Blog
So you’ve been asked to “humanize” AI content
Aug
Searches for “How to humanize AI content” have increased by 943% over the past 12 months.
More and more people are trying to “pass human scrutiny” to make AI content less bad or to trick AI content detection tools.
The process of “humanizing AI content” has become the de facto content strategy for growth-oriented companies.
I see a lot of narrow-minded authors popping up everywhere asking for advice on how to best tackle this content problem.
And I know how soul-destroying that can be because I have this shirt.
This is why it is an effort doomed from the start.
We all know the solemn and glaring problems with AI content (and if you don’t, you can read about them here ), but I don’t think anthropomorphizing a bot is the solution.
In fact, I argue that humanizing long-form AI content is an inherently flawed process.
The whole point of using AI is to boost speed, scale, and efficiency; forcing humans to edit and essentially rewrite long texts produced by AI completely undermines that value.
Before you do the research and add the “wealthy” human element, you might as well create the content yourself.
The process of passing “humanized” AI content through detector is nonsensical – you literally get an AI that is supposed to validate your humanity.
I spent hours editing AI content I was charged with so that it could pass human scrutiny, only to be tricked into believing that my content was AI… by AI?!
And I’m not the only victim. In Post on LinkedInContent consultant Kiran Shahid said AI detection tools would only validate her humanity if she wrote in the first person — a writing style that clients were unlikely to accept.
I have seen similar experiences shared in SEO Women In Tech Community…
And further Reddit…
Having humans write AI-generated content and then having the AI evaluate their humanity isn’t just weird — it’s a blatant waste of talent.
When I was asked to “humanize AI,” my workflow looked something like this:
- Receive batches of 100% AI-powered content for editing
- Make sure each article contains at least 50% “human” elements in AI detection tools
- Change all the circles to green in the SEO traffic lightweight tool (pro tip: keyword-related stuff. Supposedly works wonders in 2024!)
The articles I was sent about AI were so far from the truth that I had to rewrite them (but also had the pressure of an unrealistic deadline), and even then the end result was destitute.
When you forcefully stuff in examples, opinions, and observations that could have shaped the article if only they had been included from the beginning, you inevitably end up disrupting the flow of your content.
At this point I want to make something clear. I am not anti-AI. I apply it daily for ideation, structure, outlines, and reuse – my most critical workflow is improving the pace and readability of content I have written in “flow”. Used in this way, AI can be incredibly powerful.
But the process of “humanizing AI” — you start with 100% AI-generated content and animate it as if you were Mary Shelley breathing life into Frankenstein’s monster — is one of the most tedious tasks and extremely demotivating for writers (even though I just put it in a nippy way).
If you are someone who uses this strategy, you should know that it is a great way to slim down your talent pool.
Humanizing AI content is wasteful and incapable, but some companies will force you to do it anyway.
If you find yourself in this situation, you can make your life easier by teaching AI how to handle your content from the start.
This means that whenever you receive a batch of AI-generated articles for “humanization,” you can run them through your self-trained AI model and bring them to a more acceptable level of humanness before returning to make your own edits and additions.
Training an AI model is a clever move because it improves the AI system itself; otherwise, it’s just tedious, manual work and you might as well have written the content in the first place.
Tools like Claude allow paid users to create shared “Projects“containing up to 200,000 words of training documentation, while Custom GPT ChatGPT allows you to train up to 20 files, each with a maximum size of 512 MB.
Nishkarsha KotianSEO Specialist in Channel engine introduces brand guideline examples to customGPT and sets the rules for the bot to follow.
Similarly, Melissa PoppDirector of Content Strategy at ShakyRootrains its non-standard GPTs on knowledge documentation and adopts a process of “a few encouraging shots“by ensuring examples her best content.
Given that AI tools have “memory,” you can also practice “iterative refinement” while talking to them.
In other words, you can train them piece by piece, using tips and examples to improve the overall quality of your content. Digital PR Expert Gini Dietrich describes AI as “digital trainees” in this regard.
The thing is, training takes a lot of time, and even with the whole context in mind, if you don’t have a clear vision of what you expect from your digital trainee, you can easily get stuck in an endless loop of messages.
So training an AI model can take some of the burden off, but it is not a foolproof method.
Melissa Popp notes that customGPTs only allow her to achieve 40–50% of her goal, and the training effort often negates the benefits.
Questioning “efficiency” strategies like humanizing AI can be tough – especially if they are imposed from above.
Decision makers often don’t understand the specifics of the job and only care about making sure the numbers on the spreadsheet are green.
But your job is to make this reality impossible to ignore.
In my case, I took the following steps:
- I was regularly informed about the problems I encountered during the “humanization” process at meetings
- I tracked the time I spent on “humanizing” AI in a spreadsheet to prove it was a waste of time
- The risks of publishing substandard artificial intelligence (AI) content are outlined, providing examples of brands that have seen their SERPs drop due to manual or algorithmic penalties.
One approach I haven’t considered, but you might want to try, is to direct stakeholders toward AI-related content that doesn’t require as much human input, such as:
- Glossaries
- Product descriptions
- Frequently asked questions
- Wiki
If your ultimate goal is to boost traffic, you may also consider proposing non-AI initiatives that have been proven to boost traffic at scale.
In Ahrefs’ latest study of the 50 bootstrapped SaaS companies that will dominate SEO in 2024, the brands that saw the biggest year-over-year traffic growth adopted three primary strategies:
- creating “free tools”
- program content
- content localization
Non-AI projects don’t have to be time-consuming either. ZapierFor example, we created thousands of feature-based landing pages programmatically and then outsourced their development to partners as part of the implementation process.
This type of traffic growth strategy is much more responsive to the needs of the client/manager who needs rapid growth, giving you time to focus on your job of creating high-quality content.
Final Thoughts
Humanizing AI-generated content is unsustainable and – contrary to what many people think – ineffective.
AI should be used for ideation, planning, and drafting from the outset, not for ghostwriting.
In words Senior Content Marketing Manager at Moz, Chima Mmeje:
AI content is not the problem. The problem is when the content is so poorly written that it is obvious that it was generated by AI. AI detectors are garbage. If you hire a writer and run the content through a detector as proof that it is engaging or authoritative, then you should not be working with writers. Writers, I implore you, stop using AI to create all your content. Please apply it as a guide to facilitate you develop ideas, complete your thought process, or even spotless up your writing, but that first draft should be 100% your work.
If you are reading this and have the ability to dictate strategy, abandon the soul-destroying process of humanizing AI content.
Trust your people to write content for people.
