Getting the most out of chats

A friend, a family member or a business has built an assistant for you on Deputise. It is an AI set up by a real person, its owner, with instructions and information about one topic: a heating system, a phone, a holiday let. It is not a general-purpose chatbot.

A few things to keep in mind:

It works best on the topic it was built for. Off-topic questions get a polite redirect or a guess.
Treat its answers like advice from a knowledgeable friend: useful, but worth checking when the stakes are high. Every chat carries a note that the assistant can make mistakes.
Behind it is a real person who can step in when needed.

Stay roughly on-topic

If your friend built you a "Help with the cottage" assistant, ask it about the cottage: the heating, the Wi-Fi, the bins, the quirky shower. Ask it for restaurant recommendations and you'll get something generic at best.

If you're not sure what the assistant is for, its name and the welcome message at the start of the chat usually tell you.

Asking good questions

Be specific.

Say what you're trying to do, not just what's wrong. Instead of "the heating isn't working", try "the radiators in the lounge are cold but the boiler seems to be on".
Mention the make, model or version where it matters: "how do I block a number on an iPhone 16".
Say what you've already tried.
Add a photo if the assistant accepts images. Error screens, odd symbols and "what is this thing on the wall" all work well.
Ask one question at a time.

If the owner were reading your message over your shoulder, would they know enough to help? If not, add the missing detail.

Following up

It's a conversation, not a search engine.

If the answer isn't quite right, say so: "that didn't fix it" or "I meant the kitchen radiator, not the lounge one".
If you don't understand something, ask. "What does 'pressure gauge' mean?" is a perfectly good message.
If you're given a list of steps, work through them and report back what happened.

The assistant remembers what's been said earlier in the same chat, so you don't need to re-explain. Notes the assistant keeps about you carry over, described below.

What the assistant remembers between chats

Each chat starts fresh, except for a few short notes the assistant keeps about you.

A note gets there in one of three ways:

The assistant offers one. When the two of you have worked something out that would save time next time, such as how your equipment is set up, a fix that worked, or how you like things explained, a card under its answer suggests a note. Choose Save to keep it, Edit to change the wording first, or Not now to leave it. Nothing is kept unless you save it.
The owner adds one. The owner can write a note, or save their answer to a question you passed to them as one.
You add one yourself.

The assistant reads your notes at the start of every chat with you. Where a note and the information the owner gave the assistant disagree, it goes with the owner's information.

To see your notes, open your assistants list and choose Memory from the assistant's menu. You can edit or delete a note, add one, or clear them all. The assistant keeps up to 30 notes for you. Once it has 30, the card shows a Review memory link to your assistants list, where you can delete one before saving another.

The assistant won't offer to remember passwords, codes, or money details, and you shouldn't save them in a note yourself. It may offer to remember a health detail that affects how it helps you.

When the assistant can't help

Sometimes the assistant won't know the answer, or will go in circles, or the question needs a person. You can ask for one at any time:

"Can you get [owner's name] to help?"
"I want to talk to a person."
"Please pass this on to someone."

The assistant passes your question to the owner with a note about what you were asking and why it couldn't help. The chat confirms this and pauses until they reply, for up to 48 hours. If nobody answers in that time, or the owner closes the request without replying, you'll be told and can carry on chatting with the assistant.

Replies appear in your chat. You're notified by push where available and otherwise by email, subject to your notification preferences. With notifications off, you can still return to the chat to check for a reply.

The assistant won't escalate on its own. It might ask "shall I pass this to [owner]?" when things are stuck, and you answer yes or no.

For emergencies (medical, fire, danger), call the relevant emergency service rather than waiting on the assistant or the owner.

Knowing the assistant's limits

Anything outside its topic. A "help with my Mum's iPad" assistant doesn't know about your tax return.
Real-time facts. Some assistants can search the web, many can't. If the answer needs to be from this morning's news, double-check it.
Account-specific information. It doesn't know your bank balance, your booking reference or your calendar unless the owner has given it that context.
Things only the owner knows. If only Dad knows where Dad keeps the spare key, ask for a person.

If an answer feels confidently wrong, push back ("are you sure? I think it's X") or ask for a person.

Privacy basics

Your messages are saved, so you can come back to them and the assistant has context.
Once you ask for a person in a chat, the owner can read that chat.
Chats where you haven't asked for a person aren't open to the owner. They see totals for their assistants, such as how many conversations there have been and how many answers were liked or disliked, but not what was said.
The assistant's memory notes about you are open to the owner, who can read, add, change and delete them. Keep anything you wouldn't want them to see out of a note.
When the owner stops sharing the assistant with you, or you remove it, the notes it kept about you are deleted.
Keep card numbers, passwords and sensitive personal details out of the chat. The assistant doesn't need them, and the owner may see them.

If something in a chat is making you uncomfortable, you don't have to keep replying. Where to take it depends on the problem:

Something the assistant said: use the flag under that answer. Reporting an answer covers what to expect.
The person who built the assistant, or anything else that feels unsafe: email our safety team. Don't raise it with the owner if the owner is the problem.
An ordinary misunderstanding: ask for a person and talk it through with the owner.

Before you hit send

Is my question on-topic for this assistant?
Have I included the specifics (model, version, what I've tried)?
Am I asking one thing, not three?
Would a photo help?