Training the AI on Your Brand Voice

A reply that sounds like you doesn't just feel better — it converts and retains customers better. Here's how to give Taskra enough signal to match your voice within a few hours of training.

What "brand voice" actually means here

Voice has three layers: tone (warm vs corporate), pacing (short and punchy vs flowing) and vocabulary (the specific phrases you reach for). Taskra uses all three when it drafts.

The Brand Voice textarea

On the AI Training page, paste 2–4 sentences that show how you actually talk to customers. Real previous replies work best — don't paraphrase, don't sanitise.

• Good: "Hey! Totally fair — let me get that swapped out today. I'll DM the courier now and send you a fresh one with express ✨"
• Avoid: "We will process your request promptly." (too generic, the AI already does this by default)

Sign-off and "always say this"

If you always sign replies with a specific name or include a fixed line (e.g. "Reply any time — I read every email"), put it in the Sign-off and "Always say this" fields. The AI appends them automatically.

Tone slider — when to use which

1. 1Friendly — fashion, beauty, lifestyle. The default for most DTC brands.
2. 2Empathetic — wellness, baby, sensitive purchases. Lead with care, then logistics.
3. 3Professional — B2B, wholesale, trade accounts. Punctual, polite, not formal.
4. 4Casual — youth brands, indie merch. Emojis and contractions encouraged.

How fast does it actually learn?

The system prompt is rebuilt on every reply with your current brand voice, so changes take effect immediately — no training run, no waiting. Tweak, send a test message, see the new tone.

Things to avoid

• Don't list 10 different example replies — 2–3 high-quality ones outperform a wall of text
• Don't put confidential SKUs or policy details in Brand Voice — use the Policies and FAQs fields instead
• Don't switch tone every week. Pick a direction and let the AI settle into it