A board deck that took two minutes to generate shouldn't take two seconds to approve. That's roughly the gap a growing number of general counsel and company secretaries are now trying to close — not by banning AI outright, but by drawing a very specific line around which slides an AI is allowed to write and which ones it isn't.

It's tempting to read this as a 2026 phenomenon, a knee-jerk reaction to slide generators getting good enough to fool a room. It isn't new. It's the same argument Jeff Bezos made about PowerPoint itself, twenty-two years before AI slide tools existed — and understanding why he made it explains why the current caution isn't paranoia.

The original version of this problem had nothing to do with AI

In 2004, Bezos sent Amazon's senior team an email banning PowerPoint from executive meetings. The replacement was a six-page narrative memo, read in silence for up to thirty minutes before anyone spoke. His stated reasoning, repeated in interviews for two decades since, was that slide decks let a presenter gloss over gaps in logic — bullet points flatten the relationships between ideas and let charisma substitute for rigor. A memo doesn't have that escape hatch. If the argument doesn't hold together in full sentences, it falls apart on the page before it ever reaches a meeting.

That policy has outlasted Bezos as CEO, which is itself a data point — it wasn't a founder quirk, it was a structural fix for a structural problem. And the problem it fixed is exactly the one AI slide tools are now capable of reintroducing at scale.

AI doesn't just speed up slide-making — it speeds up the part Bezos was trying to slow down

Here's the part that gets missed in most "how to prompt better decks" advice: the danger was never that slides look bad. Polished, professional-looking AI output is precisely what makes this version of the problem harder to catch than the 2004 version. A rushed, hand-built slide with clunky bullet points at least looks unfinished, which invites scrutiny. A clean AI-generated slide with a confident headline and a plausible chart invites the opposite — a nod instead of a question.

One useful way to think about this comes from a board-deck guide published by 2Slides, which frames a board deck as a fiduciary artifact rather than a marketing document — the primary evidence directors use to exercise a legal duty of oversight. Their practical breakdown is worth sitting with: roughly 80% of a typical 20-slide board deck is structured reporting — the kind of thing AI genuinely handles well, given clean data and a solid template. The other 20% — the strategic ask, the risk slide, negotiated financial terms — is where founder or executive judgment has to live, because a single wrong adjective in a phrasing can change how a board reacts to a decision. AI can summarize what happened. It can't yet be trusted to decide how to frame what should happen next, or to own the consequences if that framing misleads.

There's also a trust cost that compounds slide by slide

A piece circulating among presentation consultants this year made an observation worth stealing: the first AI-flavored slide in a deck doesn't just read as one weak slide — it becomes the slide the audience uses to calibrate everything that follows. Once a board member spots one slide with the telltale AI fingerprint — the interchangeable icon set, the bullet points that don't actually connect to each other, the chart that looks generated rather than pulled from a live system — they don't just discount that slide. They start reading the rest of the deck more skeptically, hunting for the next tell instead of engaging with the argument.

That's a genuinely different failure mode than "the slide looks unprofessional." A slide that looks unprofessional costs you polish points. A slide that reads as machine-generated costs you the presumption of good faith for the rest of the meeting — and in a boardroom, where the entire point of the deck is to support a decision under fiduciary scrutiny, that presumption is the thing you can least afford to lose partway through.

What companies are actually restricting — and it's narrower than "AI slides"

It would be neat to say a wave of named companies have formally banned AI-generated slides in board meetings. That's not quite what the record shows, and it's worth being precise about the difference. What's actually documented is governance guidance — from firms like White & Case and Clifford Chance — recommending that companies classify meetings by sensitivity and restrict AI tools specifically for privileged, strategy-heavy, or judgment-laden sessions, while leaving routine reporting largely untouched. A 2024 poll by GC100, the representative body for general counsel across the FTSE 250, found genuinely mixed views on how far AI belongs in the boardroom at all — not consensus, and not a ban.

So the honest framing isn't "companies have banned AI slides." It's that a growing number of governance teams are converging on the same distinction 2Slides landed on independently: AI is fine for the reporting layer of a board deck, and explicitly off-limits for the layer that carries legal, strategic, or interpersonal weight. That's a narrower, more defensible position than a blanket ban — and arguably a more useful one, because a blanket ban usually just gets ignored by whoever is finishing a deck at 11pm the night before.

Why the narrower version is still the right call

I'd push back gently on the framing that this is AI aversion. It isn't — it's the same discipline Bezos was enforcing with a memo requirement, applied to a new tool. The narrative-memo policy never argued that written documents are inherently more honest than slides. It argued that removing the shortcut forces the thinking to happen somewhere visible, before the room has to react to it. Restricting AI on the judgment-heavy 20% of a deck does the same thing: it doesn't remove AI's speed advantage on the reporting slides, it just refuses to let that speed advantage extend into the parts where the whole point is that a human had to actually think it through.

Where this gets tactically useful is in how you build the deck itself. Structured slides — timelines, market maps, roadmap views — are exactly the category where a real design foundation beats a from-scratch AI layout, because the visual logic is already solved and you're not asking a generator to invent hierarchy on the fly. A pre-built timeline template handles the milestone spacing and phase logic a board expects to see instantly, which is a different problem than getting an AI tool to reason about your strategic ask. The same goes for anything geography- or market-driven — a structured map layout communicates regional exposure or expansion plans in a way a director can verify at a glance, rather than having to trust that an AI-generated graphic reflects real underlying numbers.

The line worth drawing

If you're building your own policy instead of borrowing one, the question to ask per slide isn't "did AI touch this," it's "if this number or claim is wrong, who is accountable for catching it before the board sees it." Slides where the answer is "the reporting system, verified by finance" are fine to let AI draft. Slides where the answer is "whoever wrote the sentence" are the ones that need a human hand on them the whole way through — not because AI writes badly, but because that's the one job a board deck was never supposed to outsource in the first place.