Get better answers from the AI assistant

We've learned a lot from thousands of real questions. Here's what works

These tips come from analysis of real usage data - what customers actually ask and what gets the best results. A few small changes to how you phrase a question (prompt) can make a big difference.

Keep it short

The most reliable prompts are all under ten words - clear, direct and single-minded. 

Extra words add ambiguity, and ambiguity adds risk.

One question at a time

Compound questions ⇨ two or more requests in one message ⇨ success can drop by 5%, at worst by over 20%. 

Ask one thing, get your answer, then ask the next.

The assistant remembers your conversation so you won’t lose any context.

Use specific dates

Actual dates beat relative ones

Name who you mean

Use your system’s language

The closer your question matches your actual configuration ⇨ department names, team names, absence types ⇨ the better the results.

If your department is called “Customer Services” in Ciphr, say “Customer Services” rather than “Support team”

Once you get a feel for how your data is labelled, the assistant becomes much more reliable.

When it doesn’t quite get it right

A small change in wording often makes the difference. 

One customer refined their absence query three times, then completed in minutes what normally takes hours.

Start a new chat for a new topic

This is great for drilling down, but it can get confused if you switch subjects mid-flow.

Moving from absence to salary? Start a new conversation. One click, clean slate.

It’s learning, as well as you and us

If it gets something wrong, say so in the chat. 

Think of it like a keen new starter still finding their feet. 

Every bit of feedback helps the team know where to focus next.

Quick reference