All too often, ChatGPT, Claude, and Gemini overstep their instructions because they’re so focused on making you happy. For example, an AI may jump ahead and completely rewrite a document when all you wanted was some focused feedback, or it may draft a brand-new recipe when you just wanted help substituting an ingredient.
You might think the solution is to tell the AI chatbot what it shouldn’t do in your prompt. But that could backfire. In its rush to provide a helpful response to the first several words in your request, it may deemphasize your restrictions—or skip them completely.
The trick is to structure your prompt in a way that forces the AI to stop and give each part of your request its due weight. You can even tag each segment of your prompt to tell the AI what their purposes are.
AI chatbots like ChatGPT, Claude, and Gemini are capable of recognizing “markup” elements that provide hints to the structure of your request. With markup elements, you’re essentially saying things like “This first part of my prompt is background context for my request” and “This middle part of my prompt contains instructions on what to do” and “This last part of my prompt is telling you what not to do.”
We’re going to use XML markup, and we’re going to use open/close sets of XML tags to define the areas of a prompt. In this example, one set will define the “goal” of the prompt (e.g., what you want the AI to do) and another set will define the “anti-goal” of the prompt (e.g., what you don’t want the AI to do).
Let’s say you want help writing a cover letter for a job application but you don’t want the AI to write (or rewrite) the letter for you. Instead, you simply want feedback on the letter you’ve already written.
Here’s an example “anti-goal” prompt that should keep the AI in line:
Expert Career Coach and Hiring Manager
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.
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.
This prompt, which also includes XML tags that tell the AI what its “role” should be for this prompt, should keep the AI focused on its primary task (giving you feedback on your cover letter) while stopping it from doing unwanted work (jumping ahead and re-writing your letter).
Here’s another example:
Travel Guide
Suggest 5 distinct neighborhoods to stay in while visiting Tokyo for a week. Focus on areas with great street food and easy transit access.
Do NOT build a day-by-day itinerary. Do NOT suggest specific hotels or restaurants. Keep the description of each neighborhood under three sentences.
I tried this prompt with the ChatGPT’s new GPT-5.5 Instant model and got this clean, focused response:
Shinjuku
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.Ueno
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.Asakusa…
The takeaway here is that telling your AI what not to do is one of the most powerful signals to include in a prompt, and doing it via structured XML tags helps ensure the model gets the message.
