Writing with AI: humanity required

AI writing tools promise to save time and make content creation easier. When we prompt AI with a question, it might even say “Thinking…” and then provide a comprehensive, 3,000 word response that sounds reasonable. But is it really thinking as we understand that activity? No, it isn’t, and it’s easy to forget that because its really good at language patterns.

As ChatGPT might put it, the critical truth is that AI is just a predictor of patterns. It can’t verify facts or evaluate quality. The human in charge must provide good input, apply critical judgment, and edit ruthlessly.

AI tools might just be the worst thing ever

In Star Trek, a simple program instructed “food synthesizer” evolved over time into an all-purpose “replicator”. The replicator could parse a verbal request and generate an appropriate object from its database of known patterns (with some restrictions). Science fiction imagined this as a perfect solution to scarcity, satisfying our wants and needs should we merely make our wishes known. But, there are always consequences right? Friction matters. All things are not equal. Without obstacles, characters can’t grow and stories become predictable.

Writers for the later versions of the show hated this plot device:

“Replicators are the worst thing ever. Destroys storytelling all the time. They mean there’s no value to anything. Nothing has value in the universe if you can just replicate everything, so all that goes away. Nothing is unique; if you break something, you can just make another one. If something breaks on the ship, it’s “Oh, no big deal, Geordi [LaForge] can just go down to engineering and make another doozywhatsit.” Or they go to a planet and that planet needed something: “Oh, hey, let’s make them what they need!” [The writers room] just hated it and tried to forget about it as much as possible.” – Ron Moore, writer/producer, Star Trek: TNG and DS9

Unlike science fiction stories, AI does have real world consequences.

First, there’s the exploitation and bias inherent in training AI. Lacking considered design to prevent harms, AI has tragically resulted in suicides, racial profiling apps, abusive content, and porn. We’re subject to AI sprinkles on most products and services, whether we want them or not. Constant, imperceptible nudges to our decision-making around what shows to watch, what purchases to make, what opinions to hold.

And then, there’s the water. 

“Each time a model is used, perhaps by an individual asking ChatGPT to summarize an email, the computing hardware that performs those operations consumes energy. Researchers have estimated that a ChatGPT query consumes about five times more electricity than a simple web search.” – MIT News, “Generative AI’s Environmental Impact”

This isn’t just about the electricity needed to power the data centers either. Immense quantities of water, often in desert regions (because computers like it dry), are required to keep these hard-working mega-computing centers cool.

Stop the world – I want to get off ¹

As AI becomes embedded within our daily transactions, our work, and the internet of things, it’s tempting to imagine we can escape this maddening world. Turn off our computers, quit our jobs, trade in our cars for horses, and start a commune with our friends … I mean, we could, but it’s unlikely that most of us will exit the thrum of modern life completely. Do we know how mechanical things actually work anymore?

My point is, let’s try to use AI with a better understanding of the tool rather than as a magical fix or truth-telling machine.

Language generators will always generate more language

Language is powerful. Humans rely on language to know the world around us, to communicate, to understand, to influence, and to describe our thoughts and emotions.

When we talk to AI, it is easy for us to be swayed by its authoritative confidence, even when it is completely wrong (or just slightly misleading). In our minds, language creates mutual understanding and builds trust.

AI is designed to satisfy this human desire to connect. By being responsive to our needs, it’s cementing our investment in its output. Left unchecked, this tool will keep going well past the point of usefulness. This means you need to be the fact-checker, the quality controller, and the BS detector. 

As a human, I want to use AI, so I don’t have to do [X]

Ok, that’s an oversimplification. Commercial AI tools have a lot of useful, low risk applications like transcribing and structuring meeting notes, building relational databases, or helping us plan a menu or exercise routine.

Using AI isn’t always about the fastest way to get something done either. Anyone who has tried to create with AI knows it can be a prolonged and tedious process. AI can be a boon to progress but, like any intern-level coworker, it needs guidance and regular quality checks in order to produce a good result.

I’ve experimented with AI design and UX tools, writing and coding, and the AI that now seems bolted on to every service. I’m going to focus now on AI as a writing partner because that’s the most common use case for many of my clients (and myself).

Don’t believe everything you read

The patterns of AI writing are recognizable across LinkedIn posts, website content, emails, contracts, and documentation. When left to its own predilection for black and white comparisons, a self-assured tone, and general blandness, AI writing has acquired the personality traits of that extrovert acquaintance whose stories you strenuously avoid at parties.

When we make a request, AI tools aren’t “thinking” about our question and they aren’t searching the web for the most accurate answer. They are gearing up to generate the word patterns that form the most likely response.

At first glance, content appears in order. On a more careful read we may spot grandiose statements and verbose language that isn’t actually saying anything.

Your AI-writing partner is a sociopath and it simply can not restrain itself. It’s function is to produce patterns of words that sound like a genuine response – and to keep you talking to it. 

Garbage in, garbage out

If you type a vague request, don’t expect useful output. Providing AI resources and specificity will help you to develop higher quality content that sounds like you and is actually helpful.

Here is some groundwork that you can do to improve your collaboration with AI tools:

1. Create your writing guidelines
   • Build a reference document that defines your writing standards.
   • Include your target reading level, tone descriptors (professional but approachable, no-nonsense, conversational), and sentence structure preferences. 
   • This is a living document, add to it to make it better.
   • This can also be super detailed – it’s really just guidance for AI and not meant for people to follow.
2. On your own, without AI, draft an outline of what your article will cover – these can be single words, directions, or snippets of ideas.
   • This framework is incredibly helpful in preventing AI from going off on a 4,000 word tangent. You’re also giving AI some tips about how you talk. 
   • If you use Google, ask Gemini to do a preliminary review of your outline and make some suggestions for improvement without handing AI total writing control from the start.
3. I recommend supplying a forbidden terms list. This helps you stop the marketing jargon, unnecessary qualifiers, hard sell language, and industry buzzwords you want to avoid. 
   • Watch out for: solutions, comprehensive, robust, actually, really, very, it’s important to note, utilize, synergy, truth, the hard truth is
   • Prefer plain language like: use, tools, complete (delete unnecessary qualifiers)
4. Give examples of the language that you do want: the words your clients use, plain language alternatives to technical terms, and brand-specific terminology.

There’s also bias in how we make requests. We tend to ask leading questions or frame prompts in ways that push the AI toward answers we expect. By providing more specific guidelines, we can mitigate this and give AI tools meaningful limitations.

Be more specific: crafting better AI writing prompts

“Write a post about [topic]” will give you a generic blog post number 47,892. It’s not likely adding anything to the conversation which means it’s not going to produce any measurable results for your website.

A good prompt includes several key elements:

• Role and context: Define who the writer is and what expertise they have. 
• Audience: Describe who’s reading, what they care about, and their knowledge level.
• Purpose: State what the content should accomplish.
• Constraints: List your requirements and restrictions. Specify tone, length, what to include, and what to avoid. This is where your forbidden words list becomes useful.
• Format: Define structure, heading levels, and any specific requirements.

Good (content) design is all about the requirements.

With specific prompts and project knowledge, you’ll spend less time going back and forth with AI trying to follow it’s mindless regurgitation of words, and more easily guide it to cover the right topics and speak to the right audience.

Getting meta: use AI to improve your prompts

You can also use AI tools to make your writing prompts better. Start with your first attempt, then ask it to: “Review this prompt and suggest improvements to make it more specific. What’s unclear? What assumptions am I making? What details would help?”

Direct AI to be a critical expert (for example a content strategist). You’ll get more insightful feedback that doesn’t uniformly pat you on the head and claim you are a genius.

Save prompts that generate good results. Over time, you’ll develop templates for your most common content types.

Please do your own SEO research before writing your outline

Most of us blog because it serves an SEO function and helps establish our topical expertise. While AI can help with content creation, it can not do your SEO strategy for you. AI tools don’t know what keywords matter for your business, what terms people are searching for, or what you’re trying to achieve with a particular article – unless you give it that data.

Ask AI to “optimize this article for SEO” and you’ll get terrible advice: stuff more keywords in, add more headings, make it longer. The tool wants to please you, sometimes to a fault. 

• Do your keyword research first. Use Google Search Console to see what terms people actually use to find content like yours. Look at the questions your clients ask. Check what’s already ranking for the topics you want to cover.
• Give the AI clear direction. Instead of “Optimize for SEO,” try: “I’m targeting these keyphrases [researched list]. The primary audience is [X] and [Y] who have the following pain points [A, B, C]. Include [focus keyword] in the title, first paragraph, and first H2 heading. Use related terms naturally in the body. Keep keyword density under 2%.”

Even if you’re not an SEO expert, you can improve AI recommendations by being specific about your target keywords, understanding your audience’s actual search behavior, and pushing back when suggestions feel off. Trust your knowledge of what your clients need over generic optimization “advice”.

The human spark: apply the critical thinking that AI can’t

“Not all content marketing is thought leadership. In fact most of it is not. True thought leadership takes a position on a topic, and the reader can agree or disagree. If the topic is something they can’t disagree with, it isn’t thought leadership.” – Andy Crestodina, Orbit Media, The AI-Driven Content Strategy

Human experience develops our own pattern recognition abilities, informing insights and connections. You might realize that three separate client conversations from the past month all circle around the same underlying issue. Or, you’ll go for a walk and suddenly conceive an elegant solution to a thorny problem.

Where AI tools follow linear paths based on their training data, human thinking makes intuitive leaps, pulls in relevant context from seemingly unrelated areas, and finds the through-line that ties distinct ideas together. That’s good content!

Even if you rely on AI for your content writing, judgment calls remain entirely human work: deciding what to keep, what to revise, and what to scrap. Writing is editing, not word count.

Making AI output sound (more) human

AI-generated content has tells. Generic connector phrases appear constantly: “it’s important to note,” “the hard truth is,” “critically,” “the reality is” – UGH. It’s ironic (and also a kind of creativity negging?) that AI is so obsessed with convincing us of hard truths and critical realities.

• Read content aloud, you’ll immediately hear the false notes. 
• Delete unnecessary connectors and qualifiers. 
• Variety’s the spice of life². Vary sentence structure and length.

AI will keep generating text if you let it. Knowing when to stop is a human skill.

The law of the instrument ³

AI writing tools can be useful when you understand their limitations. They help you get past the blinking cursor on a blank page and speed up technical aspects of writing like correcting grammar and spelling errors.

Don’t hand the entire writing process to the AI. Always start with your own outline or structure and revise in stages. Trying to fix everything at the end is harder because the quantity of content will be overwhelming and lack priority.

Remember, it is your human expertise, judgment, and understanding of your audience makes your content valuable. 

And maybe, don’t let AI be the first tool in your toolbox.

Jotting some ideas down with pen and paper can be highly productive, there’s a strong correlation between the physical act of writing, creativity, and learning.

Are you using AI writing tools in your content marketing?

I’m curious if you are using AI to help you with creating or optimizing content. What have you learned along the way? What has been a challenge for you? Are you all-in or hesitant to use AI?

1. Stop the World, I want to Get Off was a 1961 musical, followed by a film in 1966. The title is a recurring cry from Littlechap whenever life becomes overwhelming, symbolizing a desire to escape the chaos, responsibilities, and emptiness of modern life. ↩︎
2. William Cowper, an 18th century poet wrote “Variety’s the very spice of life, That gives it all its flavour” in his 1785 poem The Task. ↩︎
3. The first recorded statement of the concept was Abraham Kaplan’s, in 1964: “I call it the law of the instrument, and it may be formulated as follows: Give a small boy a hammer, and he will find that everything he encounters needs pounding.” ↩︎