Iterative prompting is often mentioned as the key to successful use of generative AI, but this is nothing more than a fancy new name for the high school writing process you were taught in your high school English class.
Here’s how the concepts of iterative prompting align closely with the writing process as it’s traditionally taught to students.
1. Prewriting → Prompting & Brainstorming
- In high school, prewriting involved brainstorming, outlining, and gathering ideas before drafting.
- With AI, the initial prompt serves as your brainstorming tool, generating ideas, outlines, and structure.
2. Drafting → Initial AI-Generated Content
- The first draft in writing class is like the initial AI-generated response—it’s a rough version that needs refinement.
- Just like writing a rough draft, the AI output should be seen as a starting point, not the final product.
3. Revising → Iterative Refinement
- In school, revision focused on improving clarity, coherence, and engagement—not just fixing typos.
- Iterative prompting works the same way: refining tone, structure, and details by tweaking prompts.
4. Editing → Targeted Adjustments
- Editing in the writing process meant focusing on grammar, conciseness, and clarity.
- AI users do this by fact-checking, refining specific sections, and adding a brand voice before publishing.
Getting Started With Iterative Prompting
1. Start With a Clear, Specific Prompt
The way you frame your initial prompt has a significant impact on the quality of AI-generated content. A vague prompt leads to generic results, while a precise prompt yields more targeted and useful content.
“Write an article about moving” isn’t helpful for any AI tool—or a human writer. Who is the audience? What is the purpose of the content?
A better prompt would be:
“Write a 1,000-word blog post on moving tips, covering packing strategies, hiring movers, and creating a moving checklist. This blog will be published on a moving company website with the goal of getting the reader to contact the moving company to receive a free, no-obligation quote for professional moving services.”
In my introductory journalism classes, one of the first concepts we discussed was the 5 Ws (and 1 H) framework for gathering information—Who, What, When, Where, Why, and How.
Ask yourself:
- Who is the target audience?
- What key information should be included?
- When will the reader be searching for content? (Are they ready to make a purchase or are they just gathering information?)
- Where is this content going to be published?
- Why does this piece of content deserve a reader’s time and attention? (Pro tip: You’re aiming for content that demonstrates Google E-E-A-T—which stands for Experience, Expertise, Authoritativeness, and Trustworthiness.)
- How should the content be structured to reflect the desired tone and style?
If you’re writing product descriptions or other types of marketing copy, copywriting frameworks can also be used to help you structure your initial prompt effectively.
2. Use Step-by-Step Refinement
Generating a polished piece of content often requires multiple rounds of prompting. It’s simply unrealistic to expect perfect results with the click of a button.
After you’ve created that first draft, start from the top and make tweaks to refine the output. For example:
- “Write a more creative introduction.”
- “Make the article more engaging by adding real-life examples and humor.”
- “Add subheadings and a checklist at the end.”
- “Add additional facts and statistics to the section on [TOPIC].”
- “Edit the section on [TOPIC] to avoid repetition.”
- “Rework the call to action to make it more attention-grabbing.”
Each iteration improves the quality of the content while keeping it aligned with your vision. I like to work in a Google Doc so I can easily revert to a previous version if I later decide I’m unhappy with my edits. Or, if it’s an especially important project, I’ll keep two different versions side by side on my laptop screen and read them out loud to determine what I like best. (I’ve tried many tricks over the years, but reading the text to my cat still seems to be the foolproof way to root out those robotic and clunky phrases that slip through the first draft of AI-generated content.)
3. Provide Style and Tone Instructions
AI can mimic various writing styles, but it needs guidance. Clearly state the tone and style you want for your content.
For example:
- “Write in a friendly and conversational tone, similar to a lifestyle blog. Avoid technical jargon.”
- “Write a blog post in a warm, friendly, and encouraging tone, as if giving advice to a close friend.”
- “Create an Instagram caption that feels inviting and cozy.”
- “Write a product description with a lighthearted and playful tone to appeal to young adults.”
- “Use a professional and authoritative tone, suitable for a business audience.”
You can also reference the tone of specific writers or publications.
- “Create a satirical essay, in the spirit of The New Yorker’s Shouts & Murmurs column.”
- “Write a comedic essay with the lighthearted and exaggerated humor of Dave Barry.”
- “Write a long-form article in the style of The Atlantic, balancing in-depth analysis with engaging storytelling.”
- “Craft a news report with the clarity and neutrality of The New York Times.”
If you have examples of your own writing, simply attach them to the prompt in the chat window. I typically start with five examples for the AI to mimic, but uploading additional content to reference may further improve your output.
If you’re using ChatGPT, adjusting the temperature and top-p parameters may also help you guide the output closer to your desired tone and style. Higher values produce text that is more creative and imaginative.
4. Ask for Multiple Variations
Generating multiple versions of content helps you compare different tones and styles to pick the best fit.
For example, you might say:
- “Give me three different introductions: one humorous, one formal, and one with a personal story.”
- “Provide me with five creative and attention-grabbing versions of the call to action.”
- “Suggest 10 possible headlines.”
The ability to ask for multiple variations of text without running into usage limits is one of the primary reasons I’d recommend paying for the pro version of at least one AI tool. However, if you’re working with free accounts only, you may want to try running the same prompt through ChatGPT, Claude, Gemini, etc. to compare results. Combining elements from each tool can often serve the same purpose.