A beginning guide to prompting like a writer
How to get better AI drafts with great prompts
Most people give AI the bare minimum. Their prompts are something like, “Write a blog post about our product launch.” Then they’re surprised when the output is full of buzzwords, light on substance and totally forgettable.
If you’re struggling with using AI for content writing, the issue isn’t the tool you’re using. It’s your prompting process.
AI chatbots like ChatGPT, Claude and Perplexity are incredibly literal. They give you exactly what you ask for. Ask vaguely, get vague output. Ask specifically, with clear constraints and actual direction, and you’ll get something you can actually work with.
This guide will show you how writers approach prompting to get better, more usable results. Fair warning, though — even with great prompts, AI content still requires significant editing, fact-checking and multiple iterations to get right. If you missed it, we covered what to realistically expect from AI tools in our introduction to this series.
Our recommendation: Start with a human
Before we get into prompting techniques, let’s take a step back. You’ll get better results if you write your first drafts yourself, not with AI.
Human-written first drafts lead to better final content. When you start with a person thinking through the strategy, you begin with actual insight about your audience and their problems. When you start with AI, you’re building on a generic foundation (think of all the boring content your AI has ingested during training — that’s your starting point with AI-generated first drafts) and trying to edit it into something specific and valuable. It’s backwards.
AI-generated content now outnumbers human-written content online, and quality degrades as these systems learn from their own output. The internet is becoming an echo chamber of increasingly generic material. In this environment, content that reflects genuine expertise and understanding will stand out more than ever. Companies that can produce work demonstrating real taste — meaning you understand what excellence looks like in your field and can create something original — have a significant advantage.
And drafting with AI will not save you time. Despite what you’ve been told about AI being a magical efficiency tool, creating drafts with AI and then editing them usually takes longer than just writing it yourself in the first place.
Still, we’re fielding a lot of requests from marketers who don’t have a choice about this. Their leadership has mandated AI tools. Budgets have been spent. Some marketing teams are expected to use ChatGPT, Claude or internal AI tools for content creation.
If that’s you, keep reading. We’ll show you how to get workable output so you can build on it.
How to start prompting AI for better content
Good AI output starts with good instructions. You wouldn’t hand a junior writer a one-sentence assignment and expect a polished draft. The same principle applies here.
Before you start any piece of content, give the AI clear foundational directions. If you’ve created a brand guide, now’s the time to reference it or paste the relevant sections into your prompt.
Here’s what to include in those upfront instructions:
Who the piece is for: Get specific about your audience. Instead of something like “marketers,” say “demand gen managers at Series B SaaS companies who are frustrated with lead quality.” The more specific you are, the better the AI can write for that person instead of some vague category.
What your audience is struggling with. What does your audience deal with every day? What keeps them up at night? If you don’t tell the AI about these pain points, it can’t address them in the content.
What this content should accomplish: Are you trying to educate, persuade or build connection? Do you want the reader to book a demo? Be clear about the purpose so the AI knows what kind of content to create.
How the writing should sound: Describe the tone you want. Conversational or formal? Technical or accessible? Better yet, provide examples of writing that illustrate the tone you’re going for.
The format you need: Blog post, email, LinkedIn post, landing page copy — each has different requirements. Include length targets, too. A 1,500-word blog post is structurally different from a 150-word LinkedIn post.
You might need multiple prompts to cover all these instructions. That’s fine. Just add “Don’t do anything yet, just confirm” or “Wait for instructions” at the end of each prompt so the AI waits until you’ve finished explaining everything.
Define your style rules
Tone and audience get you halfway there. The other half is style constraints at the sentence level.
One of our writers learned this the hard way after spending 20 minutes editing an AI-generated article where every single section started with “But here’s the thing” or “Here’s what’s interesting.” By the fifth paragraph, she was ready to delete the whole thing and start over. The overall tone was right. The content was useful. But repetitive sentence patterns made it sound like it was written by someone who’d read exactly one article about conversational writing and decided to copy that style forever.
The more specific you are about style constraints, the less time you’ll spend editing later. You’re teaching the AI to avoid the things you’d just delete anyway.
Tell the AI your specific style rules upfront:
Sentence structure preferences: If you hate sentences that start with certain words or constructions, say so. Examples: “Don’t start sentences with ‘But’ or ‘And’” or “No question fragments like ‘The result?’ or ‘The takeaway?’”
Punctuation preferences: Some brands use em-dashes liberally, others avoid them. Some love semicolons, others never touch them.
Forbidden patterns: AI loves certain constructions that sound smart but get old fast. This might include forced contrasts like “It’s not X, it’s Y,” or three-part lists that feel manufactured. The more you use AI, the more of these quirks you will notice. Our team has a bugaboo about hanging “-ing” participle phrases at the ends of sentences.
Banned words and phrases: Certain words and phrases show up constantly in AI-generated content. We maintain our own banned list (you can see it here), and we recommend you build one too. Every time you catch yourself deleting the same phrase, add it to the list.
If you want to save time and effort on future projects, both Claude and ChatGPT have Projects features that let you save all these foundational instructions in one place. Set up a project with your audience, add your tone and style rules once, and you won’t have to re-explain everything each time you start a new piece. Just open the project and start prompting.
The outline approach: Working section by section
Once you’ve set up your foundational instructions, don’t ask the AI to write the entire piece in one shot. Give it an outline first, then prompt for each section individually.
A client sent us a 1,200-word AI-generated blog post asking if we could “just clean it up.” The intro was actually pretty decent. But somewhere around paragraph four, the AI apparently forgot what it was writing about and started repeating the same three points in slightly different words. By the end, it had contradicted itself twice and wrapped up with a conclusion that had nothing to do with the opening hook. Cleaning it up would have required completely rewriting 80% of it.
This kind of thing happens all the time when you ask for a complete piece in a single prompt. You will likely end up with a draft that requires massive structural surgery.
Working section by section gives you control. You can course-correct as you go instead of discovering problems after the whole thing is written. The AI focuses on one clear task at a time instead of trying to juggle the entire piece in one go, and you get more thoughtful output for each section.
We’ll give more details on effective iterative drafting in our upcoming posts.
Red flag warning! AI will confidently give you wrong information
AI wants to be helpful, which means it might invent facts, misremember sources, or present assumptions as truth. This is called hallucination, and it happens regularly enough that you need to plan for it. When AI doesn’t know the answer, it often just generates something plausible-sounding instead of saying “I don’t know.”
You’ll need to fact-check everything the AI generates. Claims, statistics and quotes need verification. We’ll cover how to do this efficiently in our upcoming fact-checking guide, but the short version is to trust nothing without checking it yourself first.
What AI still can’t do (and why you matter)
Even with excellent prompting, AI has fundamental limitations. Your expertise is what makes the difference between generic output and content that actually works for your business.
AI doesn’t know your brand voice or your audience’s specific pain points unless you teach it explicitly. It can’t make strategic decisions about what to emphasize or what to cut. As a human writer, you know what’s been said to death in your industry. Your expertise shapes everything that matters — the prompts, the structure and the strategic choices about what goes in and what stays out.
Think of AI as a capable assistant who needs clear direction, not as a replacement for your judgment and knowledge.
Your prompts will improve with practice
Prompting gets easier the more you do it, and the improvement happens faster than you’d expect.
You’ll start noticing patterns in what works. If the AI consistently ignores a certain instruction, you’ll learn to phrase it differently. When a particular prompt structure produces output you like, adapt it for other projects. Over time, you build up a collection of reliable prompts and develop instincts for when to course-correct during the drafting process instead of discovering problems after the fact.
Save the prompts that generate decent output so you can reuse them. Notice which instructions the AI follows reliably and which ones need more emphasis. Pay attention to what works for your specific needs and audience.
Start with the foundation we’ve outlined here — clear upfront instructions, defined style rules, and section-by-section prompting. Give yourself a few projects to practice, and you’ll find yourself spending less time wrestling with the tool and more time on strategic work.
In our next post, we’ll cover how to incorporate source material into your prompts without ending up with plagiarized content.
If this sounds like more work than you bargained for
Getting good output from AI requires strategic thinking, detailed instructions, ongoing feedback and thorough editing. It’s essentially project management combined with teaching, on top of everything else you’re already doing.
If that doesn’t sound appealing, there’s a simpler route.
At Horizon Peak, we create AIO-optimized, human-drafted content that connects with readers and gets cited by AI platforms. No prompt wrestling, no hallucination fact-checking and no marathon editing sessions. Just strategic content written by people who do this full time. If you’d rather focus on your actual job while someone else handles the content creation, let’s talk.
Check out the other articles in this series:
Your AI writing process is broken — let’s rebuild it
How to create a simple brand guide that makes AI write like your company
Structuring inputs and source material for AI drafting success