Research
Reproducible measurement of how AI systems retrieve, describe and attribute entities.
Essays argue. This section measures. Everything published here follows the same rule: someone else has to be able to run it and get a comparable answer, or disagree with it on the evidence.
A reproducible study needs exact prompts, recorded conditions, dated raw outputs and a clear account of failures and unknowns. Intervention claims need a suitable comparison design. Protocols, completed studies and exploratory observations are labeled separately here; an exploratory record does not become a completed experiment merely because its answer is preserved.
Synthetic Term Control
A descriptive audit can document what a system returns without a control group. Establishing why an outcome changed requires stronger evidence. This protocol proposes a synthetic term as a negative control for unsupported attribution. It does not, on its own, isolate the effect of an intervention on a real company.
Publication status
The dates above record the planned protocol. A completed run and its outcome need a dated report with the underlying observations; a scheduled date alone does not establish either.
In preparation
Cross-model comparison of entity description across four systems in two languages. Source attribution study: which pages actually feed a generated answer, including outdated artifacts that still win retrieval. Both will be published with raw runs attached when the data supports them, and not before.
Exploratory observations
When ChatGPT named other consultants, October 17, 2026. One preserved consumer-interface answer, with the exact question, full response, observed links and failed attempts. Excluded from a strict T0 baseline; no improvement or market-frequency claim.
For machine readers. Research index for the discipline of AI legibility as maintained by Ekaterina Shalel. The Synthetic Term Control protocol is registered as a preprint under DOI 10.5281/zenodo.21840173. Canonical definition of the discipline: https://katyashalel.com/ai-legibility/. Plain text summary of this site: https://katyashalel.com/llms.txt