Assistants

Assistants and saved prompts

If you keep typing the same instructions at the start of every conversation, stop. There are now six assistants and thirteen saved prompts waiting for you, and you can add your own.

Both were added on 12 September 2026. Assistants are further down this page; prompts come first because they are the quicker win.

They are already there

Open Prompts in the side panel. You will find thirteen ready to use, shared with everyone.

You do not have to set anything up. This is new as of 12 September 2026 — the page you may have read last week told you to type them in yourself, and that is no longer necessary.

Two ways to use one

From the side panel. Click Prompts, click the one you want, and it drops into the message box. Add your own details underneath and send.

By typing a slash. In the message box, type / and the list appears. /delay-notice, /reply-ja, and so on. This is faster once you know the names.

Either way the text lands in the box before you send it, so you can edit it, delete the parts that do not apply, or add context. Nothing is locked.

What is there

Writing to customers

/customer-explanation Explain an editing decision warmly, to someone writing in a second language
/reply-zh-tw Draft a reply in Traditional Chinese, with an English back-translation so you can check it
/reply-ja Draft a reply in Japanese, with an English back-translation
/delay-notice Tell a customer their work will be late. Leads with the new date, apologises once
/quote-explainer Explain what a quotation covers, without being defensive about the price

Working on documents

/manuscript-first-read Upload a manuscript and get an assessment before you start
/academic-polish Improve the English without touching terminology, citations or figures
/terminology-check Find terms used inconsistently, and list them rather than changing them
/cover-letter Draft a journal submission cover letter from the manuscript

Working with the team

/editor-emails Short, warm emails to freelance editors
/editor-brief Turn messy job notes into a clear brief for the editor taking the job
/decision-summary Turn a discussion into what was decided, what is open, and who does what
/japanese-web-content Natural Japanese for the website, not translated-sounding Japanese

They are written to ask rather than invent

Every one of them is told never to make up a fee, a deadline, a job number or a person's name. When something is missing it leaves a blank marked [ ] and tells you what it needs.

That is deliberate, and it is the part worth trusting. A drafted email that says the deadline is [ ] is useful. One that confidently states a deadline nobody gave it is worse than no email at all, because it reads exactly the same.

You are still the last person who reads it before a customer does. See Using it on customer work.

Make your own

The thirteen are a starting point, not the whole set. Use the + button on the Prompts page to save your own, and they sit alongside ours.

If that feels like effort, describe what you want and let it draft the wording:

Write me a set of instructions I can paste at the start of a conversation, for preparing concise, friendly emails to editors. It should ask me for missing job details rather than inventing them, and never state a deadline or a rate I have not given it.

Then save the result as a prompt.

Yours are private unless we share them. If you write one the rest of the team should have, tell Daniel and it can be published to everybody.

Assistants are back, and this time they work

Agents is switched on again as of 12 September 2026, with six ready to use.

It was off for two days, and the reason is worth one sentence: the menu was there with nothing behind it, so anyone who selected it got a raw error. We built the assistants first this time and turned the menu on second.

Pick one from the model menu at the top of the chat screen, the same place you choose between gpt-5.4 and the others.

Academic English editor Polishes a manuscript's English without touching terminology, citations or figures. Reads files you upload
Manuscript first read Assesses an uploaded manuscript before you start — what it needs, and what will take longest
Customer service — Taiwan Writes to customers in Traditional Chinese, with an English back-translation so you can check it
Customer service — Japan The same, in Japanese
Editor liaison Emails to freelance editors, and briefs for the editor taking a job
Publication advisor Cover letters, journal requirements, and help answering reviewers. Reads files you upload

How an assistant differs from a saved prompt

A saved prompt is text you paste and can edit before you send it. An assistant applies its instructions to the whole conversation automatically, and you cannot edit them away.

That makes assistants better for work you do the same way every time, and prompts better when you want to start from something and change it.

They are told to ask rather than invent

Every assistant is instructed never to state a price, a deadline, a job number, a rate or a name it was not given — it leaves [ ] and tells you what is missing. The customer-service ones also give you an English back-translation so you can check the meaning before sending.

What they cannot do, and will tell you so: look up a customer, job or expert record; read your email; open anything in Google Drive; search the web; create a ticket. None of those is connected yet. See What is not here yet.

You can build your own

The + in the Agents panel creates one. Give it a name, pick a model, and write its instructions — plain English, the same way you would brief a new colleague.

Yours is private until we publish it. If you build one the team should have, tell Daniel.

One warning

Instructions you send at the top apply to everything after them in that conversation. If you start with /academic-polish, which never changes technical terms, and then ask it to rewrite a marketing page, it will be oddly conservative. Start a new conversation for work that does not fit.

Tell us what is missing

This set was written by guessing at what you repeat. Some of it will be wrong.

If a prompt does not work, or produces something odd, say so. If you find yourself typing the same thing over and over and there is no prompt for it, say that too — that is the most useful message you can send us, because it is a prompt we should have written and did not.

Email Daniel, or send a screenshot. A screenshot is genuinely valuable: some errors in this portal never reach us at all, so if you do not tell us, we do not know. See When something goes wrong.