Feature Request: Granular Controls for AI Assist Behavior

Currently, once AI Assist is enabled in the inbox, it replies to every message - including newsletters, out-of-office replies, and other system-generated messages. There’s currently no way to exclude certain types of messages, filter them based on content, or selectively assign conversations to the AI.

I really hope this can get some priority as it seems like an easy fix to a pretty critical issue. Yesterday in testing it replied:

Hello,

Thank you for your message. It appears that this is an automatic reply, and I won’t be able to assist you further with this inquiry. If you have any questions or need support, please feel free to reach out again after May 19th.

Best regards,
Aria
Customer Support Assistant

As you can imagine, replies like these are like liquid gold for the reddit trolls!

It’s essentially all or nothing and that creates some major limitations in real-world workflows where some messages clearly don’t need (or shouldn’t get) a reply.

What’s more:

  • AI Assist can’t be manually assigned - once it’s set to handle the inbox, it responds to everything automatically.

  • AI Assist won’t reliably reply if a human has already responded - it won’t generate smart replies or continue the thread, even if that would be helpful.

When left to run fully solo, it actually handles many tickets pretty well - but the lack of control makes it functionally buggy for mixed-team workflows. There’s no way to:

  • Prevent it from replying to known non-human messages.

  • Share the workload between human agents and the AI.

  • Let the AI follow up after a human reply.


Suggestions for improvement

To make AI Assist more usable and flexible, here are a few ways it could be improved:

1. Add “do not reply” logic to the base prompt

Let users include specific language in the main prompt to help the AI understand what to skip - e.g., messages containing phrases like “out of office,” “noreply,” “newsletter,” etc.

2. Create a dedicated “exclusion prompt” field

Keep exclusion logic separate from the tone/behavior prompt. This would be a simple way to manage reply rules without cluttering the main config.

3. Enable an “excluded terms” list

Provide a space to define specific keywords, phrases, or sender addresses that the AI should never reply to (e.g., “unsubscribe,” “mailer-daemon,” “noreply@,” etc.).

4. Implement automatic detection of auto-replies/system messages

AI Assist could be set to ignore messages that match known auto-response patterns, such as bounce notifications or vacation responders.

5. Allow manual assignment of tickets to AI Assist

Enable human agents to choose which conversations the AI should handle - ideal for support teams who want a hybrid setup rather than full automation.

6. Let AI respond after a human reply (restore smart response ability)

AI Assist should be able to continue assisting after a human response — either by regaining context and generating Smart Answers as usual, or by being re-enabled for the thread. This would allow the AI to contribute to ongoing support conversations instead of becoming inert after the first agent interaction. Manual or conditional reactivation (e.g. if the ticket remains open or unanswered for a time) would also be useful.

NOTE: I will submit a separate thread on this issue

7. Limit number of AI replies per ticket

Add controls to set a max number of AI responses per conversation - helpful for avoiding infinite loops or excessive messaging.

8. Tag-based inclusion/exclusion

Allow tickets with certain tags to be auto-excluded (or included) for AI handling - another way to filter out messages that shouldn’t get a reply.


While I recognize there are workarounds e.g. manually filtering emails before forwarding or setting up a separate property to catch messages the AI shouldn’t handle, these approaches introduce unnecessary complexity and redundancy. Ultimately, these changes would make a huge difference in making AI Assist properly usable at scale - especially for teams balancing automation with the need for occasional hands-on support. Even just a few of these would go a long way.

Hi,

Thank you so much for taking the time to share this detailed and thoughtful feedback.

You’ve raised some very valid points regarding the current limitations of AI Assist — particularly around its lack of filtering, control, and flexibility in mixed-team environments. We absolutely understand how critical this is, especially when dealing with automated messages and maintaining a smooth customer experience.

We’ll be forwarding your suggestions to our development team for review, and we’ll explore how we can enhance the AI to better support real-world workflows. Your input is incredibly valuable in helping us identify areas for improvement, and we appreciate you highlighting specific features that would make a meaningful impact.

While we can’t make any promises on timelines, rest assured that feedback like yours plays a big role in shaping our roadmap.

Thanks again for your insights!