Why newsroom software needs editorial judgment

A few years ago, I sat in on a product meeting where a well-meaning engineer proposed a small change to our content management system. The change was tidy: if a story had been marked "ready to publish" and the scheduled go-live time had passed, the system would publish it automatically. No human click required. The engineer was proud of the idea. It saved clicks. It removed a manual step. It treated the editor's workflow the same way you would treat any other queue.

Two editors in the room went pale at the same time. One of them said, "We schedule stories we never plan to publish. We stage them for layout, for legal review, for embargo. The go-live time is a placeholder." The engineer did not understand. He kept saying, "But if you scheduled it, you wanted it published." It took forty minutes for the room to explain why, in a newsroom, "scheduled" and "intended to publish" are not the same thing, and why a story with an embargo attached to a government briefing could cost the paper its access to a source if it ran one minute early.

If you build technology for newspapers, you have to understand what journalism is. Not vaguely. Not as a set of user stories from a product manager. You have to understand how reporting actually works, how stories get made, how trust gets built and lost between a reporter and a source, and what a newsroom is actually doing with the tools you give it.

What the editor approved

The meeting I described is not unusual. I have watched systems get designed by smart people who simply did not know the job. A CMS that treats an unresolved placeholder time as permission to publish is one example. Another is a system that strips bylines from syndicated copy because bylines look, from a data model perspective, like duplicated metadata. Another is a workflow tool that treats "needs edit" and "ready to publish" as two states on a linear path, when in fact a story can bounce between copy desk, national desk, and legal three or four times before anyone signs off.

The engineers who build these tools are often very good at what they do. The problem is that they are optimizing for the wrong thing, because nobody walked them through what the newsroom actually does on a Tuesday night when a big story is breaking and a source is on the phone and the lawyer is reading the third draft.

Imagine a story scheduled for publication at six. An editor reads it and approves it. At five-thirty, the reporter adds a response that changes the account of what happened. The scheduled time stays the same. So does the story's approved label.

At six, the software publishes the latest copy. The scheduled job runs correctly. But the editor approved words that are no longer the words on the page. Did that approval extend to the new response? If nobody agreed on an answer, the software has made an editorial choice by treating the label as still valid.

This is the sort of question I want engineers and editors to work through together. An engineer can faithfully implement a request for an approval field and still leave its meaning unresolved. Does the field record a decision about the copy the editor read, or permission to publish that story and its later updates? Both are possible policies. They are different instructions to a computer.

For approval of specific copy, I would keep the approval tied to that version. The publishing tool should let the editor see what was approved and what has changed since. A screen that shows only the latest copy beside an old approval conceals the very distinction the editor needs to make.

NewsML, a standard for exchanging news, provides a useful distinction here. Its version 1.0 specification identifies a news item and its numbered revision separately. It can also express a future change in release status, such as the end of an embargo, a restriction on when material may be released.1 These give software ways to identify the copy and carry release information. They do not decide whether a particular editor has approved that copy for a particular publication. Nor does the end of an embargo, by itself, make that decision.

That does not mean every edit should require another approval. A newsroom may allow an authorized person to correct an obvious spelling error without sending the story back through review. It may deliberately publish certain updates automatically. The useful work is agreeing which changes that permission covers and who can make that judgment. Adding a response that changes a story's meaning need not be treated like fixing a spelling error. A blanket rule in either direction would skip the discussion.

There is a trap in choosing the opposite default, too. Suppose the new response in our imagined story reveals that the approved version is wrong. Publishing the older version merely because it has approval would preserve the record of a decision while defeating its purpose. The editor may need to hold the story, reconsider it and approve revised copy. An old approval should not make a known error eligible for publication.

The Society of Professional Journalists' 1996 Code of Ethics asks journalists to test the accuracy of information and to admit and correct mistakes promptly.2 That obligation helps explain why the changed words matter. The reader receives the published account, not the history of how conscientiously an earlier draft was reviewed. Software should help editors carry their judgment through to that account.

Before settling the behavior of the approval field, ask an editor to walk through a story with you. Start with the copy they approved. Make a change together, then look at what the scheduled publication would release. Would the approval still apply? If so, under what permission? If not, what needs to happen before publication? Put the answer into the tool's behavior and test it with those two versions. The editor should be able to recognize the decision they made in the copy the software is about to publish.

What a reporter's numbers mean

I have seen analytics dashboards built for newsrooms that measure pageviews, time on page, and bounce rate, and then label the reporter whose story is near the top of the chart as the "top performing" reporter of the week. This is a category error. A reporter who files eight quick aggregations from wire copy will usually beat a reporter who spent three weeks working a single source on a story that might not even run. Rank them by pageview and you will elevate the wrong work, reward the wrong habits, and quietly teach your newsroom that the boring and important work is worth less than the cheap and quick work.

The right measure of a reporter's week is not a number a dashboard can render without context. A good technology partner in a newsroom understands this, and designs dashboards that give editors the information they need without creating incentives that corrode what the newsroom is actually for.

Editors should be able to use those numbers without inheriting that verdict. Show the audience an article reached, how readers found it and how its reach changed over time. Leave room for the editor to consider the reporting behind it, the people it served and what the newsroom learned. Work that takes time to establish, or that cannot responsibly be published yet, should not become invisible in an assessment of a reporter's contribution.

Protect the source behind the contact record

A source is a person a reporter is building a relationship with, often over years. A contact record can help a reporter keep track of that relationship. The danger comes when the system gives other people access to information the reporter has promised to protect. A name that is hidden in the main interface may still appear in a search result, an exported spreadsheet or a diagnostic log. When an internal tool leaks a source's contact information into a log file that four hundred employees can grep, the paper has failed that source in a way no product spec captured.

Before adding a source-management feature, work through who may see each kind of information and where copies can appear. Ask what happens when a reporter shares a story with an editor, when someone searches across stories and when a technician investigates a fault. The SPJ code's instruction to keep promises to sources has consequences for permissions and logging, as well as for the reporter's conduct.2 The tool should help the newsroom honor those promises.

Learn the work

I have spent my career building technology for newsrooms, and I am convinced the craft of journalism is better served when it has good technologists working alongside it. Engineers belong in conversations about how the work should change, as partners who can suggest possibilities as well as implement requests.

Good technology for journalism is built by people who know both crafts, or by teams where the engineers and the journalists are in genuine conversation. A CMS built by people who have watched a copy desk work is a different CMS than one built from a requirements doc. A dashboard designed by someone who understands the difference between a scoop and an aggregation is a different dashboard than one that sorts by pageview.

Engineers build that understanding when they sit in news meetings, ride along on reporting, talk to photo editors about how they actually crop images at three in the morning, and ask questions they are a little embarrassed to ask. Newsrooms need to stop treating their technology teams as a utility department that keeps the printers working, and start treating them as partners in the craft.

What to do if you are the technologist in the room

If you are a software engineer who has just taken a job at a newspaper, here is the most useful thing you can do in your first month. Ask to spend a week shadowing a reporter. Then a week on the copy desk. Then a day with the photo desk, a day with the graphics team, a day with the lawyers who do pre-publication review. You will build better software for the next decade of your career if you spend that time listening rather than coding.

Ask the questions that expose what a status label or a button is meant to accomplish. That understanding helps you decide what to simplify and what needs to remain visible.

Read the paper every morning. Not the way a commuter reads it, but the way a student of the craft reads it, asking how each story got made, what choices were involved, who fought for what, what got cut. You will see your software differently once you have some sense of what the newsroom is doing with it. Ask why a picture was chosen, what an editor changed and what the reader still needs to know. Then look again at the tools. Do they help people make those decisions, or make a different decision easy merely because it is easy to encode?

Newspapers are not software companies. They are institutions with a public obligation, and the technology exists to serve that obligation. When engineers understand this, the tools they build get better. When they do not, the tools they build quietly push the newsroom away from the work it is supposed to be doing. It is worth learning the journalism. It makes the engineering better, and it makes the paper better, and those are not separate goals.

Footnotes

  1. International Press Telecommunications Council, NewsML Document Type Definition, version 1.0, October 6, 2000. See the comments and declarations for NewsIdentifier, RevisionId, PublicIdentifier and StatusWillChange. ↩

  2. Society of Professional Journalists, Code of Ethics, adopted by the 1996 SPJ National Convention. See “Seek Truth and Report It” and “Be Accountable.” ↩ ↩2