Chattering reads your conversations and files real demand onto the roadmap for you.
What it looks for
The obvious signal is a question the AI could not answer. The stronger one is a question it could answer — where the answer was "no, we don't do that". Retrieval succeeded, so an ungrounded-only check never sees it, and it is the clearest statement of demand a customer can make.
Clustered by capability
"iPhone app", "Android app", "a phone app instead of the website" and "an app for iPad" collapse into one item with four people counted, because the matching is on the capability rather than the wording.
Filed once
Each exchange is classified once and filed in exactly one place — the roadmap, or the content-gap list — so the same customer is never counted twice.
How the filing happens
Each time the AI replies, Chattering checks that exchange right away, not when the conversation closes. An AI classifier reads what the customer asked for; for answers the AI did give, a quick wording check first picks out the ones that sound like a "no". A customer who asked "does it work with Shopify?" and heard "not yet" is filed under the same capability node as someone who asked "can I connect my store?" and got the same answer.
The analysis runs automatically on every plan except Free, where it is paused. A new item waits under Suggested by the AI in Feedback until you Accept it onto your public board or Decline it.
Where requests appear
Auto-filed requests show up in Feedback in the dashboard, alongside requests posted manually through the portal. They go through the same triage: set a status (Under review, Planned, In progress, Shipped, Declined), add tags, and merge duplicates. The source — a real conversation rather than a portal form — is recorded, but the item itself sits on the same board and gets the same treatment.
Not all signals go to the roadmap. If the AI could not answer and the question was about where or how to do something, the exchange points to a documentation gap rather than a missing feature. Those land in Help center → Content gaps, not on the feedback board. A customer asking how to reset their password is not asking for a new capability — that conversation is classified and routed separately.
Merging with portal requests
If a customer has already posted "API access" on the public portal and five conversations later surface the same need, you can merge those auto-filed items into the existing request from the Feedback board. Votes, comments, and subscribers all transfer to the surviving item. Old links to the merged request open the surviving item, so nothing breaks for anyone who bookmarked it.
This is how ten separate "Shopify integration" conversations become one request with ten people counted, rather than ten half-empty items that each look low-priority.
What does not get filed
Not every question is demand. A customer asking for a billing explanation, a password reset, or a how-to that your docs already cover is not signalling a capability gap. Those conversations close without touching the board.
What triggers filing is specifically one of two things: the AI had nothing to answer with, or it answered with a clear boundary. The line "we don't support that currently" is the strongest filing signal there is.
Reviewing what came in
Open Feedback and sort by Votes. The items at the top represent the capabilities the most customers have explicitly said they want. When something ships, mark it as Shipped from the dashboard and Chattering tells everyone who raised it in a chat — back in the conversation where they asked, not in a mass email. The loop closes without you having to track down who to tell.