A good AI voiceover starts with a script written for the ear: short sentences, one idea at a time, numbers and abbreviations spelled out the way they should be said, and a length sized to the time you have. As a rule of thumb, a script read at a moderate 130 words per minute needs about 130 words for each minute of audio, so a five-minute voiceover is roughly 650 words. This guide covers planning, timing, writing and formatting a script for AI voice tools, with a simple template you can adapt.
Speaking-rate figures come from published research and from the presets in our own calculator. Details about ElevenLabs and Murf AI come from their official websites, documentation and pricing pages, checked between September 30 and October 2, 2026. Everything else is WriterFlowLab's editorial guidance. WriterFlowLab may earn a commission if you sign up for ElevenLabs through links on this site. We have no affiliate relationship with Murf AI. See our affiliate disclosure.
Plan the voiceover before you write
Answer four questions first. They shape every later decision about length, tone and structure.
- Where will it be heard? A product demo, a YouTube explainer, a course lesson and a podcast intro all have different expectations for pace and tone.
- Who is listening? Beginners need more context and slower pacing than experts.
- How long should it run? Set a target length before you write, not after.
- What should listeners remember or do? Write this down in one sentence. If a line does not support it, cut it.
If you are turning an existing article into audio rather than writing from scratch, our guide to turning a blog post into a voiceover with AI covers that workflow.
How long should your script be?
The basic calculation is: words needed = minutes of audio × speaking rate in words per minute. Our reading time and speaking time calculator uses three speaking presets: 100 words per minute (slow), 130 (average) and 160 (fast). At those rates, a script needs roughly:
- 30 seconds: 50, 65 or 80 words
- 1 minute: 100, 130 or 160 words
- 2 minutes: 200, 260 or 320 words
- 3 minutes: 300, 390 or 480 words
- 5 minutes: 500, 650 or 800 words
- 10 minutes: 1,000, 1,300 or 1,600 words
These are estimates, not measurements. Research suggests people read aloud faster than our average preset: a 2019 meta-analysis put the average oral reading rate at 183 words per minute. Our guide to calculating reading time covers that research. AI voices also vary: their pace depends on the voice, the model and the speed setting. ElevenLabs, for example, lets you adjust speed from 0.7 to 1.2, with 1.0 as the default.
Our guidance: write to the 130 words-per-minute estimate, generate a short sample, time it, and then trim or extend the script rather than relying on the estimate alone.
Write for the ear, not the eye
Listeners cannot skim, scroll back or see your formatting. These habits make a script easier to follow when it is spoken:
- Keep sentences short. Split long sentences, and move side points into their own sentence or cut them.
- One idea per sentence. If a sentence needs a semicolon, it probably needs to be two sentences.
- Put the key point first. Say the conclusion, then the reason.
- Signpost the structure. Phrases such as "There are three steps" or "Next" help listeners keep their place.
- Repeat what matters. A short recap at the end is useful in audio, even though it would feel repetitive on the page.
- Remove visual references. Lines such as "see the chart below" or "as shown above" do not work in audio.
- Avoid parentheses and long lists. Turn them into separate sentences, or keep lists to three or four items.
Write numbers, names and symbols the way they should sound
AI voices read exactly what is written, so spell out anything that could be read more than one way. ElevenLabs recommends writing numbers, symbols and abbreviations as words for better pronunciation.
- "2026" as "twenty twenty-six" if that is how you want it said.
- "%" as "per cent" and "&" as "and".
- Abbreviations as spoken: "for example" rather than "e.g.".
- Uncommon names and terms phonetically, or with the tool's pronunciation feature. Murf has a pronunciation editor, and ElevenLabs supports pronunciation dictionaries.
Format the script for AI voice tools
- Use punctuation for pacing. Commas and full stops shape the pacing of most AI voices. Murf also lets you adjust pause length directly, per sentence or per word.
- Break the script into short sections. Sections are easier to regenerate when one line needs fixing. When we checked ElevenLabs' documentation, a single generation was limited to 5,000 characters on paid plans and 2,500 on the free plan.
- Keep delivery notes separate. Do not leave notes such as "[say this warmly]" in the text unless the tool supports them. ElevenLabs' Eleven v3 model accepts audio tags to steer delivery, but other models and tools may read such notes aloud.
- Check the character count. ElevenLabs text to speech on V2 Multilingual models uses one credit per character, so the length of your script in characters affects how much of your plan it uses. Our word counter shows both words and characters.
A simple script template
This structure works for most short explainers, product demos and lesson introductions:
- Hook (one or two sentences): the problem or question the listener cares about.
- Context (one or two sentences): why it matters or who it is for.
- Main points (two to four): each with a short example.
- Recap (one sentence): the key takeaway, repeated plainly.
- Next step (one sentence): what the listener should do now.
Example: a short explainer script. This is an illustration written for this guide. It is about 110 words long, which comes to roughly 50 seconds at 130 words per minute:
"Ever pasted text into a document and ended up with odd spaces and broken lines? You are not alone, and it takes seconds to fix. Here is how to clean it up in three steps. First, paste your text into a text cleaner. Second, choose what to remove: extra spaces, blank lines or duplicate lines. Third, copy the cleaned text back into your document. That is it. One more tip: check the result before you publish, because a cleaner fixes formatting, not facts. So, to recap: paste, choose, copy. Next time messy text slows you down, give it ten seconds in a text cleaner, and get back to writing."
Paste a draft like this into the speaking time calculator to see how close it comes to your target length.
Check the script before you generate
- Read it aloud yourself. Anything you stumble over, an AI voice is likely to stumble over too.
- Check the length. Compare the word count and estimated speaking time with your target.
- Clean pasted text. If the script came from a document or another tool, remove stray spaces and line breaks with the text cleaner.
- Generate a short sample first. Listen to the opening lines in your chosen voice, then adjust the script or settings before generating everything.
Common mistakes
- Writing for reading: long sentences and dense paragraphs that work on the page but tire listeners.
- Ignoring length until the end: cutting a finished script by a third usually damages its structure.
- Leaving symbols and abbreviations as written: they are easy to mispronounce.
- Generating everything at once: one mistake can mean regenerating the whole voiceover.
- Skipping the final listen: always listen to the full voiceover before you publish it.
Frequently asked questions
How many words is a one-minute voiceover script?
About 130 words at an average speaking pace, or roughly 100 to 160 words depending on how fast the voice reads. Generate a sample and time it to be sure.
Do I need a different script for AI voices than for human voice actors?
The principles are the same, but AI voices need more help with pronunciation, numbers and pacing, because they read exactly what is written.
Which AI voice tool should I use?
It depends on your project. Our Murf AI vs ElevenLabs comparison compares two AI voice tools side by side.
Next steps
When your script is ready, see how to use ElevenLabs for AI voice content for a step-by-step generation workflow, or read our overviews of ElevenLabs and Murf AI.