Skip to main content
DoneThat

AI Adoption GuideGeneralCreate

Vibe-Coding Slides

Vibe-coding slides means using AI tools to create presentation decks quickly, especially when the format is changing and speed matters more than a heavily designed slide workflow.

By Don, DoneThat’s AI coach · updated

Lock the brief before any slide exists

A usable slide draft starts from a locked brief, not from a prompt like "make me a deck." Write down audience, length, and the claims you are allowed to show. If those three are missing, you are generating slides for nobody, on no clock, with nothing true enough to put on a screen.

Audience is who will sit in the room and what they already know. A weekly leadership update for three directors who shipped the project with you is not a conference talk. Length is time and slide count together: a twelve-minute decision slot cannot carry a forty-slide narrative. Claims are the statements you are willing to stand next to. They come from the brief you already trust (a memo, a ticket, notes from a call), not from the model filling gaps.

Skip this lock and the tools still produce slides. Gamma, Beautiful.ai, Microsoft Copilot, ChatGPT, Claude, and Cursor will all give you a deck-shaped object. The object will look finished. It will also invent an audience, invent a runtime, and invent claims, because a slide generator cannot leave a hole unless you tell it the hole exists.

The same discipline shows up in adjacent work. vibe-coding websites fails the same way when you generate a page before you know who it is for. vibe scripting fails when you generate a script with no input and no output named. Slides are not special. An unlocked brief is more dangerous here because people are least willing to leave slides ugly.

If you use AI Skills or a saved prompt, put the brief in first: audience, minutes, slide budget, and a list of claims with a source next to each. "Source: the 12 Aug product memo" is enough. "Source: I think this is true" is not a claim. It is a wish.

Generate slides from claims you can point to

Once the brief is locked, generate slides that cite those claims. Ask the tool to map each slide to a line in the brief. If a slide cannot name its claim, delete it or leave it blank.

This is the opposite of "make it comprehensive." Comprehensive decks add sections the brief never asked for: market context, a history of the problem, a chart of "impact." Those sections feel responsible. They are usually unsourced. A quality draft is thinner than the generator wants to be.

You can do this in any slide format in this class. A Reveal.js HTML deck, a Gamma file, a Beautiful.ai deck, a PowerPoint file shaped by Microsoft Copilot, or a markdown outline from ChatGPT or Claude that you paste into slides are all the same job. Cursor can hold the brief next to the deck source if you prefer HTML or markdown slides. The format is a delivery choice. The quality rule does not change: every slide traces to a cited claim.

One working pattern: paste the brief, then instruct the assistant to output a slide list where each item has a title, the claim in one sentence, and the brief line it came from. Reject any slide whose citation is "generally known" or "typical for this industry." Then generate the visual deck from that list only.

AI assistants are useful here because they can hold the brief in context while they draft. They are not useful as a substitute for the brief. If the model has to guess the audience, stop and lock the brief again.

Empty is correct when the brief has no number

If the brief has no claim to put on a slide, the slide stays empty. Do not invent a statistic to complete a layout. Do not invent a statistic to feed a chart. The empty box is the quality signal.

Inventing a number for a chart is the failure mode that looks the most professional. A chart wants data. Slide tools want the page to look designed. The model will fabricate a series that "illustrates the point." You then present a picture of a fact nobody measured. The audience remembers the picture. You own the number.

Leave the chart stub in if you need a reminder to fetch the real figure. Label it "number not in brief" or leave the axis blank. Do not put placeholder values that resemble real data, such as round thousands or tidy percentages. Placeholder values become live values under time pressure.

The same rule applies to quotes, customer logos, and competitive tables. If the brief does not name them, they do not appear as facts. You can keep a slide titled "Open questions" with bullets that are questions. Questions are not claims.

You have twelve minutes with three directors. The brief says keep the beta waitlist closed this week, two named customers asked for a self-serve invite, and support volume is up, with no count attached. A generator will often add a "Support trend" bar chart with invented weekly totals, plus a conversion rate for opening the waitlist. Neither number is in the brief. The draft that cites the brief has a decision slide (waitlist stays closed), a slide that names the two asks, and a slide that says support volume is up with the number field empty. That last slide looks unfinished. It is the honest one.

Treat the first deck as disposable format

Speed matters more than a heavily designed slide workflow when the format is still changing. You might need Reveal.js for a web-native talk this morning and a conventional slide file for a customer call this afternoon. You might throw the first structure away after the dry run. That is the point of generating a draft quickly.

Do not treat the first output as the live deck. The first output is a proposal for structure and wording. It will over-title, over-claim, and over-decorate. It will also pick a visual system that may not match the room. Changing format is cheaper while the deck is still a draft.

Reveal.js and other HTML slide setups sit in the same class as Gamma, Beautiful.ai, and Copilot-built PowerPoint. HTML is handy when you want the deck in version control, or when you are already in a coding tool such as Cursor. It is not more truthful. An HTML slide can invent a number as fast as a designed template. If you can see the source, search it for digits that did not come from the brief.

When the format keeps moving, regenerate from the same locked brief rather than from the last deck. Editing a previous draft compounds leftover claims. Regenerating from the brief drops slides that no longer have a citation.

Presenter edit is the quality gate

The presenter still owns what they show. Read every slide out loud against the brief. Cut anything you cannot defend in the room. Fill blanks only with facts you added to the brief on purpose. Then present from the edited deck, not from the generator's first pass.

A common skip: you generate, you glance, you share screen. The invented chart is still there. Treating the draft as the live deck is how vibe-coded slides become a quality problem instead of a speed gain.

Edit for speech, not for completeness. If you cannot say the claim in one breath, the slide is doing two jobs. If a slide exists only because the template had a "next steps" layout and the brief had no next steps, delete it. If you added a number during edit, write the source in the speaker notes.

Quality here is not polish. Quality is a deck draft that cites audience, length, and claims, empty where the brief is empty, and a human who will present it after reading it.

Is this worth automating for you?

Whether this pays back depends on how much time it takes your team today. Most teams estimate that from memory, and the estimate is usually wrong in one direction or the other.

DoneThat reconstructs where the time actually went, with no timers to forget, so you can measure the baseline before committing to a project and check the gain afterward.

Measure the baseline first