Ask a room full of consultants, analysts, or marketing managers whether they used AI to build their last deck, and you'll get a lot of careful phrasing. "I used some AI tools for research." "I had it help with an outline." Nobody says "AI wrote the first draft of slide fourteen," even when that's exactly what happened. There's a gap between how much AI touches a presentation and how much anyone will admit to — and that gap is the actual story, not the tools themselves.

I've watched this play out with people who build decks for a living. The AI use is real. The silence around it is just as real, and it's not really about the technology at all.

The number everyone wants and nobody has

Here's the uncomfortable part: there isn't a clean, verifiable figure for "what percentage of corporate presentations are AI-written." No company tracks that metric publicly, and most wouldn't want to. What we do have are adoption numbers from inside individual organizations, and they're steep enough to explain why a round number like "a third of all decks" feels plausible even without a single source that proves it.

McKinsey built an internal AI platform called Lilli specifically to help with tasks junior staff used to own, including building out PowerPoint slides from a prompt and adjusting tone to match the firm's style. More than 75% of the firm's roughly 43,000 employees now use it every month. That's one firm, reporting on itself, with every incentive to make the number look good. But it's also a data point that didn't exist three years ago, and it's not an outlier — it's a preview of what happens once an AI tool gets embedded into the actual software people open first, instead of sitting in a separate browser tab.

Why "I used AI for research" is doing a lot of work

One thing I've noticed sitting in on deck reviews: people are perfectly comfortable admitting AI helped them think. They're much less comfortable admitting AI helped them write. That distinction doesn't track with what actually happened during the drafting process — it tracks with what feels safe to say out loud in a room where someone might ask "so what did you actually contribute?"

The honest answer, for a lot of decks, is: structure and judgment. Which slides make the argument, what gets cut, whether the story holds up under a client's first pushback. That's real work. But it's work that happens after a draft exists, and increasingly the draft itself is the part nobody wants to claim credit for skipping.

The part everyone leaves out: AI drafts create rework, not free time

This is where the story gets less flattering for AI, and it's the detail most "AI saves you hours" pieces quietly skip. A 2025–2026 survey of roughly 3,200 executives and employees at large companies, built on Hanover Research fieldwork and cited by presentation software vendor empower, found that only 14% of employees see a consistently positive net effect from using AI in PowerPoint once the correction time is factored back in. The same analysis, drawing on Workday's productivity research, concluded that more than a third of the value AI generates gets lost again to fixing, refining, and rewriting what it produced.

That reframes the whole "AI writes X% of decks" question. A slide an AI generated and a human then rebuilt from the ground up isn't really AI-written anymore — but it also isn't purely human-written, and there's no clean line where one stops and the other starts. The 30%-ish number that gets thrown around in casual conversation is probably closer to "AI touched the first draft" than "AI's words survived to the final version." Those are very different claims, and conflating them is exactly how the silence gets started — nobody wants to overclaim credit to a tool that then got mostly rewritten, so the simplest move is to just not mention it.

Where this actually shows up in how decks get built

The teams handling this well aren't the ones hiding AI use — they're the ones who've stopped treating "did AI touch this" as the interesting question and started treating "does the argument hold together" as the only one that matters. That's not a new idea; it's the same discipline behind how consulting firms have always structured slide arguments, long before generative AI existed to draft the first pass. An AI-generated outline still has to survive the same test a human-written one does: can you strip it down to just the slide titles and read a coherent argument, top to bottom?

Where AI genuinely earns its place is in getting a rough structure on the page fast enough that a person can spend their time editing the argument instead of staring at a blank slide. Where it quietly costs more than it saves is when the output gets treated as finished rather than as a first pass — dense, generic bullet points that need the same rework a founder's first pitch-deck draft usually needs, the kind covered in what separates a fundable pitch deck from an ignored one. AI is good at filling a slide with text. It's much weaker at knowing which of those words the audience actually needs to see, which is the same judgment call that matters when you're deciding what to cut from a data-heavy slide in the first place.

The admission problem isn't really about AI

Strip away the technology angle and this is an old workplace pattern wearing a new outfit. People have never loved admitting they used a template, borrowed a structure, or had someone else do the first pass — not because it's dishonest, but because "I built this from scratch" still reads as more impressive than "I edited something that got me 70% of the way there fast." AI just made the gap between those two stories bigger and the temptation to blur it stronger.

The decks that hold up under real scrutiny aren't the ones where nobody used AI. They're the ones where someone was willing to look at the AI draft skeptically, cut what didn't earn its slide, and rebuild the argument until it could survive being read as just a list of titles. Whether that first pass came from a prompt or a blank page matters a lot less than whether anyone actually did that second part.