Your presentation looks AI-generated. Does that matter?

You’re reading The AI Directive — practical AI intelligence for business leaders. Subscribe here if someone forwarded this.

My organisation is wrestling with a new PowerPoint problem. AI-generated presentations are becoming recognisable: generic imagery, a suspiciously perfect structure, repeated layouts and language that sounds polished without sounding like the person presenting it.

The slides may be perfectly competent. Yet the audience can still be left asking the question that really matters: Did you think this through, and are these genuinely your conclusions?

One response is to take the AI draft and make it look more human. Change the images, disrupt the layout, rewrite the headings and remove the familiar tell-tale phrases. The deck becomes less obviously machine-generated, but if the underlying argument has not changed, that is cosmetic work rather than better thinking. At worst, it is whitewashing the role AI played without fixing the reason the audience felt uneasy.

The opposite response is to accept that AI is now a legitimate productivity tool. If somebody provides meaningful data, explains the audience and objective, challenges the output and owns the conclusion, why should the fact that AI assembled the slides make the content less valid?

That is the question organisations now need to answer.

The recognisable AI aesthetic is real, but it points to something deeper than design. AI tends towards patterns that are plausible, balanced and familiar. Used with a generic prompt, it can give different people similarly neat structures and similarly reasonable language.

Research conducted with 758 Boston Consulting Group consultants illustrates both sides of this. For tasks within AI’s capability, consultants using GPT-4 completed more work, worked about 25% faster and produced results rated more than 40% higher in quality. At the same time, their ideas showed less variation than those produced without AI.

That is a useful description of the presentation problem. AI can raise the average quality of the first draft while also pulling everyone towards the same visual and intellectual middle. The polished sameness is not evidence that the content is wrong. It is evidence that the tool has done more of the shaping than the individual.

Changing a font or replacing a stock image does not solve that. The deck needs a point of view.

There is still a strong case for removing the obvious AI signals, but the purpose should not be to conceal the tool. It should be to make the presentation fit the organisation, the audience and the decision it is meant to support.

Microsoft’s own response is instructive. Its guidance for keeping Copilot presentations on-brand tells organisations to give AI representative sample slides, layouts, imagery, data visualisation and content density so the system learns how they actually communicate. Without that context, Microsoft says Copilot can fall back to generic layouts and minimal formatting.

More recent Microsoft guidance on Brand Kits makes the division of labour clearer: use AI to remove the repetitive effort of finding templates, fixing fonts and applying brand rules, then spend the saved time shaping the story, anticipating questions and deciding whether each slide communicates the point.

That feels like the right answer. Do not spend hours disguising a generic AI deck after the event. Give AI better ingredients and clearer constraints before it starts, then use human time on the parts that establish ownership.

Those parts are not mysterious. The presenter should define the decision or outcome the deck is meant to support, provide the relevant source material and data, choose the argument, test the awkward assumptions and remove any slide they could not defend without reading it aloud. Specific examples, uncomfortable trade-offs and clear recommendations will do more to make a presentation feel human than decorative imperfection ever will.

Organisations should not create an informal penalty for anyone whose deck looks as though AI helped produce it. That would reward people who are better at disguising the tool rather than people who use it well. It would also waste the productivity gain by forcing employees to recreate work manually for appearances.

The standard should be simpler: can the presenter explain the evidence, defend the argument and say what they decided? If they can, AI has accelerated production without replacing ownership. If they cannot, making the slides look more human will not rescue the work.

The visible AI style will probably become less distinctive as tools improve and organisations train them on their own templates. The management question will remain. A presentation is not credible because a human moved every box, and it is not invalid because AI did. It deserves trust when the person presenting it has done the thinking required to stand behind it.

Before approving an AI-generated deck, ask the presenter for three things: the decision it supports, the evidence behind it and the conclusion they personally own.

If those answers are clear, improve the presentation and use it. If they are not, changing the images is just tidying the problem.

This is the first of three articles exploring how organisations earn trust in AI-assisted work.

Next: AI should sharpen your thinking, not replace it.

See you Tuesday.
— Toby

These are tools I use personally. Affiliate links marked - I earn a small commission if you sign up, at no extra cost to you.

Some links in this issue are affiliate links. I only recommend tools I actually use.