Stay in Control of Copilot

Tia Kilby

M365 and Copilot trainer

Copilot can produce polished answers in seconds, but fluent does not always mean correct. Learn three simple checks to assess its outputs and know when greater scrutiny and human judgement are needed.

Data Science and AI

Subscribe to watch

Access this and all of the content on our platform by signing up for a 7-day free trial.

Stay in Control of Copilot

4 mins 14 secs

Key learning objectives:

Apply the source, sanity and audience checks to Copilot outputs
Identify uncertainty, assumptions, omissions and information that requires verification
Match the level of human review to the sensitivity and risk of the task

Overview:

The most important Copilot skill is not generating an answer—it is knowing whether that answer is safe and appropriate to use. Source, sanity and audience provide a simple quality check: trace important claims, challenge anything that does not look right, and shape the output for its intended reader.

Copilot can also smooth over uncertainty, omit caveats or make assumptions without making those gaps obvious. The level of review should therefore match the stakes. Internal, low-risk work may need only a quick check, while client-facing, regulated or decision-driving outputs demand deliberate verification and human sign-off.

Subscribe to watch

Access this and all of the content on our platform by signing up for a 7-day free trial.

Summary

Why can't you simply trust a polished Copilot response?
Copilot can combine information from multiple sources and present it in clear, confident language. That fluency can make an answer appear more complete or reliable than it actually is, so reviewing the output remains essential.

How can you establish where an important claim came from?
Start by tracing figures, names, dates, policies and other important details back to their underlying source. If you cannot establish where a claim came from, treat it as unverified regardless of how confidently it is presented.

What should you do when something doesn't look right?
Use your own knowledge and professional judgement. If a figure seems unusual, a date does not make sense or a summary conflicts with what you know, investigate rather than assuming Copilot must be right.

Why does the intended reader matter?
An output suitable for your own notes may not be appropriate for a colleague, executive or client. You remain responsible for deciding the right tone, level of detail, framing and emphasis for the person who will receive it.

What makes incomplete information particularly difficult to spot?
Copilot can make uncertainty appear tidy by presenting estimates as facts, combining different sources into one statement or leaving out an important caveat. Asking what it is assuming, what is uncertain and what might be missing can help expose these gaps.

How should the level of checking change with risk?
The higher the stakes, the more deliberate your review should be. A low-risk internal summary may need a quick read-through, while client-facing, regulated or decision-driving content requires careful verification and human judgement before it is used.

Subscribe to watch

Access this and all of the content on our platform by signing up for a 7-day free trial.

Tia Kilby

Tia Kilby is a Microsoft 365 and Copilot expert with extensive experience helping organisations and professionals understand how AI can transform everyday ways of working. She specialises in practical applications of Microsoft Copilot, showing users how to work more efficiently, automate routine tasks, and get greater value from the tools they already use. With a strong focus on real-world use cases, Tia brings complex AI concepts to life through clear, practical guidance and demonstrations. Her expertise spans Copilot across Microsoft 365, including applications such as Outlook, Word, PowerPoint and Teams, with a particular emphasis on helping users build the skills and confidence to use AI effectively in their day-to-day work.

There are no available Videos from "Tia Kilby"