Whether you’ve just uncovered a key data insight or have already walked through chart selection and storytelling, the challenge is the same: How do you communicate your message so your audience understands it and takes action?
In this tutorial, we’ll look at how to use Large Language Models (LLMs) like ChatGPT, Claude, or Gemini to communicate your findings, especially when you’re short on time, working across teams, or tailoring your message for different audiences.
To illustrate the techniques, we’ll use an example from the Superstore dataset — a fictional retail company that sells furniture and technology products to consumers.
In a previous tutorial, we analyzed the relationship between discount levels and profit margins. The key finding? When discounts exceed 30%, profitability drops sharply.
You don’t need to be familiar with that full analysis; we’ll use this one insight throughout the tutorial to show how LLMs can help you communicate clearly across different audiences. Because in most real-world scenarios, it’s not the chart that needs tweaking: it’s the message.
This is where these tools shine, giving you fast, flexible ways to shape your message without starting from scratch, even when time is tight.
By the end of this tutorial, you’ll know how to use LLMs as powerful writing assistants that can:
- Translate insights into plain language
- Rewrite messages for different tones and audiences
- Summarize chart findings quickly and clearly
- Brainstorm stronger slide titles, email blurbs, and executive summaries
You chose the right chart. You designed it for clarity. You told a compelling story. Now let’s use an LLM to help make sure your audience hears it, understands it, and acts on it.
How to Write a Good Prompt
Before we look at specific use cases, it helps to understand how prompting works. Writing a clear prompt is less about giving instructions and more like collaborating with a junior teammate who’s fast, flexible, and eager to help, but needs some direction. If we simply say, “Summarize this chart,” we might get something serviceable, but flat or generic. To get results that sound more confident and audience-ready, we need to give just a bit more framing.
Let’s walk through how to do that using a key insight from our Superstore scenario: the same one we introduced earlier about discounting and profitability.
The scatter plot shows how profitability declines as discounts increase. The pattern is clear: profit margins remain fairly stable until about a 30% discount, after which they begin to fall sharply.
Now, let’s say we want to share this finding with a regional manager who isn’t deeply familiar with analytics or very comfortable with math. We could try prompting an LLM in a few different ways, and the results we get will depend on how well we set up the context.
A vague prompt like “Explain this chart” doesn’t give the model anything to go on. It doesn’t know who the audience is, what kind of tone to use, or even what kind of output we’re looking for. We’ll likely get something robotic, cluttered with irrelevant details, or framed in a way that doesn’t match our communication goal.
Now compare that to a more thoughtful prompt:
“I’m preparing a short update for our regional manager. The chart shows that once discounts exceed 30%, profit margins decline significantly. Can you write one clear, confident sentence that summarizes this finding?”
Here, we’re giving the model:
- The basics of what the data shows
- Who the audience is
- What tone we want
A typical LLM response might be something like:
“Discounts above 30% consistently reduce profit margins, signaling a need to reevaluate current pricing strategies.”
It’s succinct, clear, and aligned with our intended audience. From here, we can keep iterating: ask for alternatives, try different phrasings, or test versions for other communication formats like slide titles or email blurbs.
One of the most useful things about working with an LLM is this flexibility. You’re not trying to get the perfect output on the first try — you’re collaborating. If the tone isn’t quite right, you can ask for something more casual. If the summary is too long, you can say, “shorten this to a slide title.” The goal isn’t to treat the model like a one-and-done answer machine. It’s to treat it like a second pair of eyes, one that can help you rework and refine your message until it feels right.
In the next section, we’ll look at real before-and-after examples using our Superstore scenario. You’ll see how small prompt changes can make a big difference, and how you can apply these same strategies to your own data stories.
What LLMs Can Help With
Starting with the insight “When Superstore discounts exceed 30%, profit margins start to drop significantly,” let’s walk through how this message can be adapted for different communication contexts. We’ll see how an LLM can serve as a flexible writing partner that helps us translate, reframe, and sharpen our message without starting from scratch each time.
To show how that works in practice, we’ll walk through four versions of the same insight, each tailored to a different audience and communication format: a team Slack message, a slide title, a stakeholder summary, and a campaign headline.
A quick note to the manager
We’re drafting a quick Slack message to our direct manager. It’s an FYI, not a formal report; we just want to flag something important from our analysis.
We write:
“Items discounted above 30% are often unprofitable.”
It’s technically true, but flat. We ask the LLM:
“Make this clearer and more confident for a quick team update.”
It responds:
“Quick insight from our discount analysis: Products with discounts over 30% consistently lose money — may be time to revisit our pricing thresholds.”
Why this prompt works:
- It sets the goal (“make this clearer and more confident”)
- It gives context (“team update”)
- It invites tone improvement without needing a rewrite from scratch
The LLM response feels more action-oriented and signals a possible next step, without overstating the case.
A slide title for an all-hands meeting
We also want to share this insight with the whole company at the monthly all-hands meeting. We have one line at the top of a slide. It needs to be crisp and clear. We could title the slide something like, “Discounts vs. Profitability”, but that lacks urgency and doesn’t help the audience quickly grasp the point.
We can get LLM support with a well-crafted prompt:
Provide three versions of a slide title for our company all-hands based on the datapoint “Items discounted above 30% are often unprofitable.”
The LLM gives us strong options:
“When Discounts Cut Too Deep”
“Profit Margins Disappear After 30%”
“The Hidden Cost of Steep Discounts”
Why this prompt works:
- It requires the LLM to provide multiple options, allowing us to select (or refine) the one that works best for us.
- It asks for a specific format (“slide title”) and a clear message (“discounts above 30% reduce profit margins”).
- The LLM fills in tone and emphasis, helping us explore which framing resonates best for a wide audience.
A summary for regional directors
We’re sending a short email to regional directors summarizing key findings from recent pricing analysis. They’re busy, so we need to sound polished and decisive.
We start with:
“Discounting patterns are affecting profit.”
Then we try this prompt:
“Write two sentences summarizing that discounts above 30% are typically unprofitable. Audience is senior managers. Tone should be direct.”
The LLM offers:
“Our analysis shows that product discounts above 30% consistently lead to negative margins. Adjusting discount limits could help preserve profitability across categories.”
Why this prompt works:
- It defines the audience (“senior managers”)
- Sets a clear task (two-sentence summary)
- Specifies tone (“direct”)
These boundaries help the LLM avoid fluff and stay focused. Now we have a summary that sounds executive-ready and a prompt we can reuse across other findings.
Helping a colleague shape a campaign message
To see another application, a marketing teammate has asked us for a quick summary of the discount analysis to help shape a new promotion strategy. They’re not looking for hard numbers; they just want a few message ideas for the campaign.
We tell the LLM:
“Suggest 3 short headlines based on the insight that 30% is the threshold for profitable discounting.”
It replies:
“30%: The Sweet Spot for Smart Discounts”
“Go Big — But Not Too Big”
“More Isn’t Always More: Where Discounts Hurt Profits”
Why this prompt works:
- It keeps the task open-ended (“suggest 3 headlines”) while still providing a key constraint (the 30% profitability threshold). This gives the LLM space to explore language variations without veering off-message.
Now our analysis is clear and ready to support front-line messaging.
In each case, the LLM isn’t doing our thinking for us. We’ve done the hard part: finding the pattern. What the LLM offers is a fast, flexible way to shape our message for different situations, tones, and audiences. It helps us say what we mean, with a little more precision and a lot less effort.
With the right framing, we can get results that sound more human and more useful.
What LLMs Can’t Do Well
Large language models can be powerful writing assistants, but they’re not analysts, subject-matter experts, or decision-makers. Just like a well-meaning teammate, they can help you communicate more clearly, but they’re not always right, and they don’t always know what matters most.
Here are a few important things to keep in mind.
LLMs don’t know your data
They don’t understand your business context, and they don’t validate the numbers you share. If you ask a model to analyze raw data, it may confidently generate a conclusion, even if that conclusion is incorrect or based on a misunderstanding of the input. That’s why in this tutorial, we’ve focused on using LLMs to communicate findings, not generate them.
You are the expert on what the data means. Use the model to help shape the message, not discover it.
LLMs can be vague, overly generic, or factually off-base
Especially when prompts are vague or lack context, LLMs may return boilerplate responses that don’t match your voice or audience. If something feels off, trust your instincts. Don’t hesitate to rephrase, clarify, or simply start again.
Also, be cautious about using any numbers the model suggests. LLMs can confidently generate information that sounds plausible but isn’t accurate — a phenomenon often called “hallucination.” It’s usually best to provide key figures yourself and treat model-generated stats or comparisons as placeholders.
Avoid sensitive or proprietary information
Most public-facing LLM tools store and learn from inputs to improve future performance. Don’t enter company-sensitive data, personal information, or anything you wouldn’t share externally. If your organization uses private or internal LLM tools, follow your team’s guidelines around safe usage.
In short: treat the LLM like a smart, fast external contractor who doesn’t know your business, your stakeholders, or your goals, unless you tell it. The better you frame the task, the more useful its responses will be. And the more critically you review its output, the more confident you’ll feel putting those responses to work.
Recap and Takeaways
In this tutorial, we explored how Large Language Models can support the communication side of data storytelling. Once you’ve done the analysis and built your visuals, an LLM can help you:
- Translate insights into clear, plain language
- Adjust tone and emphasis depending on the audience
- Summarize or rephrase findings for different formats
- Brainstorm strong messaging, from subject lines to slide titles
You already know your data. The LLM simply helps you express what you know more clearly and with less friction.
Along the way, we also covered how to write effective prompts by thinking of LLMs like collaborators: Give context, set the goal, ask for what you need, and iterate. You’re not aiming to get it perfect on the first try. Instead, it’s about shaping your message with more clarity, confidence, and control.
With thoughtful use, LLMs can help you bridge the last gap between data and decision by refining your message so it lands with the people who need it most.
