Recently, I had a brilliant idea for a blog post - but it was doomed from the start.
When my boss told me that she wanted our next cluster of blog topics to be Artificial Intelligence (AI)-focused, I thought “Fantastic. I have the perfect starter post in mind.”
A Blog Post Idea for my AI Writer
The title came to me first: “AI-Powered Web Accessibility: Current and Future Tools.” I had already done some light keyword research on this topic so had an idea of what clusters of keywords had search volume and which did not. My goal was to create the most robust blog post humankind had ever seen on web accessibility tools that would not only describe said tools but also provide use cases and links to the tools themselves. In short, I wanted to play by Google’s Experience, Expertise, Authoritativeness, and Trustworthiness (E-E-A-T) search guidelines, ensuring that my post would yield incredible results because of how people-centric and helpful it would be.
My Plan for Creating an AI-Written Blog Post
I know that AI writing tools like ChatGPT are flawed. They tend to provide thin content that’s mostly drained of humanity and sometimes, usefulness. Therefore, you must give meticulous prompts to thwart its attempts to provide one-liners that fail to communicate complex ideas. The trick, I thought, was to be prepared for its garbage output and prompt it mercilessly until it bent to my will and gave the full body of topical content I wanted.
My approach was simple:
- Use AI to write an outline
- Have my outline vetted by our website accessibility expert
- Use AI to write the post
- Prompt and re-prompt until I got what I wanted
- Present the post to our accessibility expert and blow his mind
- Bask in the glory of my useful post as the internet throws roses at my feet
But Here’s How It Went
Initially, things started off great. I created an outline, our accessibility subject matter expert (SME) gave it a thumbs-up, and I started “writing.”
As imagined, I had to put up a fight to get robust paragraphs out of my AI writer. I had to prompt and re-prompt. I battled redundancies. I scolded the machine and told it “Not good enough!” and demanded its best work. I made sure to get all those use cases in there and to name names when it came to the tools and in the end, I was happy. It had a 3,000+ word count and completely delivered on the outline that had been approved by our accessibility expert.
I thought – this post is so good I should even gate it for our accessibility guy. Make him give me his personal email address before he’s allowed to read it.
But it was too early for a celebratory beer.
I delivered the piece to him, ready to receive his praise through our internal communication channel. I wondered - should I use AI to write my “thank you” speech? Or simply offer a humble “thanks man,” with a shrug when the praise came?
How the AI Writer Failed Me
Instead of praise, our accessibility expert came at me with a sort of “uhhh... let’s have a call to discuss this.” Wow, I thought – he loves the piece so much he wants to thank me on a call!
Or... it has issues.
Spoiler alert – it wasn’t the former.
On our call he explained to me that while our AI friend did technically deliver on the structure of my outline, the piece was riddled with inaccuracies that only an actual SME would be privy to. Many of the “AI” tools it described were not based in artificial intelligence at all – they were simply programs with functionalities that seemed intelligent. And furthermore, the list of current technologies available was neither complete nor helpful and the future technologies section was full of false and incomplete information as well. Of the 3,000+ words in this piece, there were only two or three usable elements – nowhere near the epic piece I had initially envisioned.
AI Has Not Replaced Subject Matter Experts
I took the news of my failure like a man and cried into my pillow for six hours. And after that, I faced the ultimate truth of the situation:
I am not an accessibility subject matter expert.
I’m a marketing guy.
My point of failure was not in the execution of the post itself. I followed the outline that had been approved and built a comprehensive piece accordingly. Instead, my failure was the naive assumption that, as a non-SME, I could rely entirely on Artificial Intelligence to create a piece on a topic I didn’t know enough about and would still be able to sort out any inaccuracies. So, lesson learned.
How I Should Have Handled This
Honestly, I knew better. At heart, I am a writer and thus far my use of AI for the written word has been light – using it mostly to reword or tighten small paragraphs or snippets when I’m in a hurry – and to mostly decent results. Remember my bulleted plan? Well, here’s what it should have looked like:
- Use AI to write an outline
- Accessibility SME vets outline
- Prompt and re-prompt
- Working session with SME, adding or subtracting content as needed
- AI is used to tighten up the language
- Everyone basks in the glory of a useful post, gathers up thrown roses
It’s not much of a deviation from my original plan. Slightly more time-consuming, sure. But it’s critical to involve SMEs and not risk publishing false, misleading, inaccurate, or incomplete information on topics that you care about and that you want your website visitors to care about.
SMEs Should Use Creative, Thorough Prompts
In “writing” the original piece, I somewhat smugly believed my prompts were meticulous. But the truth is, as a non-SME on the topic of accessibility, I didn’t know what I didn’t know. And therefore I was limited in my ability to thoroughly and effectively prompt to excavate the really good content.
Examples:
- For the “future accessibility tools” section, I had asked AI to describe technologies that were currently being developed and one of the items was “Smart Accessibility Customization Engines” with a blurb on User-Centric Design and some of the more advanced technical functionalities under that umbrella. I, as a marketing guy (non-SME), did not know that these advanced mechanisms are current real-life and not some distant, utopian dream. If I were an SME, I could have hung the proverbial swinging lightbulb over the AI’s head and demanded answers. Demanded to read its distinctions between what it described against specific, already-existing tools.
- Similarly, the section on “Ethically-Designed AI for Bias Elimination” was supposed to break down how AI systems are being consciously designed to eliminate bias and promote fairness through being trained on diverse datasets, incorporating ethical guidelines into AI development, and continuously monitoring and adjusting algorithms to prevent and address any biases that may arise. Sounds like incredible technology, right? Except what the AI gave me was flawed and because I’m not an SME on this topic, again, I didn’t know to press it. The truth on this topic is that AI learns through data sets it is fed. In the case of accessibility data sets, the data is often not truly representative of the communities the systems are being designed to serve. Put bluntly – the people in charge of designing such things are often out of touch with those who are differently-abled. Unfortunately, my AI-written article offered no mention of this fact, which is obviously a massive problem given that it’s such a well-known issue in accessibility circles. Without the fortune of having an SME backing me up, I could have published half-true information.
A Major Takeaway
AI is great at writing- surface- level, short content, e.g., concise blurbs like meta descriptions or headlines for ads. But for longer pieces, it’s just writing what it thinks humans want to read. Computers (the AIs) are interpreting internet data sets to guess what computers (search algorithms) think humans want to read, which is an inherently flawed dynamic (for now). For writings more substantial than blurbs, and for many industries, heavy SME involvement is still very much needed.
The Final Word on AI Content Marketing
AI will not be ruling us any time soon. For now, it remains our tool. Something to use with discretion and leverage for time-saving tasks rather than relying on it entirely. It needs us as much as we need it.
Content marketing is still human Subject Matter Expert-dependent. Organizations should implement artificial intelligence tools and policies as a part of their overall digital transformation strategies .
