In brief
When news summaries, search answers and entertainment are generated individually for each person, a society loses the shared reference points it relies on to argue productively, correct mistakes and notice manipulation.
This brief sets out the scale of the change and proposes four measures that would preserve a common record without restricting legitimate personalization.
Recommendations
- Generated summaries of news should always link to a canonical source that every reader can see.
- Independent researchers should have privacy-preserving access to samples of what platforms actually showed.
- Every personalized feed should offer a one-step switch to a common, non-personalized view.
- Independent archives should keep representative samples of generated media, so that the record of what people saw survives.
The end of the same page
Disagreement is healthy only when people can point to the same thing. For most of modern history, that was easy: a newspaper printed one front page, a broadcast aired one bulletin, a search engine returned one list. People disagreed about what those things meant, but they could at least agree on what had been said.
Generative media changes that. Increasingly, the summary of a news story, the answer to a question and even the plot of a story are generated for one reader at a time. Two neighbors asking the same question may receive answers that differ in emphasis, in detail or in substance, and neither has any way of knowing.
Frothly shared-exposure panel, 2,000 adults per wave. “Same version” means identical wording in the first three sentences encountered.
Why it matters
A society without shared reference points loses three abilities at once. Disputes about facts become impossible to resolve, because there is no common text to check. Corrections lose their reach, because a correction to one version of a story does not reach readers who saw another. And manipulation becomes invisible, because nobody can compare what they were shown with what anyone else was shown.
What we propose
We do not recommend banning personalization, which has real benefits, or requiring platforms to show everyone identical content, which would be neither practical nor desirable. Instead we propose four measures designed to keep a common record alongside personalized ones.
- A canonical source. Any generated summary of a news event should link to a fixed, human-published account of it, so that every reader can see the same text behind their personal version.
- Audit access. Accredited researchers should be able to sample what a platform showed to whom, under strict privacy protections, so that divergence and manipulation can be measured.
- A common view. Every personalized feed should offer a single, clearly labelled switch to a non-personalized view of the same material.
- An archive. Independent archives, including Frothly’s Lazarus program, should hold representative samples of generated media so that historians can reconstruct what people actually saw.
Who should act
Platforms
- Implement canonical links and a common view now, ahead of regulation.
- Publish regular reports on how far personalized versions diverge from canonical ones.
Regulators
- Establish a legal basis for privacy-preserving audit access to generated media.
- Fund independent archiving of generated public-interest content.
Newsrooms
- Publish a canonical version of every story, and insist that platforms link to it.
References
- Case, H. D. (2018). Jacking out: attention and withdrawal online. New Media and Society, 20(9), 3301–3318.
- Ono-Sendai Corporation (2024). Personalization at Scale. White paper. Chiba.
- Riviera, P. (2019). The performance of illusion. Visual Culture Review, 33, 77–95.
- Sense/Net (2025). Transparency Report. Los Angeles.
About the authors
Judy Alvarez leads Frothly’s immersive media work.
Dr. Peter Riviera is a research fellow in Synthetic Media. His work concerns illusion, performance and persuasion.
How to cite
Judy Alvarez and Dr. Peter Riviera (2026). A Consensual Hallucination Shared reality in generative media. Frothly Policy brief FR-PB-2026-04. San Francisco: Frothly. https://frothly.org/research/consensual-hallucination/