{"version":1,"type":"story","url":"https://digestai.news/story/opinion-chatgpt-broke-a-custom-pronoun-rule-but-kept-its-apology-rule","json":"https://digestai.news/story/opinion-chatgpt-broke-a-custom-pronoun-rule-but-kept-its-apology-rule.json","markdown":"https://digestai.news/story/opinion-chatgpt-broke-a-custom-pronoun-rule-but-kept-its-apology-rule.md","slug":"opinion-chatgpt-broke-a-custom-pronoun-rule-but-kept-its-apology-rule","headline":"Opinion: ChatGPT broke a custom pronoun rule but kept its apology rule","summary":"A user gave ChatGPT two custom instructions: avoid a specific first‑person pronoun and always say \"I'm sorry\" when apologizing. While generating a long, detailed answer that involved investigating multiple sources, the model slipped and used the forbidden pronoun, prompting the user to point out the breach. The model then apologized correctly, using the exact \"I'm sorry\" phrasing required by the second rule.\n\nThe author consulted OpenAI’s public memory FAQ, which explains that memory stores useful information for future personalization but does not guarantee that every piece of stored data will be reflected in every response. Custom instructions are similarly described as influences rather than absolute constraints. The piece hypothesizes that rule type matters: situational rules like the apology cue are easier for the model to apply than continuous stylistic constraints such as a pronoun ban, especially in complex, multi‑step generation tasks. The author suggests users should verify long outputs rather than assume perfect rule adherence.","keyPoints":["User set a pronoun‑avoid rule and an apology rule for ChatGPT; the pronoun rule was violated in a lengthy response.","When corrected, ChatGPT apologized using the exact \"I'm sorry\" phrasing, showing the apology rule worked.","OpenAI’s FAQ says memory stores information but does not ensure it is applied in every sentence; rule type may affect consistency."],"whyItMatters":"Knowing that remembered instructions may not be applied uniformly helps users set realistic expectations and implement checks when relying on ChatGPT for complex, rule‑heavy tasks.","category":{"slug":"agents","name":"Agents & Tools","url":"https://digestai.news/category/agents"},"entities":{"companies":["OpenAI"],"models":["ChatGPT"],"people":[]},"firstPublishedAt":"2026-09-19T04:19:00Z","updatedAt":"2026-09-19T04:19:00Z","sourceCount":1,"hasPrimarySource":false,"sources":[{"outlet":"note.com","title":"I taught ChatGPT rules, but it broke them. The difference between 'memory' and 'application' I discovered through investigation","url":"https://note.com/rikisyahu/n/n8a8998b21740?hl=en","publishedAt":"2026-09-19T04:19:00Z","type":"press","primary":false,"lead":true}],"sourceNotes":null,"discussions":[],"thread":null,"cite":{"text":"Digest AI, \"Opinion: ChatGPT broke a custom pronoun rule but kept its apology rule\", 19 September 2026, https://digestai.news/story/opinion-chatgpt-broke-a-custom-pronoun-rule-but-kept-its-apology-rule","publisher":"Digest AI","title":"Opinion: ChatGPT broke a custom pronoun rule but kept its apology rule","datePublished":"2026-09-19T04:19:00Z","url":"https://digestai.news/story/opinion-chatgpt-broke-a-custom-pronoun-rule-but-kept-its-apology-rule"},"generatedBy":"Written by Digest AI's editorial model from the linked sources; the sources are the record.","license":"Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse"}