This is the lesson that matters. Everything else is ordinary document software.
Type @claude where you mean it
Put the cursor in the line you want changed and type @claude. A menu opens with two sections: Ask Claude, and Tabs. The same @ addresses the assistant or references another tab, so expect to hit the wrong one at least once.
Write the ask inline
Choosing Claude turns the line into a composer reading: tell Claude what to write or edit. Write the instruction the way you would write a review comment, because that is what it is.
Watch it become a thread
On submit it becomes an anchored comment on the right, the source text highlights amber, and the card shows that Claude is working on it. The conversation is now attached to that passage rather than floating in a chat.
Read the reply in the thread
When it finishes, Claude replies in that same thread, marks it resolved, and leaves a reply-to-reopen box underneath. The reply is short and points at what changed, because the edit itself is already in the page.
The reply box carries a Send to Claude tick, on by default. Ticked, your comment is an instruction. Unticked, it is an ordinary note for a human. That single checkbox is the whole difference between briefing and annotating, and it is easy to miss.
Two asks at once
Send a second comment while the first is still running and it does not queue silently. The card says Claude is finishing another turn and yours is next. A small thing, and the reason you do not sit wondering whether it registered.
Why this beats describing the change
House rules are the clearest case. Tell it to remove every em dash from a document and it rewrites them as colons and commas throughout, then tells you in the thread what it changed. You enforced a style rule by pointing at one sentence that broke it.
You can brief a doc
- Opened an anchored comment with @claude
- Know what the Send to Claude tick does
- Know where the answer appears
Next: what it does while you wait.