced. If you still think the risk with AI is that it can’t capture pronunciation properly, you are a year behind, especially if you are using a fit for purpose transcription tool.

The deck is where it fell apart, and the failure came in four forms.

• Contextual failure. Several inputs we had spent real time on never reached a slide though they were transcribed. The context was missing or misunderstood, which means the human-in-the-loop was clearing missing when reviewing the outputs.
• Misconstrued inputs. This one worries me most. One point had been softened into a generic statement, and another had its emphasis flipped. A missing idea is perhaps due to a missed captured but the classification of info under the wrong category is a human oversight failing to validate AI’s outputs.
• No narrative. The deck was a tidy inventory of topics. There was no argument, no sequence of “therefore,” and no answer to “so what?” A workshop deck exists to move a group from where they started to what they now believe. This one just reported that a conversation had occurred.
• Decorative design. The visuals were there to look polished, not to say anything. A chart with no point. A framework graphic that framed nothing. Design should carry positioning and tell the story, and here it sat on top of the content like wallpaper.

Also Read: Hong Kong builds a test track for robots that keep failing outside the lab

This is not a vendor story

It would be easy to write this as “my partner used AI badly.” But I use these tools constantly, and I have shipped drafts that weren’t ready, so the point is not that I’m exempt.

Research backs up what many of us have experienced. A Stanford and BetterUp Labs study, published in Harvard Business Review, gave this phenomenon a name: “workslop”, AI-generated output that looks like finished work but pushes the effort of fixing it onto the person who receives it.

That is exactly what happened here. The partner saved time, and my team inherited the cost. We spent hours checking the deck against our own notes, rewriting the narrative, and working out which statements were ours and which were the tool’s best guess of our intent.

What I do now

Two things changed in how I work, and in how I brief partners.

• First, I run a traceability check before anything AI-assisted leaves my hands. Every input from the session must map to a slide, or I must be able to say why it was dropped. It is tedious, and it catches what a skim never will.
• Second, I ask one question of every slide: what should the audience believe after seeing this? If I can’t answer in a sentence, the slide is decoration, and it goes or gets rebuilt.

Also Read: Employers are buying AI tools. Their managers are being left behind

With partners, the briefing conversation has shifted too. I now ask who the named reviewer is on any deliverable, and whether outputs will be checked against source inputs before they reach us. It is a slightly awkward question. It is far less awkward than the conversation after a bad deck.

The decision that stays human

Here is what I keep coming back to. AI can capture, summarise and format. It cannot decide what mattered most, or what the room really meant, or what story will land with this particular group of stakeholders on this particular day. That is judgement, and judgement needs a human owner.

The tool did not ship that deck. A person did. “AI slop” is rarely a tool failure. It is a sign-off failure: someone decided the output was good enough without reading it as the audience would.

So use the tools. Let them take the notes, and even draft the first version. But before you send anything, read it as the person who wasn’t in the room, and ask whether it would still make sense to them. If you wouldn’t put your name on it, don’t send it.

—

Editor’s note: e27 aims to foster thought leadership by publishing views from the community. You can also share your perspective by submitting an article, video, podcast, or infographic.

The views expressed in this article are those of the author and do not necessarily reflect the official policy or position of e27.

Join us on WhatsApp, Instagram, Facebook, X, and LinkedIn to stay connected.

The post AI took perfect notes. Then it ruined the deck appeared first on e27.