Asking AI to make a message more persuasive polishes the draft. Preparing the listener is what changes the answer. Here are four steps with prompts to paste.
Before you write, fill in a listener brief. Ask AI for the hardest objection in the listener’s voice, answer it, write versions with the same facts, rehearse out loud, and run the send check. Keep your name on every line.
Most people ask AI to make a message more persuasive after the message is already written. What comes back is smoother, a little more confident, and aimed at nobody in particular, because the model has no idea who will read it or what that person is worried about. The draft improves while the odds of a yes stay about where they were.
A team lead sends a well-written case for a new tool to a finance partner and hears nothing for a week. The writing was fine. The case answered the question the lead cared about, which was why the team needs it, and skipped the one the finance partner has to answer upstairs: what does this replace, and what happens if it doesn’t work out. Influence tends to break at that gap between the question you answered and the question the listener is holding.
This guide is about closing it before you write. It works for a hard email, a proposal, a pitch to another department, a memo to your own team or a conversation you’re dreading. You’ll learn four steps, each with something to paste, and the whole sequence fits in one short working session for a message that matters.
The four steps this guide covers. A Future Factors working method based on our own judgment, not a research-validated model.
Two of our other guides cover narrower versions of the same problem. If your listener is your boss, managing up with AI goes deeper on status updates and framing an ask. For a hard conversation with someone on your team, see preparing for a difficult conversation. This one sits across them, for any listener who has to decide something.
Write for the question your listener has to answer, not the one you want to answer.
The first step is a one-page brief on the person you’re trying to influence. AI can’t fill it in for you, because it doesn’t know your finance partner or your head of operations. What it can do is ask you good questions and turn your answers into a brief you reuse for every message to that person. Here is the finished version for the finance example above.
| Field | Filled-in example |
|---|---|
| Who decides, and who else weighs in | The finance partner decides whether it goes forward. Her director signs off, so the case has to survive being retold in a meeting she isn’t in. |
| What they own and are measured on | Keeping spend inside the quarter’s plan and not carrying costs the next quarter can’t absorb. |
| What they already believe about this | Previous tool requests from this team went unused after a few months. |
| What they’re likely to worry about | That this is a new recurring cost, with no way to tell afterwards whether it paid off. |
| What they need to be able to say upstairs | One sentence on what it replaces and how we’ll know in two months whether to keep it. |
| What I want them to do next | Approve a short pilot with a stated end date, not a permanent line in the budget. |
| What I can prove, and what I can’t | I can show the hours the current process takes. I can’t yet show the saving, so the pilot is how we find out. |
The right-hand column is an illustration of the format. Yours comes from what you know about your own listener.
The last row is the one people skip, and it’s the one that builds trust. Naming what you can’t prove yet, and offering a way to find out, gives a cautious listener something to say yes to.
| I need to get a yes from [role, e.g. a finance partner] on [what you’re asking for]. Don’t write anything yet. Interview me, one question at a time, to build a listener brief with these fields: who decides and who else weighs in; what they own and are measured on; what they already believe about this; what they’re likely to worry about; what they need to be able to say to their own boss; what I want them to do next; what I can prove and what I can’t. If I answer vaguely, ask a sharper follow-up. When I say done, write the brief in under 200 words, and mark each line as something I told you or something you inferred. |
Marking inferred lines matters. Anything the model guessed about a person it has not met stays a guess until you’ve checked it.
The brief weighs differently depending on where you sit:
This is the step that changes the result most. Before you ask AI for a better draft, ask it for the hardest objection, spoken by the listener you just described. The reason to ask directly is that assistants tend to agree with whoever is typing. A 2023 study found five state-of-the-art assistants showed sycophantic behavior across four open-ended text tasks, and that responses matching a user’s existing views were more likely to be preferred by human raters.[1] Models change between versions, so I’d read this as a reason to be explicit, not as a measurement of any one tool today.
In practice, that means “is this persuasive?” is the wrong question. It invites a pleasant yes. Give the model a role, forbid the compliments, and ask for what would stop the listener from agreeing.
| Act as [listener’s role], using the brief below. I am about to send you the message that follows. Do not praise it and do not rewrite it. Give me (1) the three objections you would raise, ordered from the one most likely to stop a yes to the least, each in one sentence in your own voice; (2) for each, what I would have to show or say for you to drop it; (3) one thing you would be afraid to ask me in front of other people; (4) any claim in my message you would want evidence for. If an objection depends on a fact you do not have, say so instead of guessing. Brief: [paste] Message: [paste] |
Run it in a fresh chat so nothing earlier in the conversation softens it. Treat the answer as a list of guesses to test, not a forecast.
Ask for the hardest objection before you ask for a better draft.
A product manager pastes a proposal into a chat and asks for something more convincing. The model returns bolder verbs and a tighter opening. Asked instead to answer as the head of engineering, it raises the migration work the proposal did not mention, which is the kind of gap a real conversation would have found. The first request polished the draft. The second found the gap.
What to do with the output: pick the top two objections. Answer them in the message, or decide to hold the answer for the room. If an objection needs a number you don’t have, that’s a gap to fill with real data, and the model shouldn’t paper over it. If every objection is generic (“budget concerns,” “timing”), the brief was too thin. The objections are still guesses from a role description, so a colleague who knows the person is worth a quick read of them.
The same facts rarely persuade three different people the same way, so the third step is writing the message as versions, each leading with what that listener needs first. Keep the facts identical and change the order and emphasis. Here is the same request, a short pilot of a new reporting process, for three different listeners.
| Finance partner | Peer team lead in another department | Your own team | |
|---|---|---|---|
| Open with | What it replaces and the end date | What it changes for their team, in their words | What is changing for them in their week |
| Evidence they’ll trust | Current cost of the process and how the pilot will be measured | A named example from a team like theirs, only if it is real | The reason behind the decision and what was considered |
| The ask, stated as | Approve a time-boxed pilot | Two named people for a short review, once | Try it for the pilot period and tell me what breaks |
| Leave out | Enthusiasm without numbers | Anything that sounds like a decision already made | Detail that only matters to finance |
| Stays identical | The facts, the dates, the commitments | The facts, the dates, the commitments | The facts, the dates, the commitments |
The fixed row is the point: versions differ in order and emphasis and keep the same promises.
Ask the model for versions that differ in what they lead with, and say so in the prompt: “keep every number, date and commitment word for word.” Then compare the versions yourself, because this is where a model’s helpfulness can drift. A softened date or a slightly bigger benefit in the version for the person who wants good news is hard to spot and hard to take back.
Write the version for the person with the least time first. If the ask, the reason and the evidence fit in a few lines, the longer versions follow. For a longer argument, our guide to creating presentations with AI covers the story-first approach.
A message gets you into the room, and the room is where the hardest question usually comes up. The fourth step is rehearsal, and the point is to practice answering, not to memorize lines. A script falls apart at the first interruption. An answer you’ve said out loud a few times holds.
| Play [listener’s role], using the brief below. We are in a [length]-minute conversation about [topic]. Open with a realistic first reaction, then ask me one question at a time. Interrupt if my answer runs long or stays vague. Do not coach me during the conversation and do not write my lines. After [five] exchanges, stop and tell me: which answers were strong, which were vague or defensive, which question I avoided, and the one sentence I should have ready for it. Brief: [paste] |
Use voice mode if your tool has one, since typing an answer and saying one are different skills. Our guide to public speaking with AI has more on practicing out loud.
A short routine keeps it from turning into open-ended chatting:
A manager about to tell a team that a project is being cancelled tends to rehearse the announcement and skip the follow-up question that matters, usually “what happens to my role?” The second round is where that one shows up, because the model asks what the manager hoped nobody would.
The limit is real. A model is more patient than most people, and it has no history with you, no bad day and no read on the mood in the room. Use rehearsal to find your vague answers. It can’t predict how the person will react.
Once the thinking is done, it’s tempting to let AI write the final message too. Some of that is fine. The risk is not only that it sounds generic. The reader may notice, and a reader who suspects a machine wrote it can judge you differently.
Researchers tested this with AI smart replies, the short suggested responses in a messaging app, across two randomized experiments. Having them available made conversations faster and the language more positive. But the more people believed their partner was using them, the less cooperative they rated that partner, a correlational result, while partners who were actually using them were rated as more cooperative.[2] So the cost came from the suspicion, not from the use. Smart replies are not a drafted pitch, so I wouldn’t stretch the numbers further. The direction is still worth planning around.
The practical answer is to make the message yours before it leaves. Do these edits:
For more on keeping your own judgment and voice in a draft, our guides on thinking clearly and communicating like a human and improving your writing at work go further. Then run the send check before anything goes out.
| Check | What a 2 looks like |
|---|---|
| 1. The ask | The listener could say in one sentence what you want them to do and by when |
| 2. The objection | The hardest objection is answered in the message, or you have chosen to hold it for the room |
| 3. The facts | Every number, date and commitment is one you can source or stand behind, and matches across versions |
| 4. Your voice | You can read it aloud without a sentence you wouldn’t say to this person |
| 5. Their time | The length fits how much attention this listener has |
| 6. The question | If asked whether AI helped, you have a plain, truthful answer ready |
Maximum 12. 10 to 12: send it. 7 to 9: fix the lowest lines first. 6 or below: it isn’t ready, and the fix is usually in the brief, not the wording. Bands are a Future Factors working guide, not a benchmark.
The tidiest way to keep the work honest is to write down who does what. For a message that matters, the split is short, and the right-hand columns are where your influence sits.
| What AI does | What the human still owns | How it gets checked |
|---|---|---|
| Interviews you and turns your answers into a listener brief | What is true about the listener, since the model hasn’t met them | Mark inferred lines and confirm them with someone who knows the person |
| Argues the listener’s side and lists objections | Which objections are real, and whether to answer them in writing or in the room | A colleague reads the top two objections, or you test them in the first conversation |
| Drafts versions that lead differently for each listener | Every fact, date and commitment, and what you are willing to promise | Compare the versions line by line against one fixed fact sheet |
| Plays the listener in a rehearsal and flags your vague answers | How you answer, and the relationship with the real person | A second round to confirm the vague answers are now clear |
A Future Factors working view. Adjust the checker to whoever would catch the error fastest.
If your name is on it, you can defend every line without the tool open.
Disclosure is the harder question. Schilke and Reimann ran thirteen experiments, covering tasks from communications to analytics, and found that people who disclosed AI use were trusted less than people who did not. They also found that penalty was smaller than the penalty when someone else exposed the use.[3] Those are judgments by people reading about a stranger, not by colleagues who know you, and the authors say it is unclear whether the effect will fade as AI use becomes ordinary.
I wouldn’t read that as advice to hide it. Follow your organization’s policy, which may require disclosure for clients or regulated work. Where it’s up to you, keep a truthful answer ready: “I used it to test the argument and tighten the draft; the numbers and the ask are mine.” That answer holds up if anyone asks.
Prompts are the simplest part to copy. The harder part is whether you keep doing the preparation a month from now, so measure it the way you would any new working habit. Use: did you run the objection pass before your last few important messages? Persistence: are you still doing it a month later, without a reminder? Impact: did the conversation change, with fewer rounds of clarifying questions, the predicted objection raised and answered, or a faster decision? A short log makes this visible.
| Message and listener | Hardest predicted objection | Raised by the listener? | What happened |
|---|---|---|---|
| Pilot request to a finance partner | A new recurring cost with no way to judge it | Yes, and the end date answered it | Approved as a time-boxed pilot, with a review date set |
| Process change to my own team | What this means for my week | Not at first; two people asked after the meeting | Added a short follow-up note, and the questions stopped |
The rows are examples of the format, not results. Your own log is what shows whether the habit pays off.
For a team, the same habit has a longer path. Training builds the capability: people learn to brief, to ask for the objection and to rehearse. Workflow design gives it somewhere to be used, for example a rule that any proposal asking for budget carries its listener brief. Reinforcement turns it into behavior, such as a manager asking “what was the hardest objection?” in the review. Adoption is what you see when that happens without being asked. Teams can start internally with a shared brief template. Working through your own real scenarios with an experienced trainer shortens the learning curve and gives a team a common way to practice.
This week, pick one message you’ll send that someone has to say yes to. Fill in the listener brief, ask for the hardest objection, and answer it before you send. Log what happens.
Yes, as long as you stay accountable for what the message says. Use AI to understand the listener, test your argument and tighten wording, and keep every fact, promise and judgment yours. Some organizations and clients have disclosure rules, so check yours before you start.
Fill in the brief with what you can see: their role, what that role is measured on, and what they’ve said in meetings or writing. Mark the rest as unknown. A short first message that asks one real question can teach you more than a long, well-argued one.
The risk moves. A leader’s tone gets read closely, so the question to rehearse is the one people won’t ask aloud, usually what the decision means for their own work. Have the model play the person least likely to speak up in the room.
Follow your organization’s policy first. Where it’s your call, answer plainly if asked, and describe what the tool did and what you decided. Announcing it in every message is rarely needed. Denying it when asked directly puts you in the exposed-by-someone-else position that the research found costs more.
Generic objections usually mean the brief was thin. Add one specific fact about the listener, such as what they pushed back on last time, and ask again. Discard any objection that you can’t tie to something real, and where you can, check the rest with a colleague who knows the person.
Sources were read on October 12, 2026: the Sharma and colleagues paper on sycophancy in language models (arXiv, 2023, version dated May 2025), Hohenstein and colleagues on AI-mediated communication (Scientific Reports, 2023), and Schilke and Reimann on AI disclosure and trust (Organizational Behavior and Human Decision Processes, 2025), where we read the journal’s abstract through a bibliographic record and the authors’ university summary because the publisher’s page blocks automated access. The four-step sequence, the filled-in brief, the send check and the influence log are Future Factors working methods based on our own judgment, not research-validated models. The scenes in this guide are illustrations, not client stories. We left out writing-speed and quality figures from workplace AI studies and any survey of how many managers struggle with communication, because we couldn’t confirm them against a primary source today.
Weekly tips. Real examples. Practical help for busy professionals.
We care about your data, check out our privacy policy.