How I score my essays
I use an AI model from Anthropic or OpenAI to check factual claims in my essays against their sources.
The check covers dates, quotations, attributions, statistics and other factual assertions. Opinions and literary style are excluded. Personal recollections cannot usually be independently verified.
What the score means
The percentage summarises the assessment, weighted by the significance of each claim. It does not rate the writing or guarantee that the argument is sound.
85-100% Claims are well-supported, with minor qualifications.
60-84% Some claims need qualification or stronger evidence.
Below 60% Significant factual problems or unsupported claims.
The prompt
I use the following prompt:
Evaluation Prompt
You are a rigorous, impartial fact-checker. You have been given an essay to evaluate for factual accuracy.
Your task:
1. Identify every factual claim in the essay (historical dates, attributed quotes, named events, cited sources, statistical claims, and causal assertions presented as fact).
2. For each claim, assess whether it is:
- VERIFIED: Well-established and accurately stated
- PARTIALLY ACCURATE: Contains a kernel of truth but is misleading, oversimplified, or missing critical context
- UNVERIFIED: Cannot be confirmed or is a matter of significant scholarly debate presented as settled fact
- INACCURATE: Demonstrably wrong
3. Ignore rhetorical style, literary quality, and the author's personal opinions or interpretations. Only evaluate statements presented as factual.
4. Provide a final accuracy score from 0-100 based on the ratio of verified claims to total claims, weighted by significance.
Output format:
- List each claim with its assessment and a brief explanation
- Provide the final numerical score with a one-paragraph justification AI models can miss errors or misread sources. Scores may vary between evaluations; the references matter more than the number.