Telling the AI Assistant What to Do

The assistant does not only wait for clients. In any conversation you can turn the reply box around and tell it what you want, in your own words.

Turning the Reply Box Around

At the bottom of any client conversation, next to the reply box, is a switch labelled Tell the assistant what to write instead.

Flip it and the box changes: instead of typing the message your client receives, you type an instruction to the assistant. The placeholder changes to Tell the assistant what to write…, and the send button becomes Ask the assistant to write this.

Type something like:

• "Offer her Thursday or Friday afternoon."
• "Tell him we're fully booked this week but Monday is open."
• "Move her to the same time next week."

The assistant reads your instruction, checks your real calendar, services and hours, and comes back with a prepared reply — and, where you asked for one, a prepared action such as a booking or a reschedule. It looks things up before answering, so you can ask it questions that depend on real information: who is free, what something costs, what this client had last time.

It can also answer you rather than draft for the client. Ask "how long after her appointment before we close?" and the answer comes back as a note above the box, with nothing prepared to send.

Changing a Draft It Already Wrote

When the assistant has a suggestion waiting for you, the same idea applies to that draft. On the suggestion, use Ask the assistant to change this reply, then say what to change:

• "Warmer, and mention the parking."
• "Offer the 4pm as well."
• "Don't book yet — just ask which day suits."

The send button reads Send guidance to the assistant. It re-thinks and replaces the draft with a new one, rather than editing around the edges. You can also just edit the wording yourself in the box — guidance is for when you want the assistant to reconsider, not for a typo.

Nothing You Ask For Sends Itself

This is the part worth knowing.

Anything the assistant prepares because you asked always waits for your approval — even if your automation level is Full auto. Full auto governs what it does with incoming client messages. It never lowers the bar on something you initiated: you look at the result and press send.

So instructing the assistant is always safe to try. The worst outcome is a draft you reject.

When It Can't

• Your AI allowance is spent. The assistant tells you rather than failing quietly. Your conversations are unaffected — you reply yourself until the allowance resets.
• A suggestion is already waiting and you didn't reference it. Rather than silently replacing work you haven't dealt with, it words your reply instead of starting a new proposal. Worth knowing: in that state it is working from the conversation alone — it will not look anything up, so it cannot answer a question about a time, a price or a policy. Deal with the waiting suggestion first, or type your instruction into the suggestion itself, and you get the version that checks.
• The assistant is switched off for your studio. It will still help you word a message, but the same limit applies — no lookups.
• Very long instructions are shortened rather than rejected. Keep them to a sentence or two — it reads them better anyway.

Common Mistakes

• Typing the client's message into the instruction box. With the switch on, what you type goes to the assistant, not to your client. The placeholder tells you which mode you're in.
• Expecting it to send immediately. It prepares; you approve. That is true at every automation level.
• Using guidance for a typo. Edit the draft directly — guidance makes it re-think the whole reply.