[{"version":"1.0","createSurface":{"surfaceId":"pat_article_en_negation_majuscules_questions_prompts_ce_que_la_preuve_soutient","catalogId":"https://a2ui.org/specification/v1_0/catalogs/basic/catalog.json","components":[{"id":"root","component":"Card","child":"body"},{"id":"kicker","component":"Text","text":{"@call":"formatString","args":{"value":"${/category} · ${formatDate(value:${/published}, format:'d MMMM yyyy')} · ${/author}"}},"variant":"caption"},{"id":"title","component":"Text","text":{"@call":"formatString","args":{"value":"# ${/title}"}}},{"id":"deck","component":"Text","text":{"@path":"/description"}},{"id":"fact_time","component":"Text","text":{"@call":"formatString","args":{"value":"${/readingMinutes} min read"}},"variant":"caption"},{"id":"fact_sources","component":"Text","text":{"@call":"pluralize","args":{"value":{"@path":"/sources"},"one":"1 source cited","other":{"@call":"formatString","args":{"value":"${/sources} sources cited"}}}},"variant":"caption"},{"id":"fact_figures","component":"Text","text":{"@call":"pluralize","args":{"value":{"@path":"/figures"},"one":"1 diagram","other":{"@call":"formatString","args":{"value":"${/figures} diagrams"}}}},"variant":"caption"},{"id":"facts","component":"Row","children":["fact_time","fact_sources","fact_figures"],"justify":"start","align":"center"},{"id":"faq_divider","component":"Divider"},{"id":"faq_title","component":"Text","text":"## Key answers"},{"id":"faq_list","component":"List","children":{"path":"/faq","componentId":"faq_item"}},{"id":"faq_item","component":"Column","children":["faq_question","faq_answer"]},{"id":"faq_question","component":"Text","text":{"@call":"formatString","args":{"value":"**${question}**"}}},{"id":"faq_answer","component":"Text","text":{"@path":"answer"}},{"id":"read","component":"Button","child":"read_label","variant":"primary","action":{"functionCall":{"@call":"openUrl","args":{"url":{"@path":"/url"}}}}},{"id":"read_label","component":"Text","text":"Read the article"},{"id":"markdown","component":"Button","child":"markdown_label","action":{"functionCall":{"@call":"openUrl","args":{"url":{"@path":"/markdownUrl"}}}}},{"id":"markdown_label","component":"Text","text":"Markdown version"},{"id":"all","component":"Button","child":"all_label","variant":"borderless","action":{"functionCall":{"@call":"openUrl","args":{"url":{"@path":"/blogUrl"}}}}},{"id":"all_label","component":"Text","text":"All articles"},{"id":"actions_divider","component":"Divider"},{"id":"actions","component":"Row","children":["read","markdown","all"],"justify":"start","align":"center"},{"id":"body","component":"Column","children":["kicker","title","deck","facts","faq_divider","faq_title","faq_list","actions_divider","actions"]}],"dataModel":{"title":"“Don't do that”, “VERY IMPORTANT”, “Can you…?”: what the evidence says about negation, capitals and questions in prompts","description":"Few studies test two phrasings of the same instruction inside an actual agent prompt. We reviewed the 2026 vendor guides, the papers and the field reports on negation, capital letters, repetition and the question form, and kept only what they state: where the evidence is strong, where it is thin, and which circulating figures we discarded.","category":"AI Agents","author":"Paul-Antoine Tual","language":"en-GB","published":"2026-11-06","updated":"2026-11-06","readingMinutes":12,"words":2610,"sources":23,"figures":3,"tags":["prompt engineering","negation","capital letters","prompt repetition","clarifying questions","AI agents","instruction following","interpretability","Claude","GPT-6 Astra"],"faq":[{"question":"Is it a mistake to tell an AI model what not to do?","answer":"Not always. Short avoidance rules such as 'no commas' held at or near 100% under task load in one study, but broad prohibitions are over-applied, and naming a forbidden word can prime it. Anthropic advises saying what to do instead."},{"question":"Do capital letters and words like CRITICAL make a model follow a rule better?","answer":"On recent Claude models, Anthropic says they cause over-triggering and advises plain wording. A 13-model study found capitals raised accuracy on factual questions by up to 8.95 points, but no study isolates capitals on behavioural rules."},{"question":"Does repeating an instruction help?","answer":"Repeating the whole prompt helped models that do not reason, in a study run on 2025 models. A short reminder of a format constraint at the end of the prompt restored 90 to 100% compliance in many settings in another. OpenAI advises stating each rule once."},{"question":"Does 'Can you…?' make an AI agent ask more questions?","answer":"No study shows it. 'Can you suggest…' can get advice where an imperative gets the action, but the interruptions we found documented come from the model, the tool, the mode and the question policy."}],"url":"https://paulantoinetual.fr/en/blog/negation-majuscules-questions-prompts-ce-que-la-preuve-soutient","markdownUrl":"https://paulantoinetual.fr/en/blog/negation-majuscules-questions-prompts-ce-que-la-preuve-soutient.md","blogUrl":"https://paulantoinetual.fr/en/blog"}}}]