Insight
Keeping Humans in Control: Why Every AI-Assisted Email Should Be Reviewed

Margaret Jhoy

AI can help a team reply faster than a person can start from a blank page. That speed is useful only if someone still reads the message before it goes out. A message sent under your organization’s name is your organization’s message, whether a person typed every word or not.
The legal profession has been unusually clear about this, and the underlying point applies well beyond law firms. In July 2024, Bloomberg Law reported in ABA Warns Lawyers to Mind Ethics Rules While Using Generative AI that the American Bar Association told lawyers to fully consider their ethical obligations when using these tools. The risks, the ABA said, range from “hallucinations” to confidentiality breaches, and competent representation requires skill and judgment.
A year earlier, in ChatGPT Doesn’t Have Ethical Obligations, But Attorneys Do, attorneys Natalie Pierce and Stephanie Goutos wrote in Bloomberg Law that lawyers should never assume AI output is accurate. They argued that it should be scrutinized as if a junior colleague or a paralegal had produced it.
They also noted that the duty to supervise non-lawyer assistance extends to tools as well as people, which means the work still has to be reviewed. Failure to do so, they warned, can have serious consequences.
AI output is a starting point
That is the right standard for any team that sends careful email, including legal aid organizations and enterprise support groups. AI can help with the volume, but the result is still a starting point.
It can save time, but what it produces is still only a starting point.
A person on the team should review it. They know the relationship, the exceptions, and the tone that is appropriate in this particular case. They can catch a missing fact, a badly placed sentence, or a reply that is technically complete and still not quite right.
Why sending without review is not a shortcut
It is tempting to think that a careful tool ought to be enough on its own. In ordinary cases, an AI-assisted reply may be close. In the case that is slightly unusual, a message that goes out unreviewed can create a problem that takes much longer to unwind than a short review would have taken.
Speed without review also creates a different kind of risk: people stop reading closely because the message usually looks fine. The review habit is what keeps the quality steady. Once that habit slips, errors become harder to notice.
What staying in control looks like
The practical design is simple. The assistant lives in your email and works around the clock.
A person reads anything that is going out, edits it if needed, and sends it. Nothing goes out without that approval.
This is not a lack of trust in the tool. It is how responsible teams use any assistance, human or otherwise. The tool can help with the work, and the person decides whether a message should be sent.
For professional services firms, that review also supports the duties they already have to clients. For other careful teams, it is a basic safeguard. Your name should leave your organization only after a person has looked at the message.
Bruce is built so that nothing goes out without approval. It lives in your email, takes repetitive inbox work off your team’s plate, and leaves the final decision with a human being.
If you would like to see how review-first email assistance works in practice, you can book a demo.
