Add prompt optimization guidance for turn detection to U3-Pro streaming docs

[devin-ai-integration]

Add prompt optimization guidance for turn detection to U3-Pro streaming docs

Summary

Adds a new #### Optimizing prompts for turn detection subsection to the "Using prompts" section of the U3-Pro streaming model selection page (/docs/universal-streaming/select-the-speech-model#using-prompts).

The new content includes:

• A benchmark table from internal testing (100 samples) comparing the default prompt vs. prompt omitted on accuracy, recall, and false positive rate
• An explanation of how the default prompt reduces FP rate from 34% → 24% vs. no prompt
• Four key takeaways for prompt design (default prompt is already optimized, concrete rules work best, filler word instructions are high-signal, recall is near-perfect across all approaches)

This was requested based on learnings from internal turn detection evaluation runs.

Updates since last revision

• Renamed "Strict punctuation rules (default)" to just "Default" in the benchmark table
• Renamed "No prompt" to "Prompt omitted" to clarify it means the prompt parameter was omitted entirely, not that the default prompt was used
• Removed the "Turn-signal framing" row from the table per reviewer feedback
• Removed the #default-prompt hyperlink from the explanatory text
• Simplified the second key takeaway to be more general (no longer references the removed row)
• Reworded the first key takeaway to clearly distinguish "prompt omitted entirely" (83%) from "built-in default prompt" (88%)

Review & Testing Checklist for Human

• Verify benchmark numbers are approved for public docs. The accuracy/recall/FP data comes from an internal Slack discussion — confirm these are okay to publish externally.
• Check the "default prompt" vs "prompt omitted" language is clear to external readers. The distinction between "the built-in default prompt" and "omitting the prompt parameter entirely" may still be confusing to users unfamiliar with the system internals. Read through the section fresh and confirm it makes sense.
• Review rendered output in preview deployment. Confirm the markdown table, bold text, and numbered list render correctly in the Fern docs site. Check heading hierarchy (#### under ###) looks right in the sidebar/TOC.

Suggested test plan: Open the preview deployment URL and navigate to the "Using prompts" section. Verify the new "Optimizing prompts for turn detection" subsection appears with a properly formatted two-row table and numbered list. Confirm the section reads naturally in context with the surrounding content.

Notes

• Requested by Daniel Ince (U07A3FQ7YR4) via Slack in #boardroom-self-service
• Link to Devin run: https://app.devin.ai/sessions/70283c12a5c5430c992d70118ac8f2d4

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment and CI monitoring

[github-actions]

🌿 Preview your docs: https://assemblyai-preview-09f2247f-a184-4055-8b9f-7dbdfccd5c5f.docs.buildwithfern.com/docs

[github-actions]

🌿 Preview your docs: https://assemblyai-preview-76c370c3-ae8a-41c0-a324-7b548fd221a1.docs.buildwithfern.com/docs

[github-actions]

🌿 Preview your docs: https://assemblyai-preview-abfd5cb9-ad18-4764-b3bb-fe0ec46f07db.docs.buildwithfern.com/docs