The Editorial Desk: A Review Process for Data Narratives

Opening Scene

No story in a functioning newsroom goes from a reporter’s keyboard straight to print. It passes across a desk first — an editor who checks whether the lede is actually the lede, whether the nut graph earns its place, whether the structure holds up, and whether any claim reads more confident than the underlying reporting actually supports. That review isn’t a formality or a bottleneck; it’s the mechanism by which a newsroom catches its own mistakes before a reader does. A reporter who’s spent three days inside a story loses the ability to see it the way a stranger will. The editor’s whole value is being that stranger, on purpose, before the story goes out.

Most data teams have a rigorous review process for their code and their pipelines. Very few have an equivalent review process for the narrative wrapped around the numbers those pipelines produce.

In Plain English

An editorial review process for data narratives is a deliberate, structured step — separate from the technical accuracy review a data pipeline might already get — that checks whether a report’s story is actually well told: is the lede correctly identified, is it stated first, is the context accurate and fairly chosen, does the structure hold up under skimming, and does the language’s confidence level match the underlying evidence’s actual strength. It requires a reviewer other than the original author, because the author’s closeness to the analysis is exactly what makes them least able to spot where the narrative has gone wrong.

The Old Way

Without a dedicated review step, data narratives tend to ship after only one of these checks, mistaking it for the whole job:

Technical review only — someone checks that the query is correct and the numbers reconcile, which catches data errors but says nothing about whether the resulting narrative actually communicates the right thing to the right reader.
Self-review by the author — the same person who wrote the analysis also proofreads the narrative around it, which rarely catches structural or framing problems, since the author already knows what they meant and reads past ambiguity a fresh reader would trip on.
No review at all, just a deadline — reports ship the moment the numbers are finalized, with narrative quality treated as whatever came out in the first draft, because no one owns the specific job of checking it.

Each of these substitutes a different, real check for the one that’s actually missing: whether the story, not just the data behind it, is right.

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

AI-generated first-draft narratives have made a dedicated editorial review step more necessary, not less, because a fluent AI draft can pass a casual read without anyone having actually verified its structure or claims. A confidently written paragraph reads the same whether it’s well-supported or subtly wrong, which raises the cost of skipping a real review specifically for AI-assisted output.
AI tools can now assist the review itself, flagging structural issues — a missing nut graph, an unstated comparison, an unordered section — at a scale that makes review of high report volume newly practical. This doesn’t replace the human editorial judgment about whether a given framing is honest or a given anchor stat well chosen, but it can triage which reports most need a human reviewer’s limited time.
The volume of reports being generated has grown faster than the number of people available to review them, making review capacity itself a real operational constraint. Teams now need to decide deliberately which reports get full human editorial review, which get AI-assisted triage only, and which ship with no review at all — a prioritization decision that didn’t used to be necessary when report volume was naturally limited by manual analyst time.

The Metaphor, Fully Extended

Newsroom Element	Data Storytelling Concept
A story passing across an editor’s desk before it runs	A data narrative passing through a dedicated review step before publication
The editor as a deliberate stand-in for the reader’s fresh eyes	A reviewer other than the original author, checking what the author is too close to see
Checking the lede, the nut graph, and the structure, not just the facts	Reviewing narrative quality specifically, separate from technical data accuracy review
A copy desk triaging which stories most need a senior editor’s attention	AI-assisted triage flagging which reports most need scarce human review time
A reporter’s own proofread catching typos but missing structural problems	Self-review by the original author catching surface issues but missing framing problems

For Beginners: What to Actually Do

Before shipping any report you’ve written, get a second person — ideally someone unfamiliar with the analysis — to read it and tell you what they think the point is; if it doesn’t match your intent, the narrative needs work, not just the reader.
Treat narrative review as a distinct checklist from data accuracy review: lede placement, context, structure, and confidence-matching-evidence are different questions than “is the query correct.”
When reviewing an AI-drafted narrative, read it as skeptically as you would a stranger’s work, since fluency in the writing tells you nothing about whether the underlying claims are well supported.
Build the habit of asking “what would an editor cut first” of your own drafts — it’s a fast way to find the parts that aren’t earning their place.

For Practitioners and Leaders: The Deeper Layer

Establish a genuine editorial review step for data narratives as a named part of your team’s workflow, staffed by someone other than the report’s author, distinct from and in addition to technical review.
Use AI-assisted structural checks to triage review capacity across a growing volume of reports, while reserving human judgment for whether framing choices are actually honest and well-chosen.
Set explicit policy for which categories of reports require full human review, which get AI-assisted triage only, and which can ship without narrative review — don’t let this default silently to “whatever there’s time for.”
Treat editorial review capacity as a real operational resource to be planned for, the same way you’d plan for code review capacity, rather than an informal favor asked of whoever’s available.

Quick Recap

An editorial review process for data narratives checks the story’s quality — lede, context, structure, confidence — as a distinct step from technical data accuracy review.
Without it, teams default to technical-only review, self-review by the original author, or no review at all before a deadline.
AI-generated drafts make this review more necessary, since fluent writing doesn’t signal whether the underlying claims are well supported.
AI can help triage which reports most need scarce human review time, but the judgment about honest framing remains a human one.

Where This Fits in the Series

This opens the series’ production arc, establishing the review process that catches narrative problems before publication. Article 11 goes deeper into one specific part of that review: fact-checking the actual claims a data story makes.