Many daily briefs that go wrong appear to have been built on perfectly good data. Anyone who checks the feed after a bad morning will usually find it did exactly what it was asked, down to the source and time stamp on every story. The trouble starts later, at each point where a person (or a script someone wrote last spring) passes the brief along and assumes the next pair of eyes will catch what this one missed. Those points decide whether a morning brief deserves anyone’s trust, and the feed rarely gets a vote. Seven handoffs seem to do most of the harm, and they’re ranked here from the most damaging to the least.<\/p>\n
First, and worst, is the handoff teams are least keen to talk about. A machine-written summary lands in an editor’s queue and gets its commas fixed, while the claims inside it go unchecked. Summaries look finished. At twenty to seven, clean grammar and a confident verb or two are often all it takes to get one approved. Journalists at public broadcasters in 18 countries went through thousands of AI answers about the news and found that nearly half carried at least one significant issue<\/a>. Sourcing alone was a serious problem in about a third of them. By lunch, an error that got past this step can be heard in meetings as something the company knows, because it arrived in the brief.<\/p>\n That risk comes back every morning in a brief that squeezes forty articles into forty sentences, and few editors will reopen forty articles before breakfast. A review policy that assumes otherwise probably turns into theater within a month. Scholars of newsroom AI have made a similar point, warning that a human in the loop is a weaker safeguard than it sounds<\/a> when that human is rushed and inclined to trust the machine. If you’ve approved a summary without opening the story behind it, you already know how that goes.<\/p>\n A narrower rule holds up better. Names and numbers get checked against the original, while the sentences that merely join them up are left alone. Quotes are the exception worth getting fussy about. A paraphrase wearing quotation marks travels well, and sooner or later someone reads it out in a meeting as though the person quoted had actually said it. Put the source link beside every sentence and most of this shrinks to a click. The narrow rule is weaker than full review. An editor will still be keeping it on a wet Tuesday in February, though, which is more than full review can promise.<\/p>\n The second handoff happens when the brief leaves the tool altogether. In regulated companies, and in plenty that merely feel regulated, the draft gets pasted into a Word file and emailed to legal or communications for sign-off. Links flatten into blue text that points nowhere. The source credits, which were the reader’s easiest way to check a claim, usually disappear with them. Legal’s markup tends to land as an attachment a few minutes before send, and whoever is on shift copies the edits into the dashboard by hand. The brief that goes out at seven is then a third version, one that no single person has read start to finish.<\/p>\n Apryse’s DOCX editor integration guide<\/a> is a sensible first read for teams that want reviewers to stay in Word without the file wandering off by email. It reads like a buyer’s checklist for a Word editor built into someone else’s product. Anyone from legal will probably skip straight to tracked changes, since the whole point is that each edit gets accepted or rejected inside the file itself. Comments and version history are in there too. Once that Word editor lives inside the brief tool, the file legal marked up is the file that goes out. That way, a lawyer’s deleted sentence stays deleted.<\/p>\n Reviewers aren’t being difficult when they insist on Word. It’s where they read contracts and press releases all day, and a request to learn a new dashboard for one email each morning is one they’ll likely ignore. Still, some copies will leave the building, and a Word file is a leakier container than it looks. Old comments, and tracked revisions someone assumed were long gone, can sit inside it unseen, and Microsoft’s guidance on inspecting documents before sharing them<\/a> walks through how to find them. It also advises running that inspection on a copy, because some removals can’t be undone. Two house rules probably cover most of the risk. Each brief gets a single file of record, and any copy that goes outside the team gets an inspection pass first.<\/p>\n Plenty of teams leave duplicates to the engineers, as if counting copies had nothing to do with judgment. Suppose a wire story about a supplier recall comes in overnight. By six the feed may hold fifteen copies of it, each under a different masthead. A bare count of fifteen, with nothing to say where the copies came from, starts to sound like fifteen newsrooms that each got the story on their own. Fifteen logos sit on top of one reporter’s work. Wire services have sold the same copy to many papers for well over a century, and it rarely mattered, because a reader of one paper never saw the other fourteen. A brief puts all fifteen in one place.<\/p>\n AI search tools get attribution wrong in a related way. A 2025 study of eight of them caught answers pointing readers to syndicated copies on aggregator sites instead of the original publisher<\/a>, in one case even when that publisher had blocked the tool from its site. A brief that credits the copy credits the wrong newsroom.<\/p>\n Readers do click those links, which makes the mistake costlier. People who get their news through AI chatbots turn out to be fairly keen clickers. Among them, 42% told researchers for the 2026 Digital News Report<\/a> that they always or often go through to the original source, and there’s little reason to think brief readers are any lazier. The grouping step should hand the editor a count of original reports, with the first publisher named and linked, instead of pointing at whichever copy happened to land first. Anything less turns repetition into apparent weight.<\/p>\n Number four moves so slowly that hardly anyone sees it happen. Back when the brief launched, somebody chose its keywords and the news categories the feed filters on<\/a>, probably in a single afternoon, and then went off to work on something else. The editor who inherits the brief sees what the filters let through and has no way to see what they hold back. A company that launched a product line in the spring can still be getting briefed on its old one in the autumn, simply because nobody added the new name to the keywords. No one chose that outcome. The query chose it months earlier, and the authority to change it never transferred with the job.<\/p>\n Older newsrooms had a crude answer to invisible omissions. Copy that didn’t make the paper got jammed onto a metal spike on the desk, and the rejects piled up there for anybody curious enough to leaf through them. Nobody would call it elegant, but every rejected story ended up there. Editors still talk about a story being spiked when it’s dropped for editorial reasons, long after the actual spikes left the newsroom. Most automated briefs keep nothing like that pile, since filtered stories never appear anywhere, so the question of what was left out can’t even be asked.<\/p>\n None of the fixes here are expensive. A rough log of excluded stories lets an editor scan once a week for what the filters dropped. The query itself can be written out in plain words at the foot of the brief, next to the name of whoever owns it, so that ownership survives the next reorganization. A quarterly meeting where someone reads the filters aloud would probably catch most of the drift, and few teams seem to hold one.<\/p>\n The fifth handoff only happens after the brief has gone out, when a source corrects its story or withdraws it. The publisher’s own page will show the change. The brief that reached two hundred inboxes at seven won’t, unless someone owns the job of chasing it. In many teams, that job doesn’t appear to belong to anyone. Forwarded copies keep carrying the error in the meantime. A large study of Twitter found that false stories were about 70% more likely to be retweeted than true ones<\/a>. An uncorrected brief can give its mistake a comparable head start inside a company.<\/p>\n The voluntary ethics code American journalists have leaned on for about a century was revised on October 4, its first rewrite since 2014, and it now holds them responsible for their work whatever tools produced it<\/a>. Daily briefs aren’t journalism in the strict sense. Their readers make decisions as though they were, though, and a correction that never reaches those readers leaves the original error sitting in their inboxes.<\/p>\n A standing line at the top of the next morning’s brief, listing whatever changed since the last one went out, is the least a team owes its readers. A team can go further and recheck every linked source a few hours after sending, while the brief is still being read and forwarded. Keeping an archive of each sent edition matters too, because a correction is hard to write when nobody can say exactly what readers were told.<\/p>\n Sentiment labels do a smaller kind of damage, but it compounds. The software tags a story as negative, the tag travels into the brief as a red dot, and by the time an executive skims it the dot can harden into the editorial verdict on that story. Automated sentiment scoring has a hard time with sarcasm and irony<\/a>, and business coverage runs on both. You’ve probably seen a dry analyst quote flagged as hostile, or a cheerful headline about layoffs marked positive because it mentioned rising margins.<\/p>\n Vocabulary is part of the trouble. A widely cited study of company annual reports found that almost three-fourths of the words a general-purpose dictionary classed as negative<\/a> don’t usually carry a negative meaning in financial writing. The paper’s title makes the point with one of them, asking when a liability isn’t a liability. Software tuned on everyday language is likely to misread an earnings story in exactly that way. Teams that keep the labels can pull twenty at random each week and compare them with a human reading, which shows fairly quickly whether the scores have earned a place in front of readers. Until they have, the labels belong in the editor’s view and nowhere else.<\/p>\n The seventh handoff does the least damage, though it does it every single day. A brief assembled overnight in one time zone gets finished by someone in another, and the phrase “the last 24 hours” means a different window to each of them. Stories published late in the evening either fall into the gap between shifts or show up twice.<\/p>\n Calendars make it worse twice a year. Britain’s clocks go back on October 25<\/a> this year, while American clocks fall back on November 1<\/a>, so for one week the time difference between a London desk and a New York desk shrinks by an hour.<\/p>\n A handover time agreed in both cities’ local hours is wrong for that week unless somebody remembers to adjust it. A written cutoff in a single reference time, agreed by both shifts and printed at the top of the brief, would likely fix most of this, and setting one up takes an afternoon.<\/p>\n2. When the Brief Leaves for Word<\/b><\/h2>\n
3. Duplicates Dressed Up as Consensus<\/b><\/h2>\n
4. Filters Nobody Remembers Setting<\/b><\/h2>\n
5. The Correction That Never Catches Up<\/b><\/h2>\n
6. Sentiment Labels Treated as Verdicts<\/b><\/h2>\n
7. The Overnight Shift Change<\/b><\/h2>\n
Collecting Stories Versus Judging Them<\/b><\/h2>\n