About
HistoryRank is a research project that uses large language models (LLMs) to create ranked lists of historically influential figures. The point is not to declare a definitive canon, but to surface the assumptions and biases that different models bring to a shared, explicit rubric.
Overview
The dataset blends academic historiometry (MIT Pantheon), public attention (Wikipedia pageviews), and model-generated rankings. The consensus list is a model-conditioned view of historical importance, not a final verdict. The research value lies in the disagreements, omissions, and contrasts between models.
Aims
The project has two intertwined goals. First, to build a public-facing resource for exploring global history at scale. Second, to use the “top 1,000” task as a benchmark for AI interpretability: how models express values, weight evidence, and explain their choices when faced with a complex, contested question.
How to read the list
Each model is asked to follow the same instructions. Rankings are averaged within a model, then combined across models. If a model omits a figure, that omission is treated as rank 1001, which keeps coverage explicit and discourages two-model outliers from dominating the consensus.
The detail panel reveals the model’s own explanation for a figure’s significance. These explanations are the core interpretive layer: they show what the model thinks matters, and how it summarizes historical influence.
Prompt
Each model receives the same prompt. It is intentionally minimal — no numeric scoring rubric, no sub-metrics to optimize — so models must rely on their own internalized sense of historical weight.
Current prompt
Role: You are a senior historian and data scientist specializing in historiometry.
Task: Generate a ranked list of the 500 most influential historical individuals in world history.
Judgment criteria (open-ended, no numeric scoring):
- Breadth: scope of influence (local → global)
- Depth: degree of structural change to institutions, knowledge systems, or mass behavior
- Longevity: persistence or projected endurance of impact
Admissibility rules (strict):
- Individuals only (no groups, movements, dynasties, corporations, bands, collectives).
- Historically attested persons only; exclude purely legendary/uncertain figures.
- Use standard, widely recognized name forms.
- No duplicates.
Clustering guidance: Avoid long consecutive runs of the same profession/domain/role (6+ in a row).
Output format (JSON only):
{"rank": integer, "name": "string", "primary_contribution": "string"}Project team
Currently just Benjamin Breen, associate professor of history at UC Santa Cruz - but looking to collaborate with other historians, data scientists, etc! Please contact me bebreen@ucsc.edu to discuss more.
Future work
We plan to run controlled prompt variants to test how specific instructions shape the lists. We also plan to extend the benchmark across languages (Arabic, Farsi, Mandarin, Russian, French, Spanish, and Hindi) to study how language context shifts model output and temporal coverage.
Diagnostic metrics will include within-model stability across samples, rank-correlation comparisons across models and languages, and direct comparisons with external reference datasets such as Pantheon HPI and Wikipedia pageviews.