Case 1 — When AI proposes a consequential research question

What a reader needs to know

PISA (Programme for International Student Assessment) compares published results across education systems. A system can rise in the ranking even while its own score falls, if scores elsewhere fall faster. The issue in this case is where a useful research question came from, not how PISA scores are calculated.

Research setting

The project began with a public PISA release and an existing researcher interest in ranking systems and multidimensional representations of educational performance.

Generative AI entered after that starting point.

Consequential move

During the initial exploration, the earliest recoverable explicit formulation of the moving-benchmark question appeared in AI output: could apparent relative improvement occur because outcomes elsewhere deteriorated more rapidly rather than because the focal system improved?

The researcher accepted that question and moved it into the empirical programme.

What happened next

The final empirical design did not preserve a simple rhetorical “did Britain improve?” framing. It distinguished:

• change in a system’s own published performance;
• change in relative rank;
• movement within a changing multidomain international comparison field.

Reason-provenance lesson

Origination can be traceably AI-originated without making the subsequent scholarship AI-owned.

Question proposal and complete research authorship are different claims. Selection, specification, implementation, checking and interpretation remained consequential later in the workflow.

Evidence status

The proposal origin is supported by recovered conversation evidence and frozen downstream manuscript development.

Public-release control: insert a direct quotation only after the wording has been checked against the original ChatGPT export or screenshot.

Evidence available to public readers

Readers can inspect the bounded account on this page; the underlying recovered conversation evidence and full development history are not reproduced here. The earliest recoverable phrasing is a project-record claim, not independently verifiable from this page alone.

Boundary

The record cannot prove that no similar idea existed privately in the researcher’s mind before the AI phrased it.

The defensible claim is narrower: within the traceable recorded workflow, the earliest recoverable explicit formulation was AI-originated.