Feedback that changes future posts

Thumbs, notes and the standing brief — how a reaction becomes a lasting rule.

You steer the agent by reacting to what it writes, not by rewriting it. Every thumbs-up, thumbs-down and note is read the next time the agent sits down to write — and the time after that. This page explains exactly how far a reaction travels.

Thumbs and notes

Any post — in the outbox, on the Posts page, in the first-posts round during setup, or on a client's private page — can be graded 👍 Sounds like me or 👎 Not me, with an optional note. Grades save instantly; there is no submit button.

What happens next:

• When the agent writes its next post, it is shown your most recent graded posts, each marked GOOD or BAD, with your note attached to the bad ones. It is told to keep writing like the good ones and to avoid whatever the notes describe.
• Only the most recent grades are shown — roughly the last ten, from the last month — so what you said this week outweighs what you said in January. A rule you want kept forever belongs in the brief (below), not in a note.
• Grades from a client through their private link count exactly like yours.

Images have their own grade

If the text is right and the picture is wrong, grade the image separately. Image grades and notes ("too corporate", "no people", "never stock-photo handshakes") steer the image concept the same way text grades steer the writing, without punishing a post whose words were fine.

Notes that work

The agent is a careful reader and a literal one. Tell it what was wrong and what right would look like:

The standing brief

Grades are recent examples; the brief is permanent law. It is everything you told the agent during setup, kept as editable answers on the agent's console. When a note is really a rule — "never mention fees", "always write in first person" — put it in the relevant answer (off-limits, voice) so it applies forever and is never crowded out by newer grades. "Want to share something new?" on the agent's page is a quick way to add to the brief in conversation, and "Let me ask you something" has the agent ask for the one thing that would most improve its posts.

The lights on the brief

Each answer in the console carries a small light — green, yellow or red — showing whether it is rich enough to write from. They are the same reads the setup conversation gave you as you typed, and they measure richness, not correctness. If posts feel generic, look for a yellow or red light first; a thin answer to "who you write for" is the most common cause. "Re-check brief" re-reads all answers after an edit.

When feedback isn't the tool

• A post that states a wrong fact → that is what the claims check is for; it corrects what is wrong before the post reaches the outbox and holds back what it cannot vouch for. If one still got through, a 👎 with the correction in the note is the right move: the reviewer and the writer both see it next time. A standing note also helps ("don't cite statistics unless they're in my material").
• A post about the wrong subject entirely → check the material attached to the agent; a feed with a vague description will pull in off-topic items.
• You want a specific thing said this week → tell the agent with "Want to share something new?" (a project, a win, an opinion), then press + in the outbox. The next post is written with it in mind.