Last updated: Oct 19, 2026 at 6:05PM
When Osello's AI isn't sure enough of a field's value, it doesn't write it. The agent asks you in the chat instead, and carries on with the work once you answer.
How sure the AI has to be
Osello sets a confidence threshold on every marketplace field: how sure the AI must be before it fills that field. Most fields need 80 out of 100. Fields where a wrong answer is costly ask for more.
The AI checks its answer against the field's threshold before writing anything. Below it, the field keeps whatever value it had. For a required field the agent asks you what to use. An optional field it has too little data for may be left empty instead.
Osello doesn't show the score anywhere. You see the agent's question, or the value once it's written.
Answer the AI's question
Questions appear in the chat that's doing the work. In the sidebar, the Chats link under AI Agents shows a badge with how many chats are waiting on your reply. Every listing and variant also has an AI Chats tab listing the chats about it.
- When the agent asks a question, type your answer and click Send.
- When it asks for your go-ahead, click Approve, or Something else and tell it what to do instead.
When an agent hands work to other agents, for example one agent per listing on a large selection, their questions come back to the chat you started, so you answer everything in one place.
When the same question keeps coming up
- If you know the answer, enter it yourself on the listing's Fields tab.
- If the AI keeps asking for a reason you can name, add custom instructions to that field. You can also tell the agent in the chat to remember your answer from now on, and it saves the rule to the field's instructions.
- If the source data was the problem (a thin description, a missing spec), add the detail to the product in your store. Osello imports the change and the next fill has more to work with.
Scheduled runs
Nobody is watching a routine while it runs. With Let the agent proceed without approval when it's confident turned on, the agent goes ahead with work that's clear-cut and still stops to ask when it isn't sure. That run waits until someone answers.
