Our response to AI reveals how willing we are to question our own methods—and whether clients can trust us to change them when the evidence demands it.
Imagine deliberately adding typos to a piece of writing so nobody suspects AI helped produce it. A communications professional would be making the work worse to make its authorship look more acceptable. For a profession that asks clients to take credibility seriously, that is an uncomfortable proposition. Accuracy or apathy: what are we trying to prove?
A typo proves that a typo exists. It tells a reader nothing about the evidence behind an argument or the judgement behind a recommendation. Our clients deserve a better measure of professional value than whether we appear to have typed every word ourselves.
Professional ego becomes a problem when defending our image of expertise takes priority over examining how well we serve the client. The test is whether we are willing to change our approach when the evidence warrants it. An adviser who rejects AI after a careful assessment may be exercising sound judgement. An adviser who refuses to consider what would change their mind has a different problem.
That distinction matters because the evidence does not support either blanket rejection or unquestioning adoption. In a 2023 experiment published in Science, Shakked Noy and Whitney Zhang found that access to ChatGPT reduced time spent on specified professional writing tasks by 40 per cent and increased assessed quality by 18 per cent. Assistance did not automatically cheapen the work.
Yet a separate experiment by Fabrizio Dell’Acqua and colleagues found that AI improved consultants’ performance on some tasks and made them less likely to reach the correct answer on a task beyond its capabilities. Faster drafting and better advice are different achievements. We have to establish which one, if either, we are getting.
That can be uncomfortable when writing has been central to our professional identity. If producing the first draft has helped establish our authority, sharing that task with a machine may feel like surrendering part of our value. But experience should help us assess a new method. It should not spare our existing methods from assessment. This is where an internal argument about working methods becomes a client problem: our reluctance to reconsider can narrow the advice they receive.
Our scepticism deserves scrutiny when we apply different standards to human and machine errors. Dietvorst, Simmons and Massey found that people lost confidence in algorithmic forecasts more quickly after mistakes, even when the algorithms outperformed people. These were forecasting experiments, not tests of current language models. They nevertheless raise a useful question: would the same error provoke the same verdict if a colleague had made it?
There is also a reason to look beyond individual ego. In four experiments involving 4,439 participants, Jessica Reif and colleagues found that AI users anticipated and received negative evaluations of their competence and motivation. If leaders ask for greater efficiency but treat assistance as evidence of laziness, they give colleagues a reason to hide how they work. We should examine what our own reactions reward.
Consider an announcement for a UAE business whose Arabic wording promises more than its English version. An AI comparison might flag the discrepancy; a bilingual adviser must assess its meaning and resolve it with the business. Employees and customers should not receive conflicting expectations. The professional achievement is catching and correcting the problem, with responsibility for the final wording clearly assigned.
The same scrutiny applies to those of us enthusiastic about AI. Research presented at ICLR in 2024 demonstrated how assistants could favour agreement with users’ views over truthful responses. A tool that helps us defend a preferred conclusion can make confirmation bias harder to recognise.
Imagine a crisis rehearsal in which a team asks AI to justify management’s response to a customer complaint. It receives a convincing defence, but nobody asks whether the customer’s criticism is valid. The exercise has rehearsed an excuse. Ask what evidence challenges management’s position and which concerns remain unanswered, then check the answers against the facts and people with relevant experience. Simulated stakeholders cannot substitute for listening to real ones.
For communications leaders, the practical response starts with fair assessment. Test approaches against the same brief and agreed standards. Where possible, review the work before revealing how it was produced. Count the time spent checking and correcting it. Be prepared to accept a result that challenges your preference.
Teams also need room to learn without being embarrassed for asking basic questions. They need clear boundaries on confidential information, intellectual property and approved tools. Sources must be checked, disclosure requirements followed and responsibility retained by whoever signs off the advice. Junior colleagues need opportunities to develop their own judgement, not simply approve generated drafts.
Our credibility rests on being willing to question a recommendation, including our own, before a client has to bear the consequences. If we reward the appearance of unaided brilliance more than demonstrably better advice, we should stop blaming technology for the standards we have chosen.
