Gains:
- Ability to convert slide titles from subject tag to action title (claim) and reduce the text in a readable form
- Ability to make a slide less of a document by rounding and contextualizing numbers and moving explanations to the speaker note
- Ability to explicitly prohibit the model from fitting numbers and verify each number with its source
Once the backbone is ready, it's time for the slide itself: title, bullet points, numbers, and what the speaker will say. The biggest pitfall here is turning the slide into a document. The audience cannot listen to you while you read the slide. In this unit, we will learn how to use AI to shorten, sharpen, and break down slide text into speech. The principle is this: the slide shows, the speaker tells.
A slide is not a document
Three basic concepts. The title is the one-sentence claim of that slide; It's not just a hashtag. A bullet is a short line of evidence that supports the claim. The speaker note is the text that is not written on the slide but that you will say. A good slide has less text and more speaker notes.
A practical limit: usually no more than 3-5 items on a slide, each item on a single line. A rough measure known as the “6x6 rule” works: ~6 words per line, no more than ~6 lines per slide. The goal is readability, not rules.
Tip: After preparing the slide, step back 10 feet and look. If you squint to read, the text is too much and the font is small.
Action title: biggest leverage
Changing the slide title from “topic” to “claim” is the single change that improves presentation quality the most. This is called an action title: it says what the slide says in full sentences.
Weak (topic)
Strong (action title)
Sales results
Sales have been growing for 3 consecutive quarters
Budget
The current budget is not enough for the end of the year
customer satisfaction
70% of complaints come from a single step
Conclusion
We recommend rolling out the pilot across the region
Artificial intelligence is very good at generating action titles, because compressing your ingredients into a single claim is a linguistic task:
Find the ONE claim summarized by the slide items below and write it as an ACTION HEAD (full sentence, 10 words max). I want the claim ("Sales are falling"), not the topic tag ("Sales"). Give 3 alternatives. Articles: [...]
Step by step: generating the text of a slide
- Write the claim of the slide (action title).
- Choose no more than 3-4 items of evidence; each supports the claim.
- Shorten the clauses: start the verb, remove unnecessary words.
- Round the number and put it into context: "≈1.3 million TL, +12% over last year" instead of "1,284,902 TL".
- Separate speaker note: comments not on the slide here.
- Verify: does each number and claim match its source?
Your role: presentation copy editor.Action title of this slide: [ ... ]Raw content: [scattered notes / paragraph]Task:1) Output up to 4 items of evidence, one line each, verb first.2) Round long numbers and add context (like % of change), but DO NOT make up the number not in the data.3) Write comments that will not go on the slide in a separate "Speaker's note" section (3-4 sentences).
To expand the speaker note:
Turn the following 4 items into a fluent SPEAKING text that I will say on stage (about 45 seconds, first person, natural). Narrate the items instead of reading them word for word. Finally, add a TRANSITION sentence to the next slide. Articles: [...]
To simplify the language:
Simplify this slide text for a non-technical audience. Remove jargon or explain it in one sentence the first time it occurs. DO NOT CHANGE meaning, KEEP numbers. Text: [ ... ]
three mini cases
Case 1 — Title change. One consultant used subject headings on all 14 slides (“Analysis”, “Findings”…). He turned them all into action titles with artificial intelligence. In a presentation he made without touching the slides, just changing the titles, the client said, "For the first time, I understood what he meant from start to finish."
Case 2 — Text diet. One engineer used an average of 68 words per slide; The audience was not reading the slide and listening to it. He told the artificial intelligence to "reduce each slide to 20 words maximum and move the rest to the speaker note." Slide word count decreased from 68 to 19; the excess went to the speaker note, the presentation became much smoother.
Case 3 — Catching the fabrication. One intern told the AI to “power these items up”; The model automatically added a number like "35% increase" even though there was no such number in the data. The intern had to spot the mistake because he didn't set the "[confirmation]" placeholder rule. In his subsequent requests, he made the instruction "generate numbers that are not in the data, write [confirmation]" the standard.
Weak prompt / Strong prompt
Weak:
Make this slide better.
"Good" is undefined; The model can fit random ornaments, even numbers.
Strong:
Your role: presentation copy editor. Edit this slide to 'less text, stronger claim'. Write an action title (≤10 words), leave a maximum of 4 single-line items, round long numbers and add context, and move the rest to the speaker note. Adding numbers that are not in the data. Raw slide: [ ... ]
The second translates “good” into measurable rules and prohibits fabrication.
Parallel structure: putting items into the same rhythm
The viewer's eye scans items written in the same format much faster. This is called parallel structure: items start with the same grammatical pattern. For example, they all begin with a verb (“We increased…”, “We decreased…”, “We accelerated…”) or they all begin with a number. Mixed patterns (one verb, one noun, one question) slow down reading and look messy.
Weak (mixed structure)
Strong (parallel structure)
cost dropped
We reduced the cost by 12%
speed issue
We reduced the delivery to 3 days
Are customers more satisfied?
We increased satisfaction by 8 points
Artificial intelligence is very practical at bringing dispersed matter into a single pattern:
Put the following items in PARALLEL structure: they all start with the same pattern (e.g. they all start with a verb and, if possible, a number). Do not change the meaning, keep the numbers, write [confirmation] what is not in the data. Items: [ ... ]
Tip: If items are different lengths on a slide, put the longest at the bottom and pull them all to a similar length. When the eye sees a neat list, it trusts the content more.
How many numbers should be on a slide?
A slide is not a table. Instead of fitting 12 issues on one slide, choose 1-3 issues that support that slide's single message. The remaining detail goes into an appendix or speaker note. The viewer can only remember one or two numbers on a slide; the rest is noise. If you really need to show a table, highlight the cell in the table you want to describe with color or frame; Let the viewer know where to look.
Common mistakes
- Writing the sentences on the slide and reading them: The audience reads, not listens to you.
- Staying on topic: An unpretentious title leaves the presentation bland.
- Raw, long numbers: Use rounded and contextualized numbers instead of "1,284,902.47".
- Leaving the speaker note blank: Cramming everything into the slide that doesn't fit on the slide.
- Relying on the number the model adds: When you say "reinforce" the model can fit; Delete the number that does not have a source.
In summary
The slide shows, the speaker explains. Reduce the text, turn the headlines into action titles (claims), leave 3-4 short bullet points at most, round the numbers and put them into context, and move everything else to the speaker note. AI is very powerful at this job of shortening and sharpening; but expressly prohibit fabrication of numbers and verify every figure with its source.
Application task
Take a full slide in your hand. (1) Turn the topic title into an action title (have 3 alternatives generated). (2) Reduce text to no more than 4 bullet points and ~20 words per slide. (3) Turn your extracted comments into a speaker note. (4) Mark the source of each number on the slide.
checklist
- [ ] The slide title is a claim (action title), not just the topic.
- [ ] There are at most 3-4 short items on the slide.
- [ ] Numbers are rounded and contextualized.
- [ ] Explanations that did not fit on the slide were moved to the speaker notes.
- [ ] I verified all the numbers added by the model with its source.
- [ ] I added the "Generate a number that is not in the data" instruction to the request.