AI Prompts for Presentations

1. State the goal, not just the topic. "Write about our Q3 results" gives a model nothing to aim at. "Convince a skeptical VP that Q3 retention numbers justify more budget for the support team" gives it a goal, and a goal produces an argument instead of a summary. Every strong prompt starts with what you want the reader to believe or do, not just what the deck is about.

2. Name the audience and what they already know. A prompt aimed at "our sales team, who already know the product cold" produces different content than one aimed at "a prospect who has never seen a demo." Skipping this is the single biggest reason AI-drafted slides come back too basic or too technical — the model is guessing at a reader you never described.

3. Specify the structure you want, not just "a presentation." Ask for a slide count, a section order (problem, evidence, recommendation, next steps), or a format ("one headline and three bullets per slide"). Structure is the part AI delivers reliably when you ask for it explicitly, and guesses at badly when you don't.

4. Put a real constraint on length and tone. "Keep each bullet under eight words" or "headlines should read as a claim, not a topic label" gives the model something concrete to check its own output against. Constraints are what separate a usable first draft from a wall of AI-flavored paragraphs you have to cut down yourself.

Drafting full slide content before you have settled on a structure wastes the most effort of any step, because a rewritten headline or a reordered section can make the bullets underneath it obsolete. Outline first, and keep the first pass to headlines and a one-line note on the evidence, not finished copy.

A prompt that works well: "Create a slide-by-slide outline for a 12-slide presentation on [topic] aimed at [audience]. Give me one claim-style headline per slide and a one-line note on what evidence goes underneath it. Do not write full bullet content yet." The constraint at the end matters — without it, most models draft full paragraphs on the first pass.

If you are starting from raw notes instead of a blank page, try: "Turn these notes into a presentation outline: [paste notes]. Group related points into slides, give each slide a headline, and order the sections as problem, why it matters now, options, recommendation, next steps." This works whether the notes are bullet fragments, a meeting transcript, or a messy first draft.

Once you have an outline, interrogate it before moving on. "Slide 4 and slide 7 make almost the same point — suggest how to combine or differentiate them" or "Which of these ten headlines is the weakest argument, and why" are both prompts that catch structural problems while they are still cheap to fix.

Asking for an entire deck's worth of content in a single prompt is the most common mistake in this stage. Models tend to spread their effort thin across all the slides at once, and the result is copy that feels uniform and slightly generic on every slide, because none of them got the model's full attention.

Draft one slide at a time instead: "Write the content for a slide titled '[headline]'. Give me 3-4 bullets, each under ten words, using these numbers: [data]. No adjectives, no filler." Handing over the real numbers or facts up front, rather than asking the model to find or invent them, is what keeps the output specific instead of vague.

When you have real source material, anchor the model to it directly: "Draft three bullets that support the claim '[headline]', pulling only from this paragraph: [paste source text]. Do not add any fact that isn't in the source." This single constraint — do not add facts that aren't in the source — is the difference between a safe first draft and one you have to fact-check line by line.

Always verify what comes back. AI drafting tools will confidently produce a plausible-sounding number or quote if the prompt leaves room for it, and a fabricated statistic on a slide is far more damaging than a slightly dull one. Check every figure and name against your source before the draft moves to the next stage.

A general instruction like "make this slide better" gives the model no way to know what "better" means, so it tends to rewrite everything, including the parts that were already fine. Naming the specific problem — vague headline, too long, too much jargon — keeps the edit targeted to what actually needs fixing.

For a weak headline: "Rewrite this headline as a specific, falsifiable claim under eight words: [headline]. Do not use vague words like 'strong,' 'significant,' or 'improved' without a number attached." For bloated bullets: "Tighten this into three parallel bullets of six words or fewer, keep every number exact: [paste bullets]."

For jargon, split the task into two prompts instead of one: "List every term in this slide a smart outsider would not recognize" first, then "Rewrite only those flagged terms in plain language, leaving the rest of the sentence untouched." Asking for the audit and the rewrite together tends to over-simplify text that was already clear.

Two things to check on anything a model rewrites: hedging words that crept in ("may," "can help," "in many cases" — cut them from a slide meant to state something with confidence), and exact numbers that got rounded off or softened into a vague descriptor during compression. Both are common side effects of a tightening pass.

1. Outline the whole deck first, headlines only, using the outline prompts above. Read it end to end before drafting a single bullet, since it is far cheaper to reorder or cut a headline than to rewrite finished slide content later.

2. Draft one slide at a time, feeding the model the outline for context plus the specific fact or source text for that slide: "Here is the full outline: [paste]. Now draft slide 3: headline '[X]', 3 bullets, using this data: [data]."

3. Fix the slide that is actually weak, directly: "Slide 5's headline is vague — rewrite it as a specific claim using this fact: [fact]." Targeted fixes on the slide that needs it beat re-running the whole deck through a generic "improve this" prompt.

4. Check the finished set for consistency: "Here are the headlines for all 12 slides: [list]. Flag any that are inconsistent in tense, tone, or level of specificity compared to the rest." This last step catches drift that is nearly invisible slide by slide but obvious once you see the full list.

A chat thread is a good place to draft; it is a bad place to build. Once you have prompted your way to an outline and slide content across a dozen back-and-forth messages, the draft is scattered across a conversation, not sitting in one document you can scroll, reorder, and compare against itself.

Eazy is built content-first for exactly that handoff. Paste your AI-drafted outline and bullets into Eazy's block editor — headings for claims, bullets for evidence, toggle lists for detail you want available but not front and center — and you have the whole deck in one place instead of one prompt reply at a time. If you started from a source document instead of a blank chat, Eazy also reads in a PDF, Word doc, PowerPoint, Excel/CSV file, or a web link and turns it into that same editable content directly.

From there, refine by chatting with Eazy in plain language — "tighten this headline," "cut the jargon in section two," "reorder these three slides" — and because Eazy already has your whole document in context, edits stay consistent across slides instead of drifting the way a series of disconnected chat prompts can. Change one line and only that slide rebuilds, so a fix on slide 9 doesn't disturb the eleven slides you already approved.

Once the words hold together, every slide is designed for you and on-brand by default, and you can apply a theme to restyle the whole deck in one click, then export to PDF or PPTX. The prompting got you a draft; the editor is where it becomes a deck.