AI Can Predict How They May Feel.

You've written an announcement that seems clear. Then someone reads it and asks, “Are we being monitored now?” That wasn't the message you thought you sent.

Most people ask AI to polish their words: make this shorter, warmer, or less like a committee wrote it during a fire drill. That's useful. But before polishing, ask a better question: How is this likely to land?

A message can make perfect sense to you and still send readers toward the wrong conclusion. A reaction X-ray is a prompt—a set of instructions for your AI assistant—that tests those possibilities before you send.

It isn't mind reading. You're asking AI to examine your language and audience context, then suggest plausible reactions. You still own the thinking, the truth, and whatever happens after Send.

Test what readers hear, not just what you wrote

Polishing asks whether a document sounds good. This review asks what readers might conclude, how they might feel, and what they might do next.

Those are three different questions. A reader might understand a new requirement, dislike it, and follow it anyway. Another might welcome it but misunderstand the next step. If you only ask whether the tone sounds friendly, you'll miss both problems.

Here's a useful way to follow the message through:

Message + audience context + current moment → likely interpretation → likely feeling → likely reaction

The important part is what comes before interpretation. Readers bring yesterday's meeting, unanswered questions, and whatever happened the last time someone announced an “improvement.”

That means the best fix may not be a warmer sentence. It may be information you assumed everyone already had.

Give AI the context readers actually have

You can use this method with any general-purpose AI assistant. Start with a short audience note, not a biography.

Describe what readers know, what they're concerned about, and their relationship to you. Add relevant timing or history, then state what you want them to understand or do.

Keep those facts separate from your hopes. “They should welcome this” tells AI what you want, not what your readers know.

Consider these two notes for the same tracker announcement:

• Context note A: The team requested fewer status meetings, already has access to the new project tracker, and knows this is a two-week pilot.
• Context note B: The team has not seen the tracker, does not know why it is being introduced, and is worried that management may use it to evaluate individual performance.

You're sending the same document, but those readers have very different starting points. You don't need to feed AI someone's private history to make that distinction useful.

The reaction X-ray prompt

Paste your audience note and document into the prompt below. You're asking for a connected review, not a scorecard for every possible feeling.

It stops before rewriting so you can check the suggested changes. That pause matters: a smoother draft can make an unsupported promise surprisingly easy to miss.

Help me check how this message might land before we improve the writing. Be skeptical but fair.
AUDIENCE CONTEXT:
[paste a short, factual context note]
DOCUMENT:
[paste the document]
DESIRED OUTCOME:
[state what the reader should understand, feel, or do]
Walk me through what the message actually says and what readers might think it means. Explain how readers might feel and why, then how they might respond right away, privately, or later. Tell me how confident you are about the important predictions. That confidence is your judgment, not measured odds or mind reading.
When it matters, consider groups who have different information or will be affected differently. Notice where supporters, neutral readers, skeptics, or people most affected might respond differently. Don't force those distinctions if they don't help.
Quote the words behind your concerns and point out missing facts. Look for wording that could sound vague, dismissive, controlling, defensive, manipulative, or unexpectedly threatening. Show me what in the message supports your concern and what you're guessing. What missing fact could change your answer? Give me another reasonable reading, too.
Compare the response I want with the response this draft might get. Explain what needs changing, starting with what would help most and why. Separate wording fixes from things I need to settle first. Flag any fix that needs me to confirm a fact, make a decision, or promise something.
Stop before rewriting. Ask me to confirm or correct your key assumptions and suggested changes. After I respond, revise the message without changing the truth or inventing facts or promises.

One memo, two very different reactions

Imagine a manager plans to send this note:

Starting next Monday, all project updates must be entered in the new tracker. Email updates will no longer be accepted. This will improve accountability, and managers will review compliance weekly. Direct questions to your supervisor.

With context note A, AI might predict cautious relief. The team requested fewer meetings and knows about the pilot. Readers may see a practical improvement, then ask where existing work belongs.

Even here, “review compliance weekly” sounds heavy for a short trial. The tracker appears to have arrived with its own security detail.

With context note B, “improve accountability” may sound like an undisclosed performance policy. Readers could feel anxious or distrustful. They might say nothing publicly while privately speculating, resisting, or doing the bare minimum.

That's a useful distinction: silence is a behavior, not proof of a feeling. If you treat no questions as agreement, you may mistake uncertainty for acceptance.

AI hasn't discovered anyone's secret emotions. It has shown how missing context changes the plausible explanations for the same response.

A better draft may require real decisions

A stronger version could look like this—but only if the sender confirms every detail and commitment:

Starting next Monday, we will run a two-week pilot using the new tracker for project updates. The goal is to replace the Monday status meeting and make current information easier to find. This pilot is not an individual performance score. We will hold a 15-minute walkthrough on Friday, and you can send questions or feedback directly to your supervisor. At the end of the pilot, we will review what worked before deciding what happens next.

This version explains the purpose, duration, benefit, support, feedback path, and what happens afterward. But neither audience note establishes all those facts.

The Monday meeting replacement, Friday walkthrough, performance-scoring statement, and review plan need confirmation. They aren't decorative details. They're promises.

Here's the deeper lesson: sometimes the writing problem is an unresolved management decision. AI can suggest offering a walkthrough. It can't put a real one on the calendar by mentioning it in a paragraph.

If the tracker will influence performance reviews, say so honestly. If that hasn't been decided, don't claim it won't. Reassurance isn't a fact just because it sounds good.

Challenge the forecast before you trust the rewrite

That same caution applies to AI's predictions. It can overgeneralize, miss private history, or misread workplace culture. It can also turn one alarming possibility into an elaborate explanation. A confident table is still not a witness.

Before accepting its advice, ask:

• “Which prediction has the weakest support?” This helps you separate a useful concern from an interesting guess.
• “What missing context would reverse your conclusion?” You may discover that readers already know something you left out.
• “What's a reasonable, less negative interpretation?” A skeptical review shouldn't become a contest to find offense.

Then sort the suggestions by what you can actually do. You can fix vague wording now, but you'll need to verify missing facts. Policy changes and new commitments need someone with the authority to make them.

Use the review to clarify your meaning, not to remove every firm boundary. Sometimes readers won't like an honest message. That doesn't automatically make the message wrong.

Use it where a misunderstanding would matter

This is worth doing for proposals, announcements, sensitive emails, policies, and presentations. It's less useful for routine notes where the context is shared and the next step is obvious. Your lunch confirmation probably doesn't need an emotional forecast.

For contracts, legal notices, or other high-stakes messages, use the review to prepare for qualified human review. Don't let a tone improvement quietly change rights or obligations.

Share only the information needed for the task, using a tool approved for that material. Remove unnecessary sensitive details. If AI needs private facts you can't safely share, take the question to an appropriate person instead.

Your next move: review one unsent email

Pick one message where the reader's reaction matters. Add a three-sentence audience note and run the prompt. Before requesting a rewrite, find the biggest difference between what you intend and what readers might reasonably hear.

If it's wording, revise it. If it's missing information, verify and add it. If it's an undecided promise, resolve that first.

Use the Action Guide to copy the context notes, run the complete X-ray, and compare the two forecasts. Then revise one real document before sending it.

You're not trying to control how someone feels. You're giving them fewer reasons to misunderstand what you mean.

Help me check how this message might land before we improve the writing. Be skeptical but fair.
AUDIENCE CONTEXT:
[paste a short, factual context note]
DOCUMENT:
[paste the document]
DESIRED OUTCOME:
[state what the reader should understand, feel, or do]
Walk me through what the message actually says and what readers might think it means. Explain how readers might feel and why, then how they might respond right away, privately, or later. Tell me how confident you are about the important predictions. That confidence is your judgment, not measured odds or mind reading.
When it matters, consider groups who have different information or will be affected differently. Notice where supporters, neutral readers, skeptics, or people most affected might respond differently. Don't force those distinctions if they don't help.
Quote the words behind your concerns and point out missing facts. Look for wording that could sound vague, dismissive, controlling, defensive, manipulative, or unexpectedly threatening. Show me what in the message supports your concern and what you're guessing. What missing fact could change your answer? Give me another reasonable reading, too.
Compare the response I want with the response this draft might get. Explain what needs changing, starting with what would help most and why. Separate wording fixes from things I need to settle first. Flag any fix that needs me to confirm a fact, make a decision, or promise something.
Stop before rewriting. Ask me to confirm or correct your key assumptions and suggested changes. After I respond, revise the message without changing the truth or inventing facts or promises.