Introduction: Why data storytelling with LLMs is essential today
Organizations now have more data than ever, but simply displaying charts is no longer enough to engage non-technical audiences. Data storytelling with LLMs bridges the gap between raw numbers and understandable narratives, transforming datasets into fluid, evidence-based stories. In this article, you’ll learn how to structure effective prompts, leverage cutting-edge models like Z.ai’s GLM-5.3, and apply industry best practices to create data stories that inform, engage, and drive action.What approach should you take to get started right away?
1. Prepare your data to be LLM-friendly
- Normalize schemas:Convert CSV, JSON, or relational databases into a single structured format (e.g., lists of dictionaries) that an LLM can easily analyze.
- Remove non-essential data:Keep only relevant columns, normalize field names, and remove missing values or irrelevant footnotes.
- Add metadata:Include brief descriptions of fields, units of measurement, and source. This helps the LLM understand context, as demonstrated by Samsung’s recent healthcare intelligence models that analyze biosignals with clear metadata embedded.
Code example (Python)
import pandas as pd
from pathlib import Path
def prepare_dataset(filepath):
df = pd.read_csv(filepath)
# Rename columns consistently
df.columns = [c.strip().lower().replace(' ', '_') for c in df.columns]
# Add a description column
df['field_description'] = df.apply(lambda row: ', '.join([f"{col}: {row[col]}" for col in df.columns if col != 'field_description']), axis=1)
# Save the cleaned dataset
output_path = Path('data/cleaned_data.json')
output_path.parent.mkdir(exist_ok=True)
df.to_json(output_path, orient='records')
return output_path
prepare_dataset('data/raw_metrics.csv')2. Write a prompt that guides the LLM toward storytelling
A well-structured prompt is the heart of data storytelling. Use the schema format to provide the LLM with a clear goal, audience, and desired tone.Example prompt for a sales dashboard
You are an experienced business analyst specializing in data storytelling.
Provided data: (paste data here or reference file path)
Task: Create a short natural language story (max 200 words) that highlights:
1. Total sales and year-over-year growth rate.
2. The best-performing quarter and why it succeeded.
3. A warning about any negative trends.
Specifications:
- Audience: Non-technical executives.
- Tone: Professional yet engaging.
- Output: Only the story, no tables or lists.
Story:This template works particularly well with large models optimized for complex tasks, such as Z.ai’s new GLM-5.3, which excels at multi-step tasks requiring extended memory. GLM-5.3 can handle the entire data interpretation workflow without the need for further fine-tuning.
When to use LLMs for data storytelling
- Dynamic reporting:When weekly dashboards need to be updated with new data, a prompt can automatically regenerate the narrative.
- Cross-functional communication:Transform technical industry data (e.g., biosignals used by Samsung’s healthcare models) into stories understandable to patients.
- Investor presentations:Generate concise, data-driven narratives that highlight growth trajectory and competitive advantage.
Okta recently demonstrated how using the Model Context Protocol (MCP) can reduce token call costs for AI agents, which is particularly useful when batching multiple data storytelling prompts. By using Okta’s identity-scoped tool lists, you can optimize the number of LLM calls while maintaining narrative quality.
Why LLM-based data storytelling works in 2026
Today’s language models do more than summarize; they capture context, tone, and visual style, making stories more engaging. Here are three reasons why this approach has become the standard:
- Scalability:Z.ai’s GLM-5.3 can handle large datasets without requiring model retraining, accelerating story creation.
- Interdisciplinary expertise:Samsung’s healthcare intelligence models demonstrate that advanced data analysis can be adapted to non-technical contexts, just like data storytelling.
- Cost efficiency:Okta’s MCP optimizations reduce API call costs, making LLM-based storytelling accessible to medium and small teams.
Concrete actions to take today
- Identify a use case:Choose a dataset that is currently overlooked in reports.
- Clean and normalize:Use the preparation script to make data LLM-friendly.
- Experiment with prompts:Start with the prompt template above and adapt the language to suit your audience.
- Validate and iterate:Compare the generated narrative with a traditional report; gather feedback and refine the prompt.
- Automate:Integrate the workflow into a Directed Acyclic Graph (DAG) using tools like Prefect or Airflow to run the prompt daily on new data.
By following these steps, you’ll transform raw data into engaging narratives that drive decision-making and meaningfully engage your audience.
Conclusion: Transform data into impactful narratives
Data storytelling with LLMs is no longer a luxury but a necessity for any organization that wants to communicate information clearly and quickly. By leveraging cutting-edge models like GLM-5.3, applying prompt engineering best practices, and utilizing optimizations like Okta’s Model Context Protocol, you can create data narratives that inform, engage, and drive action.
Start today: prepare your data, create an effective prompt, and watch as the LLM transforms numbers into impactful stories.
References and further reading
- Samsung Research presentation: Healthcare AI models for biosignal analysis.
- Okta Engineering: Model Context Protocol (MCP) and token call cost reduction.