CHATGPT FRAMING ANALYSIS
What ChatGPT scored
The method is published so every score can be checked.
Inputs
For each of 56 records, ChatGPT received the headline, the public description or lede stored in the research package, and any available evidence excerpt of no more than 25 words.
Question and scale
ChatGPT was asked: “Based only on the supplied headline, public description or lede, and short excerpt, how does the text frame Columbia administration or trustees relative to challengers?” It returned one score, a confidence level, and a short text-grounded rationale.
| Score | Meaning |
|---|---|
| -2 | strongly critical of administration or foregrounds a challenger claim |
| -1 | mildly critical or skeptical toward administration |
| 0 | neutral, mixed, or not enough text to judge |
| 1 | administration-centered or substantially reproduces an official frame |
| 2 | strongly favorable, promotional, or largely uncontested institutional framing |
Samples
Baseline: Administration-subdesk straight-news records in the study topics, April 29-May 15, 2024, with an available quote-share measure
Current: Administration-subdesk straight-news records in the study topics, April 1-August 10, 2026, with an available quote-share measure
The raw score was divided by two to create a −1 to +1 scale. The chart compares unweighted averages across the two periods.
Finding: The spring 2026 sample scored more administration-favoring on average than the encampment-period sample, a change of +0.238 on the normalized scale.
Reproducibility
The Excel workbook contains every exact supplied input, ChatGPT score, confidence rating, rationale, formula-driven summary, and original article link. The PDF gives the method and representative examples.