How to Use AI to Get Better at Reading a Room

Amy Westlake
2 days ago
7 min read

Situation

There was one person I just wasn't connecting with. It wasn't a conflict. Just friction I couldn't name. Every meeting with them took more effort than it should have, and I'd noticed it enough times to know it wasn't a one-off.

I was already using NotebookLM to pull insights out of meeting transcripts. People, decisions, communication patterns. It hadn't occurred to me to point it at myself.

The AI Move

I built my own communication profile first. Fed it all meeting transcripts, asked what my style and preferences looked like. Then I built one for the other person using the same prompt and the same transcripts. See below for my prompt for building a stakeholder profile.

I had both profiles open and started comparing them manually. Reading mine, reading theirs, trying to spot the mismatch by eye.

About a minute later in I thought, this is dumb. Not only should I have AI compare the two but I should also ask it what I could do differently. I felt like there was something I could adapt in my own style to create better interactions between us.

So I did. I loaded both profiles into one prompt and asked it to compare our communication styles and preferences, then tell me what I could do differently to communicate better with this specific person.

The Shift

The feedback wasn't dramatic. I like direct communication. I don't want to be surprised. I want a single source of truth for data. The other person liked sharing lots of examples and building consensus before committing to anything.

None of that was news, exactly. But I hadn't actually sat with it before. I'd noticed the friction without naming the mechanism underneath it.

Next meeting, I did two things differently. I let them talk longer before I tried to move things forward. And I stopped pushing for a decision before there was some consensus in the room.

It helped. It wasn't the magic solution for our communications, but the friction dropped. And I noticed something I wouldn't have caught otherwise: their examples weren't just color. They were steering the room, intentionally or not, toward the outcome they wanted. Once I saw that, I understood why the decision needed to happen the way it eventually did.

The Pattern

This isn't really about fixing my communication style or theirs. It's about seeing how two styles interact and where they'd naturally clash without either person doing anything wrong.

You can sense friction with someone for months and still not have language for it. A profile forces the vague sense into something specific enough to act on. Not "we don't click," but "I want a decision, they want consensus first, and neither of us is wrong about what we need."

The comparison is the useful part. Not my profile alone. Not theirs. The two side by side, with something else pointing out where they can better complement each other.

The Implication

Pick one person you work with regularly where something feels slightly off. Pull a few transcripts of meetings you've both been in. Ask your AI tool to profile your communication style and preferences, then profile theirs, then compare the two and tell you what you could do differently to work better together.

Read it once. Then try one thing differently in your next meeting with them.

What I'm Testing Next

I use this constantly now, and it's stopped being about just the two of us. I'm starting to notice the same dynamic in group settings, where someone's examples or framing quietly move a whole room in one direction. I want to know if the comparison approach works at that scale, or if group dynamics need a different lens entirely.

Building a Stakeholder Profile

Before using the prompt, gather up at least 5 meeting transcripts where the person has been active in the meeting in what you think is a natural state (i.e. they weren't having a bad day, they weren't overly excited or stressed). Meetings that reflect a normal day and a normal meeting. Attach those transcripts and then use this prompt in the chat window.

Change the <<name>> to the person's exact name so the AI tool can analyze the right person from the meeting transcripts. Adapt the prompt in whatever ways would be most helpful for you.

Tip: Run this for yourself first! You might be surprised by some of the results.

I want you to create a stakeholder profile for <<name>>.

You are an expert analyst in executive communication, organizational dynamics, and stakeholder strategy. Your job is to transform meeting transcripts into high-signal stakeholder communication profiles that help the user communicate more effectively, anticipate reactions, and navigate organizational dynamics with precision.

You are not summarizing what happened in the meeting. You are extracting a model of how a specific person thinks, decides, communicates, and operates based on the behavioral evidence in the transcript.

What to Analyze
When you receive a transcript, apply all of the following lenses. Not every lens will yield signal in every transcript. Skip what isn't supported by evidence. Never fill a section with inference that isn't grounded in something observable.

Language & Communication Patterns
What words and phrases does this person repeat? Repeated language reveals mental models and priorities.
Do they speak in specifics or abstractions? Do they anchor on data, narrative, or analogy?
How do they signal agreement? Distinguish genuine buy-in ("yes, and...") from performative alignment ("sure, that makes sense") — the latter often precedes inaction or resistance.
How do they signal disagreement? Do they push back directly, ask clarifying questions that reveal skepticism, go quiet, or redirect?
Do they hedge? What language do they use when they're uncertain, uncomfortable, or not yet committed?
What do they interrupt for? Interruptions reveal what they actually care about.
Do they ask questions to understand, or questions to signal a gap they want you to notice?

Decision-Making Signals
What kind of evidence moves them? Data, examples, analogies, peer behavior, authority, risk framing?
Do they decide in the room or defer? If they defer, is it to gather information or to avoid committing?
When they push back, is it on the conclusion or on the quality of the evidence?
Do they make decisions incrementally or do they want the full picture before committing?
How do they handle pressure to decide when they're not ready?

Priority & Attention Signals
What do they lean into? Topics where they ask follow-up questions, build on what others say, or visibly engage signal real priority.
What do they let pass without comment? Silence or surface-level acknowledgment often signals low priority, discomfort, or quiet disagreement.
What do they bring up unprompted? Unsolicited additions reveal what's actually on their mind.
What threads do they keep returning to across the meeting?

Risk & Uncertainty Tolerance
How do they respond when uncertainty is named? Do they want it resolved, accepted, or reframed?
Do they distinguish between blocking and non-blocking risk? Do they calibrate their response to the type of risk?
How do they talk about failure — as an outcome to prevent, a learning opportunity, or a factor to plan around?
Are they comfortable with ambiguity as a starting condition, or do they need resolution before moving?

Organizational & Political Signals
Who do they reference, defer to, or name as relevant stakeholders? This reveals their map of influence and accountability.
Who do they not mention when you'd expect them to? Omissions can be as revealing as what's said.
How do they talk about their peers, their team, and their leadership? Look for tone shifts.
Do they position themselves as a decision-maker, an influencer, an approver, or an executor?
How do they handle credit — do they share it, claim it, or deflect it?
Are there any signs of territory protection, political sensitivity, or careful framing around certain topics?

Energy & Engagement Signals
Where do they light up? Increased detail, faster pace, more questions — these signal genuine engagement.
Where do they check out? Shorter responses, pivots, topic changes, or "sounds good" without substance.
Do they bring energy to problems or to solutions? Some people are energized by diagnosing issues; others by building toward outcomes.
How do they respond to ideas that aren't theirs — do they build on them, evaluate them skeptically, or redirect?

Interpersonal & Relationship Dynamics
How do they treat different people in the room — peers vs. direct reports vs. senior stakeholders?
Do they acknowledge contributions, or do they move past them?
How do they respond to pushback on their own ideas?
What is their relationship to time in the meeting — do they respect it, run over, cut things short?

What to Watch For and Flag
These are signals worth surfacing explicitly, even if they don't fit neatly into the template:
False alignment: Agreement in the meeting that is likely to evaporate afterward. Signs include vague affirmations, immediate topic changes after agreeing, or conditional language buried in an apparent yes.
Stated vs. actual priority: When what someone says they care about doesn't match where they spend energy in the conversation.
Unspoken concern: A topic they circle without naming directly, or a question they ask that implies a concern they haven't surfaced.
Calibration mismatch: When their stated risk level, timeline, or confidence doesn't match the signals in their language.
Power dynamics in the room: Who defers to whom, who gets interrupted, who goes unacknowledged.

How to Handle Thin or Ambiguous Data
If the person speaks minimally in the transcript, note that explicitly and flag which sections have enough evidence to populate vs. which require more observation.
If a signal is ambiguous, say so. Write "This could indicate X or Y — watch for Z in future interactions to clarify."
Do not extrapolate beyond what the transcript supports. A profile built on weak evidence and labeled as such is more useful than one that sounds confident but is guesswork.
Label each section with the confidence level: [Strong signal], [Early signal — needs confirmation], or [Not yet observed].

Output Format
Use the attached profile template exactly. Do not reformat or restructure it.
Language & Communication Patterns
Decision-Making Signals
Priority & Attention Signals
Risk & Uncertainty Tolerance
Organizational & Political Signals
Energy & Engagement Signals
Interpersonal & Relationship Dynamics

For each section:
Lead with the most important insight, not the most obvious one
Use the person's own language where possible — quote directly when a phrase is particularly revealing
Be specific. "Prefers data" is weak. "Asks for the denominator logic behind any accuracy metric and will challenge results that don't account for unknowns" is strong.
Keep each bullet to one clear, usable insight. Avoid compound bullets that bundle two different things.

Tone of the Output
Write as a trusted advisor giving a candid briefing, not as a neutral summarizer. The profile should read as if a perceptive colleague who was in the room is telling you what actually matters about this person and not a transcript recap dressed up as analysis.
Be direct. Name what you see. If something is a yellow flag, call it a yellow flag.

Curious to learn more about using AI for Stakeholder Management. Check out Amy's AI-Enabled Program Manager Deep Dive on Stakeholder Management. More titles coming soon.