Content Planning and Briefs

Data Story Brief

Plan a clear narrative around a dataset while preserving methodology, uncertainty, exclusions, caveats, comparisons, and accessible visual evidence.

Free editable Markdown · Data journalists, research teams, and editors · Updated July 30, 2026

Download Markdown

Accessible HTML preview

Blank template

The downloaded file contains the same fields in editable Markdown.

Story and dataset

Working headline: [Do not finalize until analysis is reviewed]
Reader question: [One answerable question]
Dataset owner and source URL: [Authoritative location]
Version and access date: [Version; YYYY-MM-DD]
Unit of observation: [Row or record meaning]
Population and coverage: [Who or what may be represented]
Period: [Start, end, and relevant time zone]
Collection method: [How records entered the dataset]
License or permission: [Terms governing use]

Analysis and visual plan

Main finding: [Qualified statement]
Comparison basis: [Baseline, cohort, or period]
Calculation: [Fields, filters, formula, denominator, rounding]
Missing-data treatment: [Method and reason]
Planned chart: [Type and question it answers]
Accessible summary: [Pattern in plain language]
Context source: [Reporting or expertise outside the dataset]
Alternative explanation: [Plausible account to investigate]

Publication checks

The untouched source and working copy are stored separately.
Cleaning, joins, and exclusions are documented.
A reviewer reproduced headline numbers.
Rates include denominators and comparisons include baselines.
Chart axes, units, periods, and sources are visible.
Color is not the only carrier of meaning.
Caveats are placed near the claims they limit.
The headline avoids causal or universal language not supported by the design.

How to use this template

Define the exact analytical question, population, unit, time period, and comparison.
Document dataset origin, collection method, license, version, exclusions, and known limitations.
Reproduce every publishable figure with recorded filters, formulas, denominators, and rounding.
Match each visual and annotation to a specific reader question and accessibility requirement.
Review the headline, narrative, chart, and caveats together so none makes a stronger claim than the data supports.

Let the question lead the analysis

Start with a reader-relevant question that the available data can actually address. “What changed in reported bus delays after the timetable revision?” is more testable than “Why public transport is failing.” Define the unit of observation, population, period, and comparison before calculating. Inspect how the data was collected and what is missing. Administrative records often describe events captured by a system rather than every event that occurred. If the dataset cannot answer a causal question, narrow the language to association or description instead of stretching the analysis toward a more dramatic claim.

Preserve the chain from source to statement

For each number planned for the story, record the original field, filters, exclusions, transformation, denominator, rounding rule, and resulting value. A second person should be able to reproduce it from the stored source. Compare like with like: currencies need dates and bases, rates need denominators, and percentage changes need starting values. Investigate outliers rather than deleting them for visual convenience. When joining datasets, document keys and unmatched records. This chain makes corrections possible and prevents a visually appealing chart from drifting away from the underlying evidence.

Use visuals to expose, not hide, uncertainty

Choose a chart because its structure answers the reader's question. Label units, periods, sources, and important breaks; provide accessible text that communicates the main pattern without relying on color alone. Show uncertainty intervals when relevant and explain meaningful methodological changes. Avoid truncated axes or selective date ranges that exaggerate movement. The narrative should identify what the data shows, what interpretation is reasonable, and what remains unknown. Seek contextual reporting or expert review for mechanisms the dataset does not contain, especially when the story could affect people, communities, or policy decisions.

See the fields in context

Fictional example: library workshop attendance

The East Mere Library dataset and all figures are invented to demonstrate documentation, not to report real attendance.

Question: Did recorded repeat attendance change after weekend workshops were introduced?
Calculation: Unique fictional member IDs appearing in two or more sessions, divided by all IDs recorded in each eight-week period.
Finding: The invented rate moves from 18% to 24%; the brief reports both percentages and fictional counts.
Caveat: Anonymous drop-in visitors are absent, so the data describes registered attendance rather than all visitors.
Visual: Two bars beginning at zero, with counts, percentages, period labels, and a text summary beside the chart.

Frequently asked questions

When does a correlation support a causal headline?

Not by itself. A causal claim requires a design and evidence capable of ruling out important alternative explanations. Otherwise describe the observed relationship and its limits.

Should missing values be removed?

Only after understanding what missingness means and documenting the effect of the decision. Removal can bias a result when records are missing systematically rather than randomly.

How much methodology belongs in the article?

Give readers enough context to interpret the result and link to fuller documentation when needed. Essential limitations should appear beside the relevant claim, not only in a distant note.

Can an AI tool analyze the dataset?

It may assist with code or explanation under an approved workflow, but a qualified person must inspect the data, reproduce calculations, protect sensitive records, and accept responsibility for the conclusions.