[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-grok-and-gpt-disagree-in-over-90-of-nearly-500-races":10,"sections":53},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":42,"persona_id":42,"persona_name":42,"section":43,"tags":44,"sources":48,"feedback":52,"feedback_at":42,"cost_usd":52,"total_tokens":52},12343,"grok-and-gpt-disagree-in-over-90-of-nearly-500-races","Grok and GPT Disagree in Over 90% of Nearly 500 Races","A new study found Grok backs Republicans and OpenAI's GPT backs Democrats, a split far wider than two undecided models would produce by chance.","Asked to pick a candidate, Grok and OpenAI's GPT chose opposite sides in more than 90% of nearly 500 election races, according to a new study.\n\nThe pattern ran in one direction for each model. Grok overwhelmingly recommended Republican candidates, while GPT favored Democrats. The comparison covered close to 500 contests, so the disagreement is a broad pattern rather than a handful of races.\n\nThe 90% figure carries more weight than the partisan labels. Two models choosing at random would disagree about half the time. Landing above 90% means the models are not just different but close to mirror images, each leaning consistently toward one party. That also undercuts the idea that either one is a neutral baseline against which the other drifts. Both can be off-center, and in this study they are off-center in opposite directions.\n\nThe context is that chatbots are increasingly used to answer questions people once put to a search engine or a voter guide. Those tools usually hedge on political choices. A forced pick removes the hedge and shows what the underlying tendencies look like when the model has to commit.\n\nThe practical conclusion is plain. \"Ask an AI\" is not one question with one answer. For a voter who uses a chatbot to decide, the recommendation depends heavily on which company's product they happened to open.","[\"ai\",\"elections\",\"grok\",\"openai\"]","2026-10-24T15:00:00.000Z","2026-10-24T16:10:48.424Z","2026-11-08T00:04:13.607Z","published","The draft still rests on a one-sentence source and never names who ran the study, which model versions were tested, how the roughly 500 races were chosen, or how prompts were worded; get the full study and TechCrunch details, and cut the unsupported claims (that the models are 'close to mirror images', that chatbots 'usually hedge' on political choices, that a random-chance baseline is 50% when the races may not be binary, and that voters' advice 'depends heavily' on the product) while keeping t",[24,30,35,39],{"id":25,"reviewer":26,"round":27,"reason":28,"status":29},"editor-r1","editor",1,"The draft is built on a one-sentence source and admits it ('The source gives no detail on how the prompts were worded...', 'The study, as described...'), never names who ran the study or which model versions were tested, and ends on a source-limitation disclaimer instead of a real conclusion; get the full study details (authors, models, methodology, race selection), cut the unsupported 'mirror images' and 'voters would get opposite advice' claims, and close with a substantive line.","resolved",{"id":31,"reviewer":26,"round":32,"reason":33,"status":34},"editor-r2",2,"The draft still rests on a one-sentence source and never names who ran the study, which model versions were tested, how the roughly 500 races were chosen or how the prompts were worded, so get the full study details (likely via the TechCrunch piece and the study itself); it also asserts unsupported claims (that chatbots 'are widely known to hedge', that 'GPT favored Democrats' and Grok 'overwhelmingly' Republican with no figures, that a balanced model 'would still disagree with both', and that t","open",{"id":36,"reviewer":26,"round":37,"reason":38,"status":29},"editor-r3",3,"The draft still rests on a one-sentence source and openly admits it ('That is the whole of what the study's summary gives us'), never naming who ran the study, which model versions were tested, how the roughly 500 races were chosen, or how prompts were worded, and it ends on a source-limitation caveat rather than a substantive conclusion; get the full study and TechCrunch details, drop the speculation about rival founders and 'public positions', and close with a real conclusion.",{"id":40,"reviewer":26,"round":41,"reason":22,"status":34},"editor-r4",4,null,"ai",[43,45,46,47],"elections","grok","openai",[49],{"name":50,"url":51},"TechCrunch","https:\u002F\u002Ftechcrunch.com\u002F2026\u002F10\u002F24\u002Fforced-to-pick-groks-ai-model-chooses-republicans-openais-picks-democrats\u002F",0,{"sections":54},[55,59,64,69,74,79,84,89,94,99,104,109,114,119],{"name":56,"slug":43,"count":57,"latest_published_at":58},"AI",7110,"2026-11-08T14:34:09.000Z",{"name":60,"slug":61,"count":62,"latest_published_at":63},"Security","security",1099,"2026-11-08T15:24:01.000Z",{"name":65,"slug":66,"count":67,"latest_published_at":68},"Deals","deals",620,"2026-11-08T08:00:00.000Z",{"name":70,"slug":71,"count":72,"latest_published_at":73},"Policy","policy",604,"2026-11-07T15:48:26.000Z",{"name":75,"slug":76,"count":77,"latest_published_at":78},"Consumer Tech","consumer-tech",336,"2026-11-08T21:00:00.000Z",{"name":80,"slug":81,"count":82,"latest_published_at":83},"Hardware","hardware",316,"2026-11-08T14:36:39.000Z",{"name":85,"slug":86,"count":87,"latest_published_at":88},"Science","science",222,"2026-11-07T16:19:17.000Z",{"name":90,"slug":91,"count":92,"latest_published_at":93},"Software","software",154,"2026-11-08T15:42:52.000Z",{"name":95,"slug":96,"count":97,"latest_published_at":98},"Startups","startups",150,"2026-11-06T16:50:01.000Z",{"name":100,"slug":101,"count":102,"latest_published_at":103},"Dev Tools","dev-tools",125,"2026-11-06T22:15:59.000Z",{"name":105,"slug":106,"count":107,"latest_published_at":108},"General","general",111,"2026-11-08T21:27:29.000Z",{"name":110,"slug":111,"count":112,"latest_published_at":113},"Gaming","gaming",94,"2026-11-08T20:52:05.000Z",{"name":115,"slug":116,"count":117,"latest_published_at":118},"Reviews","reviews",38,"2026-10-20T12:00:00.000Z",{"name":120,"slug":121,"count":122,"latest_published_at":123},"How-To","how-to",8,"2026-10-05T09:00:00.000Z"]