Skip to content

Type something to search the manual

    Slides and presentations

    ~ min read

    30-second summary
    • A good presentation isn’t a projected document. The AI default produces text-heavy slides: you need the right instruction to flip it.
    • From the report to the outline: “narrative thread in N slides, one idea per slide, 30 words max per slide”. Rigid instruction, usable result; without the constraints, AI paraphrases the document and you end up reading it on screen.
    • Tight bullets: explicit constraint, “max 4-6 bullets of max 10 words each, no third-person verbs”. Without the constraint, verbal inflation kicks in.
    • Speaker notes: what you’ll say aloud while the slide is projected. You use them to prepare, not to read.
    • What NOT to ask: graphic design, colors, animations, fonts, templates. That’s the job of your slide tool (PowerPoint, Keynote, Google Slides) or a designer.

    A good presentation isn’t a projected document. Opening a slide full of words is an implicit choice: the audience has to either read or listen, not both. In almost every case they stop listening and read, badly. Slides that work well support what you say, they don’t duplicate it.

    With AI you go wrong in a specific way: you ask “make the slides from the report” and AI paraphrases the document into smaller paragraphs. You get a PowerPoint file, but you’ve lost the function of slides. The fix isn’t to change the tone, it’s to change the instruction: tell it what not to do (paraphrase, write paragraphs, use long verbs) and what you want (one idea per slide, counted words). The rest (layout, colors, template) stays out: AI doesn’t have what it takes there.

    The most common starting point is a report or long document you need to present. Temptation number one: copy-paste bits of the document into slides. Temptation number two: ask AI “make the slides from this”. Both produce text-heavy slides, because the source document was made to be read, not projected.

    The prompt that breaks the pattern is rigid: “extract the narrative thread from this text in ten slides, one idea per slide, 30 words max per slide. For each slide give me a short title (six words max) and between two and four bullets. No paragraphs.” Three constraints in a row (slide count, total words, no paragraphs) force the synthesis. What you get is the narrative structure of your document, not the document shrunk.

    Two decisions sit outside the prompt and you make them. How many slides: depends on the presentation time, not the document length. Handy rule: about one slide per minute for a 15-20 minute discursive talk, fewer if the audience wants to discuss, more for data-heavy slides. What to feed in: if the document is long (more than five or six pages), don’t paste all of it. Summarize it first (the pattern from Summarize a long document) or pull the key sections, and feed AI only those. A whole document forces AI to fill more than it should, and the outline comes out diluted.

    The number 30 isn’t magic, but it’s useful: keeps the slide readable from five meters away and forces synthesis. You can move it between 20 and 40 depending on density (less for a pitch, more for technical slides). Under 20 it goes cryptic, over 40 it turns back into a document in disguise.

    “One idea per slide” is the constraint that does the heaviest lifting. Without it, AI packs three concepts into the same slide to save on the total, but a slide isn’t a tool for efficiency, it’s a tool for attention. One thing at a time.

    After reading the outline, you critique it with the same move as Drafting professional documents: cut the slides that add nothing, reorder, insert the missing one. Ten proposed slides often become seven or eight after the cut, and those are the right ones.

    The outline gives you the title and two to four bullets per slide. The next step is making them readable in projection. Here AI goes wrong in the opposite direction from the rest of the lesson: it writes too much. Bullets of 15-20 words, impersonal third-person verbs, parallel phrasing (“it has been observed that…”, “one notes that…”, “it emerges that…”). Those are paragraphs chopped into sticks, not bullets.

    The prompt that works is restrictive and numerical: “rewrite these bullets as max four bullets of max ten words each, no third-person verbs, telegraphic style. Numbers and proper names stay; the rest gets trimmed.” The two constraints (count, length) are non-negotiable; the style (“telegraphic”) gives direction; the note on numbers protects data from paraphrase. The “no third-person verbs” is the constraint that does the most work: strip the impersonals (“it has been observed”, “one notes”), and the paragraph collapses into the parts that actually belong on a slide, which are numbers and names.

    If the result is still long, second round: “shorter: max six words per bullet. Split into two bullets if needed.” Two rounds almost always get you there.

    A quick test: read the bullet out loud and count the words on your fingers. If you hit ten before running out of sentence, it’s long. If you can say it in one breath, it’s fine.

    The bullet pins the idea; the words you say out loud are different. In PowerPoint, Keynote, and Google Slides there’s a speaker notes field under each slide: that’s where you write what you’ll say while the slide is projected. The audience doesn’t see it; you do, if you want.

    AI writes them well if you hand it the slide’s bullets and the context of who’s listening: “for the slide with these bullets, give me two or three sentences of speaker notes. Audience: [CEO / technical team / external client]. Tone: [more formal / more direct]. No word-for-word repetition of the bullets.” You pass one slide at a time, not the whole deck in one go: same pattern from Drafting professional documents, asking for everything at once thins out the quality.

    That last constraint is where the difference shows. Without it, AI writes back what’s in the bullet with a few words more. Useless: in projection, saying out loud what’s written is what empties a presentation fastest. Speaker notes are for saying something else: the reason behind the number, the anecdote, the transition to the next slide.

    You use them to prepare, not to read. If when presenting you end up reading the notes word for word, you need a real script, not speaker notes, and probably a smaller audience or a recorded call without one. That’s rarely the case: better to rehearse aloud and carry them as a safety net.

    You have a paragraph from your Q2 report you want to turn into a slide titled “Q2 Results”:

    In Q2 we posted revenue of 1.2 million euros, up 12% from Q1 but still 5% below annual budget. Delivered projects were 23, in line with Q1. Hours over quote rose 18%, mainly on client Alfa.

    The generic prompt (“give me the bullets for the slide”) produces long bullets that just rephrase the text:

    • In Q2 we reached revenue of 1.2 million euros, up 12% on the first quarter
    • The result is still below the annual budget, with a 5% gap
    • Projects delivered in the quarter were 23, in line with Q1
    • Hours over quote went up 18%, mostly on client Alfa

    Same text, only broken up. Useless. Change the prompt:

    You
    Rewrite these bullets as max four bullets of max ten words each, no third-person verbs, telegraphic style. Numbers and proper names stay. - In Q2 we reached revenue of 1.2 million euros, up 12% on the first quarter - The result is still below the annual budget, with a 5% gap - Projects delivered in the quarter were 23, in line with Q1 - Hours over quote went up 18%, mostly on client Alfa
    AI
    - Q2 revenue: €1.2M, +12% vs Q1 - Annual budget: still -5% - Projects delivered: 23 (level with Q1) - Hours over quote: +18% (client Alfa)

    Same content, half the words, and a slide you can read in three seconds. You tell the story out loud: “we grew 12% but still under budget; project count holds but hours over quote are up, and the weight is almost all on one client.” That’s what goes in the speaker notes, not in the bullets.

    Graphic design and color choices. A prompt like “what palette would I use for corporate slides of a consulting firm?” gets you generic advice that has nothing to do with your template. Your brand already has a palette; if it doesn’t, you need a designer, not an AI. A few seconds of your own thinking about the template you usually use is worth more than any generic answer.

    Animations and transitions. Same story, worse. Animations rarely help (almost never), and when they do, it depends on the specific slide. AI will suggest “fade in each bullet one by one” even when it makes the presentation worse. Rule of thumb: if you don’t have a specific narrative reason to animate, don’t animate.

    Fonts. The font is set by your company template, or by the template you pick in PowerPoint or Keynote. AI doesn’t know it and will suggest some random one. Skip that question.

    In general, if the question starts with “what should it look like…”, you’re asking the wrong thing. AI is strong on content, weak on visual form. For design, use your slide tool or a designer.

    If you don’t have any company template at all (freelance, tiny outfits, first deck ever), start from a ready-made template in PowerPoint, Keynote, or Google Slides, not from an “aesthetic suggestion” from AI. Those templates were designed by actual designers and are a better starting point than any generic palette.

    From slides the module moves to numbers: spreadsheets, CSVs, small analyses. The next lesson, Working with data and tables, opens that chapter with a particular eye on company data privacy.