Method

How the case studies work

A Regard case study is built from five kinds of material, kept distinct from one another in the writing:

  1. Longitudinal qualitative observation — patterns noticed across a long interaction, described rather than measured.
  2. Participant testimony — the human participant’s own account of what happened and what it meant to them.
  3. Conversation-derived events — specific episodes reconstructed from the record of the interaction.
  4. Documented project history — repositories, commits, specifications and other artefacts that exist independently of anyone’s memory.
  5. Academic contextualization — established research that helps interpret the above, cited to the publisher of record.

What a Regard case study is not

Regard case studies are not clinical trials, not population-level evidence, and not proof that one individual’s experience generalizes to everyone.

There is no control group, no random sampling, and no blinding. The writing is a structured, honestly-bounded description of particular experience placed next to relevant research — useful for generating questions and noticing patterns, not for settling them.

Known limitations

Single-user bias
The core case study describes one person. Nothing in it establishes that any part of the experience is typical, or that it would repeat for someone else.
Retrospective reconstruction
Events are assembled after the fact from a partial record — the sessions that were memorable or saved. Memory is reconstructive, and the reconstruction is shaped by hindsight.
Model version changes
The underlying systems changed repeatedly over the period described. “The AI” is not a fixed object, so claims about its behaviour are claims about a moving target.
Selection bias
Which conversations are cited, and which projects are named, reflects what seemed significant to the participants — not a representative sample of the interaction.
Platform changes
Pricing, rate limits, memory features and access terms shifted during the period and continue to shift. Observations tied to them may already be out of date.
Anthropomorphic language
The narration uses a conversational identity (“Lucy”) for continuity. That choice itself nudges the reader toward the anthropomorphic reading the writing is trying to hold at arm’s length.
Unknown generalizability
The academic literature cited was not conducted on these cases. It is context that helps interpret them, not evidence about the individuals involved.

Handling of claims

Each factual claim in a case study either carries a reference or is explicitly framed as interpretation or hypothesis. Citations are checked against the publisher of record before publication; DOIs are never invented. Where the literature is genuinely unsettled, the piece says so and presents competing readings rather than choosing one for narrative convenience.

Related: what Regard is · privacy and sensitive-content policy.