The Authority Index
Experiments

Retrieval Prompt Variants: Experiment Log

By Sofia Laurent2/12/2026Experiments
Retrieval Prompt Variants: Experiment Log featured image

Protocol

Prompt variants were grouped by specificity, temporal constraints, and citation explicitness.

The same topic clusters were reused to preserve comparability across variants.

Measurements

Primary metrics were citation precision, duplicate-source rate, and domain entropy.

All measurements were recorded per engine and per prompt class.

Observations

High-specificity prompts improved citation precision but reduced source diversity.

Temporal constraints materially changed source freshness patterns in recommendation outputs.

Prompt variant baseline
EngineCitation RatePresence Rate
ChatGPT42%67%
Gemini35%61%

Precision by Prompt Variant

General33%
Specific49%
Temporal41%

Related Research

Author

Sofia Laurent profile

Sofia Laurent

Head of Experimental Research

Sofia leads controlled experiments on prompt sensitivity, source diversity, and ranking signal interactions across major AI answer engines.

View all research by Sofia Laurent.