Give feedback — improve Wavo

Last updated: May 19, 2026

Give feedback — improve Wavo

Per-message feedback: thumbs up / down icons under each assistant message. Backed by PATCH /api/v1/chat/messages/:messageId/feedback. Body: feedback_rating SmallInt (-1 = thumbs down, 0 = neutral, 1 = thumbs up), optional feedback_comment text. Stored on chat_message row. Aggregated by team for prompt + model tuning. Don't only flag mistakes — also confirm wins. For deeper issues: create_support_ticket tool inside chat.

Who is this for

Anyone using Wavo regularly. Especially: early adopters whose feedback shapes how Wavo evolves.

How to give feedback

Thumbs up / thumbs down

Below each Wavo response:

  • 👍 Thumbs up — answer was good

  • 👎 Thumbs down — answer was bad

Click → instant feedback recorded.

Add a comment

After thumbs up / down: optional text input for context:

  • "Wrong number — actual ROAS is X"

  • "Perfect, this is exactly what I needed"

  • "Tool call worked but explanation was unclear"

Comments help the team understand why the rating, not just that it happened.

API

PATCH /api/v1/chat/messages/:messageId/feedback:

{
  "feedback_rating": -1,
  "feedback_comment": "Wrong ROAS; postback data not included."
}

feedback_rating is SmallInt in range -1..1:

  • -1 thumbs down

  • 0 neutral (rarely used)

  • 1 thumbs up

Stored on the chat_message row.

What to give feedback on

Thumbs down (bad)

  • Wrong data — Wavo cited a number that doesn't match the truth

  • Hallucinated — Wavo made up something that doesn't exist (rare; injection-defense + persona instructions try to prevent)

  • Wrong action — Wavo planned to do X but you asked for Y

  • Confusing explanation — answer was correct but hard to follow

  • Tool failure mishandled — tool returned error but Wavo didn't explain

  • Too verbose or too brief

  • Wrong language — Wavo replied in a different language than your input

Thumbs up (good)

  • Direct + accurate — got the right answer fast

  • Useful follow-up suggestions — chips genuinely helpful

  • Tool chain worked — multi-step planning ended in correct outcome

  • Good prompt-craft suggestion — Wavo helped refine an ambiguous question

Don't only flag mistakes. Wins help the team understand what's working.

How feedback is used

Per-team feedback aggregated for:

  • Prompt tuning — system prompt iterations based on common rating patterns

  • Model selection guidance — which models perform best for which task types

  • Tool improvements — tools with low-rating output get reviewed + improved

  • Documentation — articles like this one updated when feedback reveals confusion

Aggregated, not per-user-tracked for performance management.

When to escalate beyond feedback

Feedback is for iterative improvement. For urgent issues:

Wrong actions executed

If Wavo executed something incorrectly (e.g. paused wrong campaign despite approval):

  1. Give thumbs down + comment

  2. Check ai_action_log for what payload was confirmed (payload_hash)

  3. Compare to what you intended

  4. If genuine bug: open support ticket via Wavo: "Open a support ticket about this incorrect action"

Persistent hallucinations

If Wavo repeatedly invents data:

  1. Try switching model (Sonnet → Opus or Gemini Pro)

  2. Add explicit instructions ("cite source", "if unsure, say so")

  3. Give thumbs down with specifics

  4. Open support ticket if pattern continues

Security concerns

If you observe Wavo:

  • Exposing what looks like a secret that got past sanitization

  • Performing actions you didn't approve

  • Crossing team boundaries

Don't post details in chat. Use direct support channel (Help Center → Contact Support) with specifics. Security issues handled separately from product feedback.

The support ticket tool

Wavo can open tickets for you:

You: "I think Wavo's answer is wrong. Can you open a ticket?"
Wavo: [invokes create_support_ticket]
Wavo: "Ticket #12345 created with our conversation context attached."

Tool: create_support_ticket (LOW write mode — no approval needed).

Tickets flow to Pylon (apps/backend/src/routes/api/pylon.route.ts):

  • Context: last messages of conversation

  • Subject: derived from your description

  • Assigned to support team

  • You + team can follow up via Pylon UI / email

Tone in feedback comments

Helpful patterns:

  • ✅ "Wrong number — should be X based on tool Y"

  • ✅ "Confusing — tried to explain ROAS but mentioned CTR midway"

  • ✅ "Perfect — saved me 10 min"

Less helpful:

  • ❌ "Bad" (no detail)

  • ❌ "Useless" (no specifics)

  • ❌ Personal frustrations (Wavo + team can't act on those)

Specific feedback = actionable improvement.

Feedback frequency

Don't worry about over-feedbacking:

  • Thumbs up / down is a single click — no friction

  • Comments optional

  • Aggregated weekly / monthly for prompt iteration

If you give thumbs up to most answers and thumbs down to occasional misses: that's the right pattern. Pure thumbs down without occasional wins suggests systematic prompt-craft issue worth investigating.

Per-message vs session-level

Currently: per-message only. No session-level feedback.

If you want to give overall feedback on a session: comment on the most representative message with session-wide observation.

Privacy of feedback

Feedback is per-message tied to:

  • The message ID

  • Your user_session_id (who gave feedback)

  • Timestamp

Used by:

  • Workspace admin (audit usage patterns)

  • Wavo product team (aggregated for tuning)

Not exposed to other users in your team unless explicitly shared.

Anonymity

Feedback is not anonymous to admins of your workspace. If you want to flag something sensitive without attribution: use the support ticket flow (mark as confidential).

Common questions

Does my thumbs down change Wavo's behavior immediately?

No. Feedback aggregates for periodic tuning. Doesn't change in-session behavior. For in-session correction: tell Wavo directly ("that's wrong because X, please redo").

Can I edit my feedback after submitting?

Re-submit a new rating; the new one overwrites (PATCH overwrites the row's fields).

Can I delete feedback?

Set feedback_rating to 0 (neutral) + comment empty — effectively cleared.

Does feedback affect my credit usage?

No. Feedback is free.

Where do I see my past feedback?

Not surfaced in standard UI. Admins / support can query if needed.

Related