Best Of

When I began adapting content from online learning materials into scripts for AI avatars, I expected the process to feel familiar, like writing narration for a video or voiceover. I quickly realized that scripting for synthesized speech is a different kind of challenge. What looks good on paper can sound robotic out loud, and tiny changes in punctuation can shift the whole tone.

Whether you're building branching scenarios, integrating avatars into course content, using text-to-speech for accessibility, adding voice to knowledge checks, enhancing onboarding experiences, or creating engaging feedback in Brightspace, here are a few things I’ve learned that might help you bring more life to your AI voices.

✏️ Write How It Sounds, Not How It Reads

AI avatars don’t just read your script, they perform it. Unlike a human narrator, though, they need clearer cues for pacing and tone.

• Use commas and ellipses to guide rhythm and pause.
• Choose contractions (“we’re” instead of “we are”) to sound more natural.
• Break up longer sentences to avoid a flat, monotone delivery.

✅ Try this: Read your script out loud. If it feels clunky or too stiff, it’ll sound even more awkward when the avatar reads it.

📣 Watch for Tricky Pronunciations

AI voices often get tripped up by:

• Proper names
• Acronyms
• Homonyms (like “read” vs. “read”)

One example that stands out was the word “present," which needed to be written as “preezent” to sound correct. Spelling things out phonetically or adjusting the sentence can make a big difference.

✅ Pro tip: You can improve pronunciation by playing with phrasing, punctuation, or even using phonetic spellings. I found that simplifying the surrounding sentence often worked better than trying to force a technical fix.

🧠 Use AI to Help Troubleshoot Pronunciation

Sometimes, no matter how many tweaks you try, the AI still doesn’t say it quite right. When that happens, I use generative AI tools to brainstorm alternatives. By feeding in a tricky line and asking for clearer or more phonetic versions, I can quickly explore new options without getting stuck.

• Ask AI to suggest phonetically friendly rewrites
• Try prompts like: “Make this easier for text-to-speech to pronounce”
• Use it to compare different wording approaches

✅ This kind of quick brainstorming has saved me time and helped smooth out some of the more frustrating hiccups.

✂️ Keep It Short and Sweet

Brevity goes a long way. Long or complex sentences tend to sound dull or confusing when read by an avatar.

• Break ideas into smaller chunks
• Use line breaks for natural pauses
• Don’t be afraid of fragments as they often sound more natural

✅ Think of each line as one breath. What would you say in a single moment, out loud?

💬 Tone Is in the Text

If you want your avatar to sound friendly, excited, or warm, your words need to carry that tone.

• “Let’s explore this together!” works better than “We will now examine…”
• Choose active verbs and language that sounds human
• Use exclamation points carefully — they can boost energy when used sparingly

Final Thought: You’re Not Just Writing, You’re Directing

Scriptwriting for AI avatars is part writing, part sound design. You’re directing a performance, even if your “actor” is synthetic. I've found that writing for AI voice is a distinctly different skill from writing text meant to be read.

If you’re trying this in your own courses, I’d love to hear how it’s going. What’s worked? What’s still frustrating? Let’s keep learning from each other and sharing the quirks we encounter along the way.