Writing speaker notes for every slide with AI

A webinar script written as one long document is quick to produce. It gets harder when every slide needs its own speaker notes that last exactly as long as the slide stays on screen and lead neatly into the next one. AI speaker notes can save you hours here, as long as you feed the tool the right information and test the result out loud. This guide covers how slide and spoken text work together, how long the notes for each slide should be, and what to listen for when you read them back.

Why each slide needs its own notes

A continuous script is where you work out structure and argument; see writing a webinar script and writing a webinar script with AI. Once your slides exist, a second step pays off: you break the script into sections and attach each one to exactly one slide.

That gives you some practical advantages:

• Slide changes land on time. When the notes end, the next slide appears. Nobody keeps talking over a picture that vanished half a minute ago.
• Edits stay small. If a price or an example changes, you rewrite one slide’s notes instead of a whole page. With a recorded voice-over, you only re-record that section.
• Running time becomes predictable. The word count per slide tells you, fairly accurately, how long the whole webinar will run.
• The AI does better work. Small, clearly bounded tasks give cleaner results than asking for forty minutes of text in one go.

What the slide shows and what the notes say

Slide and notes do different jobs. The slide is the anchor for the eye: one sentence, one number or one image. The notes carry what doesn’t fit on the slide, such as the reasoning behind it and an example.

The first sentence picks up the slide. When a new slide appears, people listen for the connection straight away. If your opening sentence uses the slide’s keyword, eye and ear meet. Don’t read the slide aloud, though. Your audience reads faster than you talk and will drift off.

The notes follow the reading order. If the slide shows “before” on the left and “after” on the right, talk about them in that order. If bullet points appear one by one, each point gets its own short paragraph.

Only refer to what’s visible. A line like “as you can see in the top right” only works if something is actually there. AI drafts love these phrases because they’re common in talks, so check every one.

Here’s a small example. Say you’re running a webinar on invoicing for freelancers, and the slide shows “4 hours a month” in large type.

Weak: “On this slide you can see four hours a month. Four hours a month are lost, and that’s a lot of time.”

Better: “Four hours a month. Here’s where that comes from: if you send twelve invoices a month and each one takes twenty minutes to write, chase and file, you’re there. And that’s before a single late payment.”

The better version explains the number and lets people check the maths. The slide stays simple.

How long the notes for each slide should be

Length follows from how long the slide stays up and how fast the words are spoken. In a calm presentation many people speak at very roughly 120 to 150 words per minute. Synthetic voices usually sit in a similar range and can be set a little faster or slower. The table below assumes about 130 words per minute. The timings are rules of thumb, not fixed targets.

Watch the edges. Notes of more than about 150 words keep a content slide on screen for over a minute, so splitting it usually works better, as the article on slides that work on a phone explains. Under ten words, slides flick past and it feels rushed; merge two slides or add an example.

Give the AI a word count per slide rather than a duration in seconds. Language models only roughly stick to length targets, so count afterwards.

Transitions between slides

Ask for notes slide by slide and each one often starts from scratch, opening with “In this section” and closing with a little summary. Played back to back, it sounds like twenty separate mini talks.

Good transitions use the end of one slide and the start of the next:

• The ending sets up what’s next. The last sentence asks a question, makes a promise or draws a conclusion that the next slide answers.
• The opening picks up the thread. The first sentence on the new slide connects without repeating the previous line.
• Section breaks should be audible. When you move from problem to solution, or from content to offer, the transition can be more explicit: “So that’s why it happens. Now for what you can do about it.”

Watch out for stock phrases like “moving on to”, “as mentioned earlier” or “let’s take a look at”. Once is fine. If the same phrase turns up on every third slide, listeners notice the pattern.

An example of a transition that works:

End of slide 8: “Twenty minutes per invoice doesn’t sound like much. The real question is where those minutes go.”

Start of slide 9: “Mostly into chasing. Writing the invoice is the quick part.”

Working with AI, step by step

Your preparation matters more than the tool. Here’s a process that works:

1. Build a slide list. One row per slide with its number, the text on the slide, a short description of the image, the slide type and a target word count. A simple table will do.
2. Set the frame once. Audience, tone, who is speaking (you in the first person, or a synthetic voice) and a list of facts that must be right. Add any words you don’t want to hear.
3. Work in batches. As a rule of thumb, five to eight slides per request is a manageable size. With each batch, include the last two sentences of the previous slide so the transition fits.
4. Fix the output format. Ask for “Slide 7:” followed by the notes and nothing else. That way you can paste the text straight into your presentation.
5. Check transitions separately. Once everything is written, give the AI only the closing and opening sentences of all slides and ask it to flag repeated phrases and abrupt jumps.

A prompt for one batch might look like this:

You’re writing speaker notes for slides 9 to 14 of a webinar on invoicing for freelancers. The audience is self-employed people with no accounting background. A synthetic voice will read the notes, so write numbers and abbreviations the way they’re spoken. Use only the facts in the list below and don’t add figures of your own. Don’t read out the slide text; explain it with an example. Stay close to the word count given for each slide. The last sentence of each slide should lead into the next. Slide 8 ends like this: “…”

The sentence about facts matters most. Speaker notes often contain numbers that aren’t on the slide, and that’s exactly where invented details creep in. The article on spotting AI hallucinations shows how to catch them. Also make sure any number that appears on a slide is identical in the notes.

The read-aloud test

Speaker notes are finished when they’re easy to say, and the only way to find out is to say them. Open the slides, put the notes alongside and read each slide aloud with a stopwatch.

Listen and look for these things:

• Time per slide. If the measured time is far off the plan, cut or add. Speaking faster isn’t the fix.
• Breath. Any point where you need air in the middle of a sentence gets split.
• The pointing test. Does every “here”, “above” and “this bar” match the slide? Does each revealed point appear as you mention it?
• Audio only. Close your eyes and have someone read you a few slides. People listening with their phone in a pocket need notes that make sense without the picture.
• Picture only. Mute the sound and click through. Someone watching without audio should still be able to follow the thread. The article on webinar subtitles goes further.

If a synthetic voice will read the notes, try a few typical slides before recording the whole thing. Names, abbreviations and numbers are the most likely to come out wrong. The article on AI voices for webinars covers pronunciation in detail.

Checklist for your speaker notes

• Every slide has its own notes of a suitable length.
• The first sentence picks up the slide without reading it out.
• Each set of notes ends with a transition, and no stock phrase repeats noticeably.
• Every number matches your fact list and the slide.
• Words like “here” or “above” match the image and the reveal timing.
• You’ve read the notes aloud and timed each slide.
• Tricky words have been test-recorded if a synthetic voice is speaking.

Frequently asked questions

Should the AI write the whole script at once or slide by slide?

Both, one after the other. First write the script as connected text so the argument holds together. Then split it into notes for each slide and refine them. Starting with individual slides makes it easy to lose the thread.

What if the notes for one slide run too long?

First check whether there are two ideas in there. Usually there are, and then you split the slide. If there’s only one idea, cut repetition and side clauses until the length fits.

Should speaker notes be written in the first person?

If you’re speaking yourself, yes. If a synthetic voice is reading, a neutral presenter without “I” is usually the more honest choice, unless the voice clearly introduces itself as a virtual host. An AI voice shouldn’t pretend to be a specific real person.

Final thought

Good speaker notes turn a stack of slides into a talk. AI gives you a quick first draft for every slide, provided you hand it a word count, your facts and the end of the previous slide. The final polish happens when you read aloud with a stopwatch, because that’s where you hear whether picture, words and transitions fit together.