# Why Well-Written Prompts Are Almost Always Structured > Source: > Published: 2026-06-30 06:44:59+00:00 [Career Growth](https://www.dataleadsfuture.com/tag/career-growth/) # Why Well-Written Prompts Are Almost Always Structured A Plain and Intuitive Explanation People are arguing again about whether prompts should use HTML or Markdown. Last year, the debate was Markdown vs. JSON. But to me, the answer has nothing to do with the format itself. Markdown, JSON, HTML, they all work well for the same reason: structure. As for which one to pick, it depends on which format appeared more frequently in your model's pretraining data. Prompt engineering is not about feeding the model more information. It's about helping the model and the human put limited attention in the right place. Structured language does exactly that. With just a few markup symbols, it pre-allocates attention across the text. Through Markdown, JSON, and HTML tags, language stops being a linear stream of words that only flows left to right. It becomes something with hierarchy, emphasis, attention weights, and a multi-dimensional shape. Structured text does not just change how an LLM pays attention. It changes how you pay attention, too. Let's look at a passage in plain language first: We're going on a picnic on July 1st, and the destination is Central Park. We're driving. You take the ring road from the south side of the city, and I'll drive straight down from the north. We'll meet at Entrance 2 of the park at 8 AM. Remember to bring tuna sandwiches, bottled water, and a Bluetooth speaker. I'll bring the picnic blanket, the tent, chicken sandwiches, and bug spray. Also, don't forget to book the park tickets in advance on the app. You can only book one day ahead, so search for Central Park in the app. Now let's try structuring that same text, and strip out the natural language filler (conjunctions, prepositions, all the connective tissue, since the structure already shows the relationships and none of that is needed anymore). Here's what it looks like after Markdown rendering: ``` ### Picnic Location Central Park ### Meeting Time and Place - July 1st, meet at 8 AM - **Entrance 2** of the park ### Getting There - You: drive via the ring road - Me: drive straight from the north side ### What to Bring - You: **tuna** sandwiches, bottled water, Bluetooth speaker - Me: **chicken** sandwiches, picnic blanket, tent, bug spray ### Important Reminder - Must book in advance on the app, search *Central Park* - **Can only book one day ahead** ``` Now, if someone asks which entrance to meet at, you can find the answer so much faster, right? Through categorization and emphasis, the important information already pulls your attention toward it. It doesn't get buried inside a wall of text that means nothing at first glance. All of that change comes from tweaking just a handful of symbols to introduce a complete structured syntax: ``` ### Heading **Bold** *Italic* - List 1. Ordered list ``` So back to the original point: why do structured prompts work better? Because structured text is no longer natural language. Natural language lets attention drift, for both LLMs and humans. Structured language anchors attention on purpose. A lot of decisions and responses depend on where attention lands, not on how smart you are. Here's a simple example. You need to buy a bag of pasta at the grocery store. You're incredibly smart, but you're hungry right now. Your attention goes straight to the serving suggestion photo on the packaging, and it looks delicious. So you grab a bag without thinking twice. Standing next to you is a nutritionist. Whose attention goes straight to the ingredient list and nutrition facts. His reaction: too many calories, I'm trying to lose weight, putting it back. You didn't miss the calorie count because you can't read a nutrition label. You missed it because the pretty food photo on the package caught your eye first. But if someone told you beforehand that high-carb foods make weight gain easy, you'd probably think twice next time you reach for pasta. That story makes two points: - Attention drives what happens next. - Attention can be shaped, at least to some degree. Guiding attention inside a prompt matters a lot, then. Rather than hoping an LLM will notice the ingredient list on its own, just call it out directly in the prompt: ``` ### Pay Close Attention!! 1. **Do not** focus only on the cover image. 2. **Always check the ingredient list**. ``` You mark it because you're not leaving it to chance. That's the simplest logic behind structured prompts, and it's logic both you and an LLM can follow. So back to the question at the start: HTML or Markdown? It doesn't matter. What matters is whether you're using structure to anchor attention for the model and for yourself. The symbols are just the tool. Structure is the whole point. Thanks for reading and subscribing. If you found this helpful, feel free to share it with a friend. ## Further Reading Here are prompts and LLM tips I often use at work, sharing them with you: You pour your heart and soul into building an AI agent system, only to find out after launch that it's just another barely functional demo product. Sound familiar?