The AI-Written Email Works. Until Someone Knows.

TAYLECT: AI Leadership Brief

Where AI strategy becomes a leadership advantage.

Edition 27 · October 2026

AI-assisted messages can be effective and still leave people questioning the person who sent them. For leaders, the bigger question may be what happens when we stop finding our own words.

Executive Brief

The Signal: In a study involving more than 1,100 working professionals, heavily AI-assisted messages from a supervisor were still considered effective. But perceptions of sincerity dropped sharply. With light AI assistance, 83% of employees viewed the sender as sincere. With heavier assistance, that fell to between 40% and 52%. A separate 2026 study involving more than 1,300 people found that revealing AI authorship negatively affected impressions of the sender, even when the message itself remained unchanged.

Why It Matters: AI can help us communicate more efficiently. But efficiency and trust aren’t the same thing. And if we increasingly rely on AI to find the words for us, we should be asking what happens to our own ability to communicate.

The Leadership Decision: Establish where AI should support communication, where it can draft routine messages, and where the leader needs to do the thinking and writing. The distinction matters more than the technology.

Imagine a manager sending a message to the team after a difficult quarter.

It’s thoughtful. It acknowledges the challenges, recognizes individual contributions and thanks everyone for pushing through.

The team appreciates it.

What they don’t know is that AI wrote most of it in about thirty seconds.

Does that change anything?

The message still says the right things. The employees still feel recognized. The manager still intended to express appreciation.

But would they feel the same way if they knew how it was written?

That’s where the research gets interesting.

The message works. The perception changes.

Peter Cardon of the University of Southern California and Anthony Coman of the University of Florida studied how more than 1,100 full-time professionals responded to workplace messages written with different levels of AI assistance.

The scenario involved a congratulatory email to a team. Participants evaluated messages attributed to either themselves or a supervisor.

One finding stood out.

Even when AI was heavily involved, the messages were still considered efficient and effective.

But perceptions of the sender changed.

When a supervisor used AI lightly, for tasks such as grammar and editing, 83% of employees perceived the sender as sincere.

With heavy AI assistance, that figure dropped to between 40% and 52%.

Professionalism ratings also declined, from 95% with light assistance to between 69% and 73% with heavy assistance.

Another finding is worth paying attention to.

People were more forgiving of their own AI use than they were of their supervisor’s.

Think about that for a moment.

We may consider AI a perfectly reasonable tool when we’re using it ourselves, yet expect something different from the people leading us.

Perhaps that’s because a leader's communication carries more than information.

It carries intention.

When someone thanks you, apologizes or recognizes your work, you’re not simply evaluating whether the sentences are well constructed.

You’re also evaluating whether they mean it.

The study has limits. It examined a particular workplace scenario, not every kind of leadership communication. We can’t assume the same results would apply to performance reviews, conflict resolution or difficult conversations.

But the researchers also found that participants considered emotional and relational messages less appropriate for AI assistance than factual communication.

That distinction matters.

What happens when people know?

A separate 2026 study by Jiaqi Zhu and Andras Molnar explored how people respond to messages when they believe AI was involved.

More than 1,300 adults evaluated AI-generated messages, including an apology, under different authorship conditions.

Some participants were told a human wrote the message. Others were told AI wrote it. Some were given uncertain information, while others received no authorship information.

When participants believed AI had written the message, they evaluated the sender more negatively.

Words such as lazy and insincere appeared in their assessments.

When they believed a human had written it, impressions were more positive.

What I found particularly interesting was the group that received no information about authorship.

Their impressions were just as favourable as those of participants who believed a human wrote the message.

The words hadn’t changed.

What changed was what people believed about the effort and intention behind them.

That doesn’t mean every AI-assisted message is dishonest. Nor does it mean leaders must disclose every use of AI.

It does suggest that people care about more than the finished product.

They care about whether someone took the time to mean what they said.

And that’s a difficult thing to automate.

We should be asking another question.

Most discussions about AI-assisted communication focus on the recipient.

Will people trust the message? Will they recognize that AI wrote it? Should the sender disclose its use?

Those are legitimate questions.

But I’ve been thinking about the person writing the message.

What happens to our ability to communicate when we gradually stop doing the difficult part ourselves?

Consider how easily a new habit can develop.

You need to send an email. You open an AI tool, describe what you want to say and receive a polished draft.

You make a few edits and send it.

The next time, you do the same thing.

Eventually, starting with AI becomes automatic.

There’s nothing inherently wrong with that. For routine communication, it may be an entirely sensible use of the technology.

But what happens when the message is difficult?

When you need to apologize sincerely, explain a decision people won’t like, or tell someone they’ve fallen short of expectations?

Those situations require more than polished language.

They require reflection, judgment and sometimes the discomfort of figuring out exactly what you believe.

If AI consistently handles that first step, are we still practicing those skills?

We don’t yet have enough evidence to answer that question specifically for leadership writing.

But research from other areas gives us a reason to pay attention.

The skill question: what the research tells us

Microsoft and Carnegie Mellon researchers surveyed 319 knowledge workers about their use of generative AI.

Participants who reported greater confidence in AI also reported engaging in less critical thinking during AI-assisted tasks.

That finding is an association based on self-reported behaviour. It doesn’t prove that AI causes people to lose critical-thinking ability.

Still, it raises a useful question about what happens when confidence in the tool begins replacing scrutiny of its output.

In another study published in PNAS Nexus, researchers Shiri Melumad and Jin Ho Yun compared how people researched a topic using traditional web search versus an AI-generated summary.

Participants using AI spent less time engaging with the material, reported less ownership of their advice and produced less distinctive responses.

The study involved controlled tasks, so we should be cautious about applying its findings directly to everyday workplace behaviour.

Then there’s an Anthropic experiment involving 52 mostly junior engineers learning an unfamiliar coding library.

Participants who used AI assistance scored an average of 50% on a subsequent comprehension assessment.

Those who worked without AI scored 67%.

But the most useful finding wasn’t simply the difference between the two groups.

It was how participants used AI.

Those who delegated the work tended to understand less—participants who used AI to ask conceptual questions or seek explanations preserved more of their understanding.

Coding isn’t leadership communication. And a short experiment doesn’t establish what happens to someone’s skills over several years.

I wouldn’t make that leap.

What these studies suggest, taken together, is a distinction worth exploring:

AI can help us complete a task without necessarily helping us develop the ability to complete it ourselves.

And the way we use it may make all the difference.

Support. Expectation. Dependence.

This brings me to something I’ve been exploring through the Taylect Decision Dependency Framework.

The framework describes three stages in how our relationship with AI can evolve:

Decision Support → Decision Expectation → Decision Dependence

It’s a leadership lens, not a research finding. But I think it applies particularly well to communication.

Decision Support

You write the message.

AI helps you improve it, challenge your reasoning, identify an unclear sentence or suggest a better structure.

You remain responsible for the thinking, the sentiment and the final words.

Decision Expectation

You begin expecting AI to produce the first draft.

Even when you know what you want to say, starting from a blank page feels unnecessary.

You still review the message, but AI increasingly determines its initial structure and tone.

Decision Dependence

You find it difficult to begin an important message without AI.

You rely on it not just for wording, but to decide how to express appreciation, disappointment, disagreement or empathy.

And because the output usually sounds polished, you may stop questioning whether it actually sounds like you.

These stages aren’t inevitable, and people can move between them depending on the task.

But the progression is worth recognizing.

Dependence rarely announces itself.

It develops through small decisions that each seem perfectly reasonable.

The leadership challenge isn’t to avoid AI assistance.

It’s to recognize when assistance is becoming substitution.

Three decisions leaders should make this quarter

1. Decide which messages deserve your own words.

Not every message needs the same level of personal involvement.

Routine updates, meeting summaries and scheduling communications are reasonable candidates for AI drafting.

Recognition, apologies, performance feedback and difficult news deserve a different approach.

That doesn’t mean AI has no place in preparing those messages.

It can help you organize your thoughts, identify ambiguity or consider how your words might be received.

But the substance should come from you.

If you’re thanking someone, you should know exactly what you’re grateful for.

If you’re apologizing, you should have done the reflection yourself.

And if you’re delivering difficult feedback, you should be able to explain the reasoning without relying on a generated script.

2. Use AI to strengthen your thinking, not replace it.

For an important message, try writing a rough version before opening an AI tool.

It doesn’t need to be elegant.

It just needs to capture what you actually mean.

Then use AI to challenge it.

Ask what might be misunderstood. Ask whether the message is clear. Ask what you’ve overlooked.

You can even ask AI to explain why it recommends a particular change.

The goal isn’t to prove you can write without assistance.

It’s to ensure the thinking remains yours.

3. Apply the transparency test.

Before sending an important message, ask yourself:

Would I be comfortable if the recipient knew exactly how I created this?

This isn’t a research-backed test or a recommendation to label every AI-assisted email.

It’s a personal leadership check.

If the answer is yes, you’ve probably maintained a reasonable level of ownership.

If the answer makes you uncomfortable, ask why.

Perhaps AI did more of the emotional work than you intended.

Perhaps the message expresses a sentiment you haven’t fully considered.

Or perhaps it simply doesn’t sound like you.

Whatever the reason, that discomfort may be telling you something useful.

The real advantage is knowing when not to delegate

I use AI. I see its value. And I believe it will keep changing how we communicate at work.

This isn’t an argument for returning to a world where every email must be written from scratch.

It’s an argument for being more deliberate about what we hand over.

Leaders are already expected to make decisions about budgets, priorities, people and risk.

Communication deserves that same judgment.

Because when someone receives a message from their leader, they’re not just receiving information.

They’re receiving a signal about attention, effort and care.

And sometimes, taking the time to find the right words is part of what makes those words matter.

Key Insight

A message can be effective without strengthening the trust behind it. Leaders need to understand the difference.

The Question

Which messages in your organization would feel different if the people receiving them knew exactly how they were written?

And a more personal question:

When was the last time you wrote something important from a blank page?

What did you notice?

One Resource Worth Your Time

Professionalism and Trustworthiness in AI-Assisted Workplace Writing: The Benefits and Drawbacks of Writing With AI

Peter Cardon and Anthony Coman International Journal of Business Communication

This research supports the sincerity and professionalism figures discussed earlier.

It’s worth reading because the findings are more nuanced than a simple argument against AI-generated communication.

The researchers distinguish between different levels of assistance and message types. That distinction is central to making responsible decisions about AI use at work.

University of Florida research summary: https://news.ufl.edu/2025/08/writing-ai-work/

AI is not a technology race. It is a decision-making advantage.

AI winners won’t write faster. They’ll know which words have to be their own.

Dwayne Taylor Founder & Editor, Taylect

The views expressed are my own and do not reflect those of my employer.