I sat through six pitch decks last week. Every single one had an AI slide. Not because every single product meaningfully uses AI. Because founders have figured out that investors expect to see it, and now we’ve got a cargo cult situation where people are bolting “powered by AI” onto products the way they used to bolt “blockchain-enabled” three years ago.

It’s the same playbook. Different buzzword. And investors are starting to see through it.

The Three Types of Bad AI Slides

I’ve been categorising these for my own amusement, and they fall into predictable patterns:

The Architecture Astronaut. This slide shows a complicated diagram with boxes labeled “AI Engine,” “ML Pipeline,” and “Neural Network Layer.” It looks impressive until you ask what each box actually does, and the founder says something like “well, that’s where the intelligence happens.” There’s no specificity because there’s no substance.

The Future Tense Slide. “Our platform will incorporate AI-driven personalisation, machine learning-based predictions, and automated insights.” Every verb is future tense. Nothing has been built. The AI is aspirational, not actual. This is fine if you’re pre-seed and being honest about your roadmap. It’s not fine if you’re presenting it as a current capability.

The GPT Wrapper. The product calls an API, usually OpenAI’s, with a prompt template that took twenty minutes to write. The entire “AI” component is a few lines of code that any developer could replicate in an afternoon. It’s not that API wrappers can’t be valuable. They can, if the prompt engineering is sophisticated and the workflow around the AI is genuinely well-designed. But most of the time, it’s a thin layer over a commoditised API with no defensibility.

What Actually Impresses Investors

I talk to angel investors and VCs regularly, and here’s what they tell me they’re looking for on the AI slide:

Proprietary data. If your AI is better because you have data that nobody else has, that’s a real moat. A startup that’s trained a model on five years of Australian residential construction defect reports has something defensible. A startup that fine-tuned GPT-4 on publicly available documentation does not.

Measurable improvement. Show before and after. “Our AI reduces claim processing time from 14 days to 3 days” is compelling. “Our AI provides intelligent insights” is not. Numbers matter. Be specific, even if the numbers are early and based on a small sample.

Technical depth when asked. You don’t need to fill your pitch deck with technical details. But when an investor asks “how does your AI actually work?”, you should be able to give a clear, specific answer. If the best you can do is “we use machine learning,” you haven’t done enough work to deserve the AI label on your deck.

Honest limitations. The founders who impress investors most are the ones who say “our model is currently 82% accurate on standard cases and about 65% on edge cases, and here’s our plan to improve that.” That shows real understanding. Claiming 99% accuracy on everything is either dishonest or a sign you haven’t tested rigorously.

The Question You Should Ask Yourself

Before putting an AI slide in your deck, ask yourself this: if you removed every AI component from your product, would it still be useful?

If the answer is no, you might be building an AI company. Good. Make sure your AI is real and differentiated.

If the answer is yes, you’re building a product company that happens to use some AI features. That’s also good. But frame it that way. Your AI slide should be about how AI enhances your core value proposition, not about the AI itself.

The worst outcome is when a founder can’t articulate their core value proposition without leaning on AI terminology. That usually means there isn’t one.

A Better AI Slide Template

If you’re going to include an AI slide, here’s what it should cover in four or five bullet points:

1. What the AI does. One sentence, specific. “Our model predicts residential property maintenance costs within 12% accuracy based on building age, condition reports, and local climate data.”
2. What data it uses. Where the training data comes from and why it’s hard to replicate.
3. What it replaces. The manual process or existing solution that your AI improves upon.
4. Current performance. Actual metrics, not aspirational ones.
5. What’s next. One specific improvement you’re working on.

That’s it. No architecture diagrams. No buzzwords. Just clear information about what you’ve built and why it matters.

The Bigger Problem

Here’s my real concern. The AI slide plague isn’t just wasting investor time. It’s eroding trust. When every pitch deck claims AI capabilities, investors develop antibodies. They start discounting genuine AI innovation because they’ve been burned too many times by vaporware dressed up as machine learning.

That hurts the founders who are doing real, hard, genuinely innovative AI work. They have to fight through a wall of scepticism created by everyone else’s BS slides.

If your product doesn’t need AI, that’s fine. Most products don’t. Build a great product that solves a real problem and let your results speak for themselves. That pitch deck will be more compelling than any AI slide could ever be.

Jack Reeves is a startup founder and technology advisor based in Australia.