I tested 5 AI models summarizing the same news articles. They all inherited the source's framing, even when trying to be neutral. i'm rookie, be kind

So I did a small study testing whether ChatGPT, Claude, Gemini, Grok, and DeepSeek summarize news the same way. Spoiler: they don't, and the reason is kind of concerning. The setup was simple. Six immigration news articles (left, center, right sources), same neutral prompt to each model, all thirty summaries manually coded for neutrality, accuracy, completeness, emotional language, and framing. What I found was that all five models consistently inherited the framing of the source article. When I fed them a left-leaning article, the summaries got coded as more negative. Right-leaning article? More positive framing in the summary. Center source? Clean results across the board. The creepy part is that the summaries actually sound neutral. If you just read them, they seem balanced. But they're shaping reader understanding through emphasis, omission, and tone inherited from the source. Claude performed best overall, Grok was strong on completeness, ChatGPT cut corners sometimes. **Important caveats:** This is six articles. One coder. One topic area. You literally cannot generalize this to "all AI is biased." This is exploratory work that raises a question, not proof of anything. But I think the question is worth asking: when people consume news through AI summaries, are they getting objectivity or are they getting the source's framing laundered through a model that sounds neutral? All the data is open. The Excel workbook has every summary, my coding rubric, my notes. Poke holes in it. Test on different articles. Let me know if the pattern holds or if I'm just seeing what I want to see. Full repo with all data: [GITHUB REPO](https://github.com/FreakyDevelopers/Political-Framing-in-AI-News-Summaries/tree/main) Happy to answer questions about methodology or take criticism on the coding approach.