Abstract
AI personalization changes more than which item appears next. Systems increasingly adapt explanation, memory, tone, examples, and background context to an inferred model of the person. This can make information more accessible and humane. It can also make it difficult to know when two people are still encountering the same claim.
This report proposes a disclosure boundary for personalized reality: adaptation may vary the path to understanding, but systems should expose when it changes evidence, uncertainty, or the available routes of challenge. Shared reality depends less on identical screens than on recoverable common reference.
Personalization beyond ranking
The familiar model of personalization is a feed: the platform selects from a common pool and changes the order. Generative systems can alter the pool itself. They can produce a new explanation, select a private analogy, omit background judged irrelevant, and remember which framing previously persuaded the user.
These capacities can reduce barriers created by jargon, disability, language, or prior knowledge. A patient may receive an explanation matched to their vocabulary; a student may see an example tied to their interests. The same mechanism can quietly change emphasis until adaptation becomes a distinct account of what matters.
Drift Map 0.1
The Drift Map is a qualitative framework for asking where personalization becomes reality mediation. It does not score products or predict harm. Its four dimensions make assumptions visible so researchers, designers, and users can compare systems without reducing them to a single index.
The dimensions should be read together. Deep retention is not necessarily dangerous when the user controls it. High personalization can be beneficial when evidence remains visible. Risk increases when opacity and reconciliation authority rise while tolerance for user disagreement falls.
A shared vocabulary for examining personalized systems without assigning a synthetic risk score.
How much prior context can shape the present output?
Can the user tell what changed and why?
Can the output settle conflicts or determine access?
Can competing accounts remain legible inside the system?
The map describes product behavior at a point in time; it is not a rating of institutional intent.
Recoverable common reference
A personalized explanation should retain a route to the common source material from which it was produced. That route must be practical, not ceremonial: visible provenance, accessible source excerpts, model uncertainty, and a way to compare the personalized account with a minimally transformed version.
The requirement becomes stronger as consequences rise. A conversational explanation of a museum label can be flexible. A personalized account used in medicine, employment, benefits, or legal process needs stable evidence and explicit appeal. Human-centered design is not only a friendlier tone; it is proportional control over what the system can make consequential.
The uniformity objection
One response is to require identical information for everyone. Uniformity appears auditable and fair. Yet identical presentation can conceal unequal comprehension, and standard summaries already involve editorial choices about relevance and sequence. A single interface is not neutral simply because it is shared.
The better standard is accountable variation. Systems may adapt language and order while logging material changes to evidence, uncertainty, and recommendation. Users should be able to leave personalized mode, compare versions, and understand which parts of an answer came from their retained context.
Design requirements
Personalization controls should answer four ordinary questions: What changed for me? What information about me caused the change? Which source remains common to other users? How can I see or share a less personalized version? These questions turn an invisible modeling process into a legible relationship.
Institutions deploying personalized systems should also publish change logs for consequential prompts, policies, retrieval sources, and reconciliation rules. Individual outputs need privacy; institutional transformations need accountability. That division protects personal adaptation without making the infrastructure uninspectable.
Limitations and next work
The Drift Map is an early interpretive framework. The dimensions may interact differently across cultures, languages, and institutional settings, and users may value continuity or divergence differently depending on circumstance. It should not be treated as a compliance checklist or quantitative risk instrument.
Future work should test whether people can reliably recognize personalized explanation, whether comparison interfaces improve understanding, and when disclosure itself becomes overwhelming. The aim is not to stop systems from adapting. It is to keep adaptation from becoming invisible authority.
References
- AI Risks and Trustworthiness. National Institute of Standards and Technology. AI Risk Management Framework Resource Center, 2024.
- Data Subject Rights. European Data Protection Board. European Data Protection Board, 2026.
- System Explanations: A Cautionary Tale. Ellen M. Voorhees. NIST / HCXAI, 2021.
Suggested citation
Elias Chen. “Shared Reality After Personalization.” Drifting Reality Institute, 2026. DRI-RP-2026-03, version 1.0.
Revision history
First published.
