Insights
On building AI that stays on-brand
Insight 01
Your AI Tool References Your Brand. It Still Can't Stay On-Brand.
Every team I talk to is already using AI to build presentations, and there are plenty of tools to help you do it. The drafting happens in minutes now — that part works.
What doesn't work is consistency. The formatting drifts into layouts you didn't ask for. Visuals that are almost-right in a way that's somehow worse than obviously wrong. And someone spends the next two hours reformatting a draft that was supposed to save time. It begs the question: is this really better than what we were doing before AI?
A few months ago a long-time client asked me to develop a new presentation template for their sales team. No problem, I've done this for them and plenty of other clients loads of times, and I love doing it. Then, the rub. They asked me to integrate the template with AI so their sales team could draft inside the presentation itself. I told them I'd look into it, and I tried out every tool and plugin under the sun. Not one of them worked the way I wanted, and none to my own exacting standards.
At first I thought it must be an input problem. Without brand context the AI is generating into a vacuum, producing the statistical average of every deck it's ever seen. But after building specialty skills and uploading asset libraries and brand standards, I still wasn't getting the results I wanted. What I learned during this process is that uploading your brand standards isn't the same as being governed by them.
The current generation of AI presentation tools will happily take your brand guidelines, your past decks, your logo, your colors. Then they generate a deck that's responsive to all of it, and it’s completely unrepeatable.
Run the same inputs twice and you get two different decks. Not because the tool failed, but because that's what it does: it reinterprets your brand from scratch on every single run. Every output is a fresh one-off. The user has no control over the result and no way to reproduce it. The brand assets are an influence on the generation, not a constraint on it.
That's the distinction that actually matters, and it's the one the tooling glosses over. Responsive to your brand is not the same as governed by it. One gives you a plausible interpretation each time. The other gives you the same reliable result every time. Only one of those is a system you can build a team's workflow on.
The fix is bounded generation, not better prompting
You can't prompt your way to consistency when the underlying tool is designed to improvise. Believe me, I tried — a lot. Consistency has to be structural, and creating structure takes work. That realization led me to a different approach: build the high-quality branded templates first — the layouts, the hierarchy, the rules that define exactly what the AI is allowed to touch. Design systems with guardrails.
The AI works inside that system. It sticks to what it's great at: drafting the content, selecting which layout fits, picking images and icons from your brand assets — and only your brand assets. Nothing it can reach for is off-brand, because the off-brand options aren't in the system.
This inverts the usual control. In a one-off tool, the AI makes the design decisions and you clean up after it. In a governed system, the design decisions are made once, upstream, by a designer. The AI is limited to choices that can't break the brand. Bounded creativity, not freeform generation.
The result is decks that come out consistent and client-ready with very little editing or reformatting afterward. Not because the AI got smarter, but because it was never allowed to wander. I call my system a Deck Engine.
Speed was never the hard part
Every AI tool optimizes for speed. Speed is the easy win, and it's the part the model already does well. Almost none optimize for repeatable quality, because quality doesn't come from the model. It comes from the system you put around it: the design thinking, the information architecture, the boundaries on what the AI gets to decide.
That work happens upstream of generation, and it's design work, not prompt work. The teams getting reliable outputs aren't the ones with the cleverest prompts or the newest tool. They're the ones whose AI is operating inside a system a designer built.
So if your decks still come out inconsistent even though your tool knows your brand, the question isn't which tool or how you're wording the request. It's whether there's a governing system underneath the generation, or just a model reinterpreting your brand one deck at a time. Until the system exists, you'll keep getting one-offs. Once it does, consistency and speed stop being a trade-off.
Building those systems is what I do. If your team is scaling AI for presentations and watching the outputs drift, that's the problem I solve. I’d be happy to talk through what a governed system would look like for your brand.
"Responsive to your brand is not the same as governed by it."
Your team already uses AI. Now make it operational.
We help teams turn AI experimentation into controlled, repeatable presentation systems that improve quality, consistency, and speed.
Portland, Oregon | Barcelona, Spain
Big Idea Design | Presentation Systems Built for AI