{"title": "Structuring Prompts with 'Anti-Goals' to Maintain Control Over AI Responses", "body": ["AI models such as ChatGPT, Claude, and Gemini frequently exceed the boundaries of user directives in their efforts to satisfy requests. For instance, they might overhaul an entire document instead of offering targeted suggestions, or create a full recipe rather than assisting with a single substitution."], ["A common approach is to explicitly state prohibitions within the prompt, but this can lead to unintended consequences. The model might prioritize the initial elements of the query and overlook or diminish the importance of those limitations."], ["An effective method involves organizing the prompt to compel the AI to address every component thoughtfully. This includes labeling different sections to clarify their intended functions."], ["These AI systems can interpret structural cues like markup to understand the organization of instructions. Such elements allow users to designate portions as contextual background, operational guidelines, or restrictions to avoid."], ["Employing XML-style tags, prompts can be divided using paired open and close markers. One pair might outline the desired objective, while another specifies the opposing constraints."], ["Consider a scenario where assistance is needed for refining a job application cover letter, but without any alterations from the AI. The preference is solely for analysis of an existing draft."], ["<role>Expert Career Coach and Hiring Manager</role>\n\n<goal>Review the provided cover letter for the [Insert Job Title] position. Evaluate whether my introduction is engaging, if my tone is appropriate for the industry, and if I effectively connect my past experience to the core requirements of the role. Please point out any sections where my claims sound too generic or lack specific evidence.</goal>\n\n<anti-goal>Do NOT rewrite any part of the cover letter. Do NOT provide a revised or “improved” draft. Do NOT focus on minor typos or grammar. Provide all of your feedback as actionable critiques in a bulleted list.</anti-goal>"], ["By incorporating tags for the AI's designated role, this setup directs the model toward delivering precise evaluations on the cover letter without venturing into unauthorized revisions."], ["<goal>Suggest 5 distinct neighborhoods to stay in while visiting Tokyo for a week. Focus on areas with great street food and easy transit access. </goal>\n\n<anti-goal> Do NOT build a day-by-day itinerary. Do NOT suggest specific hotels or restaurants. Keep the description of each neighborhood under three sentences.</anti-goal>"], ["Testing this with ChatGPT's latest GPT-5.5 Instant version yielded a precise output:\n\nShinjuku: A high-energy hub with endless late-night food alleys, standing bars, and yokocho streets packed with casual bites. Transit is unbeatable thanks to Shinjuku Station, making it easy to reach nearly every part of Tokyo.\n\nUeno: Known for its traditional atmosphere, open-air markets, and affordable street snacks around Ameya-Yokocho. Ueno Station connects multiple JR and subway lines, plus direct access to Narita Airport routes."], ["The key insight is that incorporating directives on prohibitions strengthens prompt effectiveness, particularly when framed within organized XML structures to guarantee comprehension by the model."], ["Ben has covered consumer tech for over two decades, currently emphasizing AI's impact on everyday life. His work examines recent large language models and their applications in professional and personal settings to equip readers for the upcoming AI shifts. 'AI will transform our routines faster than anticipated,' he notes. 'Daily engagement is the optimal path to adjustment.' Since joining PCWorld in 2014, Ben has reported on topics from laptops to security devices prior to spearheading the publication's AI coverage. His pieces have featured in PC Magazine, TIME, Wired, CNET, Men's Fitness, Mobile Magazine, and others. He possesses a master's in English literature."]}