How can AI help you write better?

I've written recently about how I use AI for writing. This week I’d like to share some details on how I use generative AI in my writing workflow.

Spoiler alert: I don't let AI write for me. Whether I'm writing for a client, for my blog, or here on LinkedIn, I do the writing myself, and I use AI only as an assistant.

And that's the top piece of advice I have for writers working with AI chatbots: Treat them like interns or research assistants you are giving assignments to.

Simply put, do not expect the AI to complete any work for you at the same level as you. Instead, treat it as a helper and partner. Provide clear instructions and be prepared to review, correct, and enhance most, if not all, of its output.

It helps to think of AI as a colleague who is pretty clever and has access to an unbelievably large quantity of source material but who is utterly lacking in common sense or any notion of originality.?

You know that coworker who is kind of a know-it-all and also a bit of a con artist? The one who will make up answers and pretend to know things even if they don’t? That’s your chatbot coworker.

If you understand these limitations and can work around them, AI chatbots can be powerful collaborators.?

For example, I've used AI to help me brainstorm ideas, start my research, fill in gaps in outlines, and generate alternative phrasings for me to consider. I take what it generates with a big grain of salt, but the ideas and phrases it generates often inform the outlines and drafts that I create.

Important note: If you’re writing for clients or publications, make sure they know you are using AI in your writing process. Some won’t care; others will have policies against AI use that you need to respect. For example, The Hill and Entrepreneur now have policies against AI content, and Harvard Business Review requires writers to disclose any AI they use.?

Give the chatbot a clear, detailed assignment

Whether you’re using a chatbot for research or help with writing, put as much specificity into the prompt as you can.

The pointers I shared on how to start a writing project with a team are equally valid when your “team” is you plus a chatbot.

As with an assignment brief, it’s useful if your prompt includes a few specifics:

• A working title
• A short description of the content you’re working on — just a sentence or two describing the idea is enough, or a full abstract if you’ve got that.
• The purpose for which you’re writing this piece. What’s the goal you hope to accomplish?
• The target word count you’re aiming for
• The target publication, and/or any details on the audience that you’re hoping to reach
• Any notes on the tone or style you’re hoping for (formal or casual, technical or general-interest, use of “I, we, you, they” pronouns)
• Any supporting notes and resources, such as transcripts, white papers, blog posts, or news story links.
• You can also list any stats or data points you’d like the story to include.

Warning: Don’t ever put anything confidential into a chatbot prompt or into a customized chatbot, like a custom GPT on ChatGPT. Chatbots can leak information, through attacks known as prompt injection and other means.

Revise your prompt iteratively

If you don’t get what you want, revise your prompt and try again.

If the chatbot provides a response that seems unusually well-articulated but is lengthy, unoriginal, and lacking in specificity (the default for both ChatGPT and Claude), you can modify and resubmit your prompt with specific instructions to address the issue.

For example, here are some instructions you might add to your revised prompts:

• Limit your responses to five sentences or less.
• Don’t use more than three sentences in each paragraph.
• Vary sentence and paragraph length to provide variety and musicality to the writing.
• Use direct, conversational language, with short sentences and varied sentence structures.
• Eliminate fluff from your response. Get right to the point.
• Avoid jargon in your response.
• Avoid the following words: delve, showcasing, underscore, comprehensive, crucial, intricate, pivotal. (These are 7 words that suggest a text was written with AI. I would add a few others to that list: transformative, ensure, critical.)
• Provide examples, preferably with a human dimension, in order to give your writing a more emotional, relatable aspect.
• Provide sources for each argument, with links.

Check everything

As I wrote last week, generative AI platforms are dream machines: Their job is to make up stuff. While they often produce factual statements, they don't do any fact-checking, so errors do slip in -- sometimes egregious ones.?

It’s up to you to check everything and make sure that the AI isn’t bamboozling you. That includes clicking through on any links and checking that whatever it links to actually corresponds to the AI summary. You don’t want to be those lawyers who got in trouble in 2023 for citing nonexistent precedents in their legal brief.

One thing you might try: Use one chatbot to check the work of another chatbot. For instance, you could take a paragraph from ChatGPT and feed it to Claude with this prompt:

• You are a fact-checker. Please alert me to any factual errors or inconsistencies in the following text.

I used this prompt for the first two paragraphs in this section, and Claude's answer was pretty good. Its summary: "Overall, while the language is somewhat colloquial and uses some generalizations, the core message about the need for human verification of AI-generated content is accurate and consistent with current understanding of AI limitations."

NOTE: This prompt can surface potential issues, but I wouldn't rely on any chatbot for fact-checking, ever. Always check any facts yourself.

I'd love to hear how you're using AI in your writing. Let me know in the comments!

***

BUT WAIT THERE'S MORE. For more AI prompts and tips for writers, check out my blog post, "8 generative AI prompts to help you write better." Many of the tips are only for subscribers of my newsletter, but it's free to subscribe!