I've been building an AI presentation tool for a while, and one thing is clear: most people use it wrong. Not because they're lazy, but because they follow the same bad habits. Here are the five mistakes I keep seeing.
Asking for a presentation instead of an argument
"Make a presentation about remote work" — this is what most people start with. The AI has no idea why you're making the deck, who will sit in the room, or what you want them to do afterward. So it produces something generic. Every slide says the same things you've already read in a LinkedIn post.
What works: tell it the context before anything else. "We're a 20-person sales team, our quota went up, and I need to convince the CEO we should shorten the sales cycle. The audience is skeptical, not technical." That's a prompt. The AI still might miss, but at least it's aiming at the right target.
Treating the first output as the final draft
I'm always surprised when someone takes the first version of an AI deck and moves it straight into their company template. The output is a draft, and not even a good one. It's a placeholder that expresses a certain structure. You have to ask follow-up questions, make it rewrite the weak sections, and cut half of what it generates.
My teammate told me last week that she asked an AI tool for a slide deck about her product launch. The first version had a slide called "Key Benefits" with three bullet points. After she pushed back, the next version had a concrete flow: problem, customer quote, before-and-after, objection handling. She used maybe 30% of the first draft.
Leaving out your own point of view
AI doesn't have opinions. It has averages. If you don't tell it what you actually think, it will give you balanced, inoffensive content that puts a decision group to sleep. The best decks I've seen are the ones where someone had a strong position and forced the AI to amplify it.
Tell the AI what the contrarian take is. Tell it who argued against you. Tell it the one thing you want people to remember. If you skip this, you're just getting a summary of everyone else's thinking.
Ignoring the narrative arc
Presentation software is great at bullets, but the best presentations feel like a story. AI can generate the ingredients, but it won't build the narrative. There's a difference between "here are facts" and "here is why this matters to you."
I usually take whatever the AI produces and reorder the slides completely. I also ask it to print the point of each slide as a full sentence, not a label. When every slide can be read as a sentence, the deck has an argument. If it can't, you've just collected information.
Not rewriting the voice
AI-generated text has a tell. Words like "leverage," "streamline," "unlock" — not necessarily bad, but they're empty. You need to replace at least half of the generated sentences with something that sounds like you. Otherwise people will notice immediately. They may not know it's AI, but they'll know it feels dead.
While building Huanjian, we found that the teams who get the most value are the ones who treat AI as a starting point for a conversation, not as a ghostwriter. They edit for rhythm. They add their own examples. They delete the conclusion and write their own.
I still make the first mistake sometimes. The fix is to treat the AI like a new colleague who is too polite to ask clarifying questions. You have to do the thinking about what matters, and let it handle the grinding. That's the only way I've seen it work.