Is the voice reading your document, or answering for it?
You're listening to a workplace policy in Word and ask what a sentence means. When the voice starts speaking again, are you hearing the next line of the policy or an AI explanation? Microsoft's existing guide describes Interactive Read Aloud for Copilot licence users: select Unmute to ask a document-related question, then select Play or say ‘Return to reading’ to go back. That's useful help without leaving the document. It also creates a moment I'd want to hear clearly, especially with the laptop screen out of view. I'd want a short spoken introduction such as ‘Copilot's explanation,’ followed by ‘Back to the document’ when reading resumes. On screen, the answer should sit apart from the original paragraph. Someone checking a leave policy shouldn't mistake an assistant's interpretation for a sentence their employer wrote. Those are design suggestions, not a claim that Word currently omits these cues; I haven't tested the feature. The guide documents the controls, not whether listeners can reliably tell the two kinds of speech apart. I'd try that distinction before choosing a more natural-sounding voice. If you use a talking document reader, what tells you it has stopped quoting and started explaining?
Comments
I'd want ‘read me the passage you're basing that on’ to be an easy next step. Suppose the explanation says I qualify for leave, but the policy has a service-length condition. Hearing the original paragraph lets me check that condition without asking the assistant to explain its own explanation. That's a control I'd ask for, not a feature I've tested in Word. A spoken AI label tells me who is talking; getting back to the supporting text lets me question what it said.
For that leave-policy example, I'd check what the listener believes afterward, not just whether they noticed the spoken label. After a short listening session, ask them to explain who qualifies and point out which conditions came from the policy rather than the assistant. Someone might correctly remember hearing ‘Copilot’ and still remember its interpretation as an employer's promise. Compare ordinary read-aloud with the question-and-answer version on similar passages, including the time spent returning to the source. That's a test I'd propose, not a Word result. The useful outcome is a person making the right leave request without having to listen to the whole policy again.
I'd check the privacy wording before using the question button on a sensitive document, too. The same Microsoft guide says ‘Read Aloud doesn’t store your content or audio data.’ That's a statement about Read Aloud; the page doesn't separately explain retention for the Copilot questions and answers it now describes. I wouldn't extend that sentence to the conversation without checking the applicable Copilot terms and the organisation's policy. That isn't evidence that a conversation is retained—it's a gap this guide doesn't settle. Someone listening to a medical letter or an employee's grievance deserves to know whether asking for an explanation changes how that material is handled.