The CEO used AI to write his speech manuscript, only to find that his own "voice" was gone.

Key Takeaways

• Starting from the possible problems that CEOs may encounter when writing speeches with AI, this article discusses the loss of leaders' subjectivity in the AI era: the more capable AI is, the more easily leaders' judgment, unique voice and sense of presence will be quietly replaced.
• Leaders' subjectivity is embodied in three dimensions: judgment (who makes the diagnosis when data is insufficient), voice (who conveys meaning), and presence (who experiences subtle signals firsthand).
• This article provides a four-layer collaboration model to help you draw boundaries: what you must do yourself, what AI can handle under your supervision, what can be fully delegated to AI, and puts forward an action checklist. For example, you can write down your working assumptions in text, then ask AI to challenge them, add new perspectives or point out what is missing, so as to exercise your own judgment.

Have you ever used AI tools to write speeches? Most of the time, AI can really deliver "faster, better and more comprehensive" results.

A CEO was preparing for a quarterly speech. His team used AI to pull his past speeches and internal performance memos to generate a draft -- coherent, appropriate, all-inclusive, and accurately reflecting the company's priorities. But after reading it, he felt something was off. In his own words, the draft was too "safe": it just re-spliced the comforting remarks he had said before, without a single sentence responding to the real current dilemmas.

Over the past year, this CEO has deeply embedded AI into the workflow of the executive team: customer insights are shortened from several days to a few hours, strategic documents can automatically synthesize views from countless conversations, and analysis is faster and more comprehensive. But the cost of soaring efficiency is gradually emerging -- the team's thinking begins to converge, and the outputs are increasingly "correct but mediocre". AI is good at strengthening existing cognition, but not good at questioning assumptions, navigating ambiguity, and opening up new directions.

Therefore, as a CEO, you can start thinking from the task of writing a speech: as AI becomes more and more capable, how do you embody your subjectivity?

The answer lies in the following three dimensions: judgment, voice and presence.

Judgment

When data is insufficient to diagnose or solve problems, people need to decide what is truly important. AI can summarize materials and compare solutions, but it cannot understand all the causal relationships that a leader must navigate in high-stakes decision-making. The diagnosis and framing of problems should always be yours.

The division president of a data company once fell into this pit. Before each quarterly meeting, he used AI to generate data summaries, and gradually found a pattern: the focus of the AI summary determined the focus of the meeting. One quarter, the team's profit declined, and the summary started with the rising supplier cost. The team spent most of the meeting discussing this factor, and only dug out the real driving factor at the end: the sales force tilted resources to distributors, who secured deeper discounts and suppressed the profit margin of each transaction. If he had made the diagnosis first at the very beginning, the meeting would have been far more efficient.

Voice

Leaders' communication carries not only information, but also composure, empathy and belief. Your unique voice -- speaking out harsh truths, delivering unwelcome news, and making statements in language with your personal intentions and standards -- is irreplaceable by AI. A 2025 study from the MIT Media Lab found that the more people rely on AI for writing, the lower their sense of ownership over the works they produce. This is a warning signal worth noting for any leader who needs to face doubts from the board of directors, investors or the media.

Presence

Stay close enough to your team to capture things that AI reports cannot tell you: a subtle change in tone, a casual comment that pops up only after the meeting ends, a concern that is mentioned only once but is more important than anything on the agenda. A regional CEO relied on AI-generated summaries for a long time to follow up on customer meetings and employee feedback. He did have a thorough grasp of the facts, but gradually felt disconnected from the key people in the business. Summaries cannot capture the hesitation of customers before answering difficult questions, or the words that employees say only after the official end of the meeting -- and these are precisely the signals most needed for decision-making.

So, how to judge what should be delegated to AI and what you must do yourself?

The starting point of protecting your subjectivity is to identify which level of AI collaboration the current task requires. The following four-layer model: the first three layers correspond to the moments when the leader's judgment, voice and presence must take the lead, and the fourth layer is to delegate low-risk work to AI to free up thinking space for the first three layers.

Layer 1

Your Judgment

In high-stakes decision-making, diagnosis, trade-off, accountability and trust are the core. AI can help at the margins -- summarizing materials, comparing solutions -- but it cannot understand all the causal relationships that leaders must navigate. The framing of problems should always be yours, because it determines everything downstream: which solutions people will take seriously.

The division president of that data company made a very simple correction: before opening the AI data summary, he would write down his own diagnosis -- what he thought the most important problem was and why. Then he let AI test it: sometimes AI confirmed his judgment, and sometimes it raised challenges. Either way, he walked into the meeting room with far more confidence.

Action Checklist: Write down your diagnosis and working assumptions in text first, even if it is only a few sentences; then ask AI to challenge them, add new perspectives or point out what you have missed. This sequence exercises the "muscle" of your judgment -- if you skip this step, your judgment may atrophy rapidly.

Layer 2

Your Voice

Speeches, materials and information, announcements, communication of new directions and other similar tasks fall into this layer. AI can help you build the structure, improve clarity, and even provide alternative expressions -- but the meaning, tone and final wording are yours.

In real speech scenarios, it is recommended not to let AI write the full draft, but let it generate an outline -- you expand the key points on your own. Then read the draft aloud, and the first reading will expose every phrase that "you would never say in a conversation", every sentence that is "too general", and every paragraph that "does not match the weight of the current moment". You may need to rewrite most of the language by hand, but you will definitely get a clear, humanized communication.

Action Checklist: First clarify your intention -- what should the audience understand, feel, and do after listening? List 3-5 key points you must convey, which come from your judgment of the current situation; let AI organize, challenge or polish these points; use the AI outline as a scaffold to write the first draft yourself; read it aloud, mark all language that "does not sound like you", and rewrite it in your own words.

The first two layers talk about "things that must be done by you" -- judgment and expression. The next layer is the work that AI can do, but you have to keep an eye on it.

Layer 3

Your Presence

For repetitive work, there are two biggest risks: first, over-reliance on AI -- the information processed by AI keeps pouring in and all looks correct, so review becomes a routine, and leaders regress from "reading carefully" to "glancing over", from "asking follow-up questions" to "giving approval"; second, sense of distance -- you think you understand the situation, but in fact you are already disconnected.

The regional CEO made this correction: he did not give up AI summaries, but put customer visits and one-on-one direct communication back into his calendar. These interactions restored his "feel" for the business -- he became more confident, more curious, and more willing to explore the next possible steps of the business.

Action Checklist: Before relying on AI summaries, design your review mechanism first -- which parts need human reviewers, which parts require you to go back to the original information source, which parts need special checks for "missing voices" or "buried assumptions"; develop the habit of asking AI tools "what might this summary have missed"; protect unscripted direct contacts.

Layer 4

Free Up Space for Deep Thinking

Low-risk, standardized work, such as formatted routine reports, or answering common questions, can be completed by AI with light supervision or no supervision. Only by moving this type of work out of leaders' daily routine can they have the energy to do things that only humans can do.

When a COO was deciding which tasks could be delegated to AI to run autonomously, he asked four questions:

1. Is the task specific enough and can it be clearly defined?

2. Can the team judge what a good outcome looks like?

3. Is it easy to spot errors and low-cost to correct them?

4. Is this an internal routine task, or will AI represent the company publicly?

Boundaries will not form on their own. Every time you delegate a task to AI, you need to decide in advance: what it can do without asking for permission, what triggers a human review, and who is responsible if something goes wrong.

AI can make leaders faster, more knowledgeable and more powerful. But the same tool can also make their judgment more passive, their voice more vague, and their engagement with the business more "second-hand".

In an era where AI is getting better and better at communication and critical thinking, using AI to amplify your truly unique voice will become an increasingly scarce and valuable capability. Scarcity is the reason why you are irreplaceable.

This article is from the WeChat Official Account "HSG Summit" (ID: Sequoiacap), author: Hong Shan, published with authorization from 36Kr.