Show the Wire Photo, Then Write the Caption

Opening Scene

A photo desk runs a striking image from a flooded district above the fold, and only below it, in the caption, explains what the reader is looking at and why it matters. The picture does the work of establishing that something real and significant happened; the caption’s job is just to make sure the reader interprets it correctly. Now imagine the reverse: a caption asserting “this flood was the worst in a decade,” printed above a photo that, on its own, doesn’t obviously support that claim. The reader’s trust immediately splits its attention between believing the claim and checking whether the evidence backs it up. The order in which evidence and explanation arrive changes how skeptically a reader receives the point.

Data reports face the identical choice on every chart: show the evidence first and interpret it after, or state the interpretation first and hope the evidence backs it up.

In Plain English

Evidence-first sequencing means presenting the underlying data — the chart, the table, the raw comparison — before stating the interpretation, so the reader can form their own initial impression and then have it confirmed or sharpened by the explanation that follows. This isn’t a rule that evidence must always come first; sometimes, as Article 2 established, the headline claim genuinely needs to lead. The distinction here is at the level of an individual chart or finding within the piece: once you’re past the headline, does this specific piece of evidence get shown before or after its meaning is asserted, and which order actually serves this particular claim?

The Old Way

Reports without a deliberate choice here tend to default to one extreme or the other, often without noticing they’ve made a choice at all:

Interpretation with no visible evidence — a claim stated in prose with a chart included mainly as decoration, positioned such that a skeptical reader can’t easily check the claim against the underlying picture.
Evidence with no interpretation at all — a chart presented and left to speak for itself, on the assumption that the pattern is self-evident, when in practice different readers will draw different, sometimes contradictory, conclusions from the same chart.
Evidence and interpretation in the wrong order for the claim being made — leading with an assertion for a finding that’s genuinely surprising and needs the evidence shown first to be believed, or leading with a chart for a routine, low-stakes finding where a quick stated conclusion would have served the reader faster.

The underlying mistake in all three is treating “evidence” and “interpretation” as one unit rather than two decisions — what to show, and in what order to show it relative to what you’re claiming about it.

What’s Changing (and Why AI Is the Reason)

AI-generated narratives increasingly produce the interpretation first, by default, because language models are built to generate prose conclusions more naturally than they select and place supporting visual evidence. Left unchecked, this systematically biases automated reporting toward the “claim first, evidence after” pattern, regardless of whether that’s the right order for a given finding.
This makes deliberately choosing evidence order a task that increasingly falls to whoever reviews AI-generated narrative output, not just to the person drafting from scratch. A reviewer now needs to ask, for each AI-drafted claim, whether it would be more trustworthy and more persuasive with the evidence surfaced first — a question the drafting model itself isn’t reliably asking.
For surprising or high-stakes findings specifically, evidence-first sequencing has become more important, not less, as AI-generated claims have become more common and more fluent. A confidently worded AI-generated claim reads exactly as authoritative whether or not it’s correct, which means showing the underlying evidence before or alongside the claim is one of the few remaining signals a reader has for judging whether to trust it.

The Metaphor, Fully Extended

Newsroom Element	Data Storytelling Concept
A striking photo run above the fold, caption below explaining it	Evidence — the chart or data — shown before its interpretation is asserted
A caption asserting drama over a photo that doesn’t obviously support it	An interpretive claim stated before its supporting evidence, straining reader trust
A photo run with no caption at all, left to speak for itself	A chart presented with no interpretation, leaving readers to draw their own, possibly wrong, conclusions
An editor choosing photo-then-caption order deliberately per story	Choosing evidence-then-interpretation order deliberately, per finding, not as a blanket rule
A wire service drafting the caption before the photo has even been selected	AI-generated narrative defaulting to interpretation-first prose regardless of whether that order suits the claim

For Beginners: What to Actually Do

For each individual chart or finding in your report, decide deliberately whether to show it before or after stating what it means — don’t default to one order out of habit.
Reserve evidence-first sequencing especially for surprising or high-stakes claims, where a skeptical reader will want to check the picture before accepting the interpretation.
Never leave a chart with zero interpretation attached, on the assumption the pattern is obvious — different readers reliably see different things in the same chart.
When reviewing AI-drafted narrative, check whether a stated claim would be more trustworthy shown alongside its underlying evidence rather than asserted first.

For Practitioners and Leaders: The Deeper Layer

Establish evidence-first sequencing as the default review expectation for any surprising or consequential claim in your team’s reporting, while allowing interpretation-first sequencing for routine, low-stakes findings.
Recognize that AI drafting tools systematically bias toward interpretation-first prose, and build a specific review step to catch and reorder this for claims where evidence-first would serve readers better.
Treat evidence-then-interpretation sequencing as a trust-building tool specifically for claims a skeptical or unfamiliar audience might otherwise doubt.
Train reviewers to ask, of every AI-generated claim, “would I trust this more if I saw the underlying data before the sentence” — and reorder accordingly when the answer is yes.

Quick Recap

Evidence-first sequencing shows the data before asserting its meaning, letting the reader form and then confirm their own impression; it’s a per-finding choice, not a universal rule.
Without a deliberate choice, reports default to unsupported interpretation, uninterpreted evidence, or evidence and interpretation in the wrong order for the claim being made.
AI-generated narrative tends to default to interpretation-first prose, which reviewers now need to specifically check and correct.
Evidence-first sequencing has become a more important trust signal as fluent, confidently worded AI-generated claims have become more common.

Where This Fits in the Series

Article 6 covered the turn as a narrative device; this article covers the ordering choice around any individual piece of evidence within a piece. Article 8 addresses a related discipline: keeping each chart to one story, rather than layering several claims onto a single visual.