I build presentation tools for a living. So when AI writing tools started flooding the market, I was skeptical. Most of what I saw looked like a robot had vomited bullet points onto a template. Then last month I had to prepare a strategy review for our investors, and I decided to actually give it a real shot. Here's what happened.
The first output is always a template zombie
I asked an AI tool to create a 10-slide deck on our product roadmap. What came back was structurally perfect and intellectually dead. Every slide had a neat title, four bullet points, and absolutely nothing worth hearing. "Key insight: we need to focus on customer feedback." No kidding.
I've seen this problem from both sides now. As someone who reviews countless decks, I can spot an AI-generated slide from across the room. The writing is too balanced. Every argument has a counter-argument. Every risk has a mitigation. It's like reading a press release written by a committee of really polite people.
My colleague Lena summed it up when she saw my first attempt. "This looks great," she said, "and I don't believe a single word of it." That stung because it was true.
What actually works: using AI as a thinking partner
The turn came when I stopped treating AI as a presentation writer and started treating it as a sparring partner. Instead of asking for a deck, I asked for pushback. Here's my draft narrative, I told it. Tell me where the logic breaks. Give me three reasons my growth projection is optimistic. That was genuinely useful.
Last week I was preparing a client review for a manufacturing company. I had a slide claiming we'd reduced their manual reporting time by 40%. Before I put it in the deck, I asked the AI to interrogate that number. It asked whether the baseline was measured before or after we'd already automated a few tasks. I checked, and yes, the baseline was crooked. That one question saved me from a very uncomfortable meeting.
My take is that AI is best at the parts I don't enjoy thinking about: challenging my assumptions, finding missing logical steps, suggesting alternative structures. It's useless at the part that matters most, which is having a real opinion based on real context.
What I've started doing is writing my rough outline first, often just a few messy bullets. Then I ask the AI to suggest two or three different narrative flows through that material. Then I ignore most of its suggestions and keep maybe ten percent. That ten percent is usually a transition I wouldn't have thought of or a way to frame the problem that makes more sense to my audience.
While building Huanjian, we've found that people don't actually want AI to write their presentations for them. They want AI to help them structure their chaotic thoughts. The prompt matters more than the model. "Write a deck for me" gives you the template zombie. "Here are my messy thoughts, help me find the story" gives you something you can actually use.
Where I've landed
I still write my own slides. The words, the examples, the numbers that make my argument convincing. But I now use AI every time I present to ask one question: what am I missing? It's not that the AI knows anything I don't. It's that it forces me to articulate my reasoning clearly enough to be challenged.
Honestly, I'm still figuring out the right balance. There are days I use AI to draft a full speaker's note and then edit the hell out of it. There are other days I go completely manual and don't touch any AI tool. The difference seems to be whether I already know the story. If I do, AI helps me refine it. If I don't, AI just gives me confident nonsense.
I've stopped worrying that AI will replace presenters. It won't replace anyone who has something to say. It will replace the people who had nothing to say anyway and were just filling slides with words. That's a trade I'm willing to make.