Reply length
Cap how long the AI's drafts get and how it handles long threads. Shorter is almost always better.
Reply length
The single most common complaint about AI support replies is that they are too long. Ochre lets you cap length and tilt the AI toward concision.
How length is controlled
Three settings on AI → Behavior:
- Target length. Short, medium, or long.
- Hard cap. A word or character ceiling that the AI will not cross.
- Long-thread behavior. What to do when the conversation is already 10+ messages deep.
Target length
- Short. One to three sentences. The AI gives a direct answer and stops. Good for chat widgets and FAQ-style email.
- Medium. A short greeting, the answer, an optional link to a longer article. Good default for email.
- Long. Multi-step explanations with bullets when needed. Use only for technical channels with heavy debugging.
The default is medium. Most teams should not move from this.
Hard cap
The hard cap is an absolute ceiling. The AI rewrites until it fits. If the answer genuinely cannot fit, the AI shortens by linking to a knowledge base article instead of inlining the explanation.
Common caps:
- Chat: 80 words.
- Email: 200 words.
- Slack Connect: 60 words.
Caps can be set per channel.
Long-thread behavior
When a conversation is already long, the AI has three options:
- Mirror. Match the customer's most recent message length.
- Shrink. Shorter than the customer's message.
- Summarize first. Add a one-line "tl;dr you asked X, here's Y" before the answer.
Most teams use mirror with a hard cap as the safety net.
Why short usually wins
Three reasons:
- CSAT correlates with reply speed and clarity, not word count. A three-sentence "yes, here's how" beats a 200-word "thank you for reaching out" letter.
- Customers do not read past the first paragraph. Every word after the answer is wasted.
- Short replies cost less. Shorter outputs mean fewer tokens billed to your provider.
You can verify the third point in your AI receipts, where every reply shows tokens out and dollars.
Bullets, lists, and links
The AI uses bullets when the answer is genuinely a list (steps, options) and prose when it is not. You cannot force bullets on every reply, and you should not. Forced bullets read like the AI does not know the answer.
Links to KB articles count as part of the reply. The AI prefers "here is how to do X: [article link]" over a 100-word inline explanation, when the article exists.
Per-channel length
Length and target should match the channel:
- Chat widget: short, hard cap 80.
- Email: medium, hard cap 200.
- Slack Connect: short, hard cap 60.
- Forum or community: medium, no hard cap.
Set per-channel overrides under the Behavior page.
How length interacts with voice
Length and voice are independent. A "warm" tone with a "short" length is fine: the AI will say "Yep, you can do that under Settings → Billing." Tone is not how many words.
Testing length
The Playground shows the exact word count of every draft. Iterate until five sample replies hit the length you want without sounding clipped.
Failure modes to watch for
- Truncation. If the hard cap is too tight, drafts read like they got cut off mid-sentence. Loosen by 20 words and re-test.
- Padding. If the target length is too long for the channel, the AI fills space with "thank you for reaching out" and "please let us know if you have any further questions". Tighten target.
- Mismatch. Long replies on a chat widget feel wrong even when accurate. Per-channel overrides fix this.
Recommended setup
- Target: medium.
- Hard cap: 200 words on email, 80 on chat.
- Long-thread behavior: mirror.
- Re-test after a week of receipts.
Related
Was this article helpful?