If you haven’t used ChatGPT’s Deep Research tool yet, get ready to have your hair blown back, and then set on fire.

On Saturday morning, I asked the tool to do some research for an article I wrote on the benefits of private debt for institutional investors. I gave it an earlier draft, then clicked “Send.” In 13 minutes, it cheerily produced a 3,000-word research paper, complete with copious sources, tables, pull quotes, and a full bibliography. It cited only reputable sources like PitchBook, Deloitte, and McKinsey.

I copy-pasted that paper into a Google Doc, downloaded it, and fed it into a custom GPT designed to write articles for the client, complete with their brand guidelines and corporate details. In about two minutes it created a 1,200-word blog post that, for all intents and purposes, appeared to outperform anything a human writer could do without spending two days on craft, and invoicing over $1,000.

But there was one glaring problem.

The research was a joke.

The Deep Research AI gave me copy like:

According to CIO Nicole Musicco, the move was driven by the ability to deliver “steady distributions, real downside protection, and better relative value than equity.” The company reallocated billions and sent a clear signal that institutional portfolios are rethinking how they can drive innovation and income at the same time.

That’s not bad. If a human writer gave me that, I wouldn’t blink. But when I asked for a source link, the custom GPT did a little finger pointing to the research document. And the sources referenced by the research document didn’t contain any such quote. Nor did I find it anywhere online.

The same thing happened with four other direct quotes and exact statistics in the same article section. When I asked ChatGPT what happened, it apologized and hedged. In other words, ChatGPTs vaunted deep research tool had made up its quotes and data.

Used by a trusting editor, AI has the potential to destroy your brand credibility overnight. In 2024, a reporter from Cody Enterprise in Wyoming resigned after he used AI to fabricate quotes and details in his stories. His deception was outed, and he admitted to using AI. But he — and his paper — didn’t know his article had fake news in it. Their only real crime was trust.

Thankfully, working with the GPT and clarifying my needs, I was able to retool my own article with verified sources and accurate pull quotes and data. And it didn’t take two days, or even half a day. It was more like half an hour.

There are two key takeaways here.

First, for content managers, marketing managers, and editors: Vanilla AI almost always hides a poison pill. You can’t, can’t, can not trust that what it gives you is true. You must put an expert writer in the process. Someone who knows how to find and verify information from respected sources and how to weave it into a compelling narrative in the right voice, tone, and engagement level. That person needs to be there to guide the AI and tell it where and when it wanders off the rails, and show it what to do instead.

Second, if you’re a writer who is horrified (like most writers I see online today) by the rise of AI and the death of professional writing, you have a clear and exciting way forward. I think it’s more than a little glib and insensitive to say you can get a job as a “prompt writer.” I think that cheapens the years (and possibly decades) of work you’ve put into learning your craft.

I also think there’s a real danger of experienced writers deciding their skills have been rendered null and void. That’s anything but true.

The real path forward is to be the expert guiding the machine at every step along the way. In my view, the way to think of it is that you likely won’t command the same rate you used to for a 1,500-word article, but you might not need to.

Writers — and especially freelance writers — have always targeted a certain hourly rate. I believe that by working as an expert human in the loop, you can still earn that rate, guiding AI tools to write high-quality copy, and potentially higher-quality copy, faster than was possible before AI came along. In other words, writers are increasingly wary of AI taking their jobs, and publishers should certainly be excited by the efficiencies it offers. But both parties should fight hard to put a human expert in the driver’s seat.

How are you using AI to enhance your content creation, and what role do you see for human experts?

Can you really check for AI-generated content?

No. But you can check for bad writing.

When AI hit the scene in 2023, nobody cared whether a blog was written by AI. If it was cheaper and it got the job done, where was the harm?

Then the questions started popping up. Was it ethical? Would it steal bread from the mouths of human writers and their families? Was it plagiarism? Was it accurate? Would Google penalize it?

It’s still a bit too early to tell if AI is taking a concrete crusher to the job of content writing. In a survey by content guru Jennifer Goforth Gregory, most writers said they’re earning more today than ever. And most are using AI tools to get more work done. Yet on LinkedIn, you’ll see story after story about writers losing their careers to AI. It does feel a lot like a pandemic.

Next, originality. Yes, AI-written content is plagiarism — or not — depending who you ask. One study found AI can “reuse words, sentences, and even core ideas.” Then again, “Since chatbots generate new text ... GAI could be considered more like ghost writing than plagiarism.” The bottom line is, AI learned to write by analyzing work by human writers. To which AI proponents answer, “So did you.”

What about accuracy? AI has a spotty record here. It can craft precise responses just as often as it can hallucinate. Most AI tools now come with a disclaimer like, “ChatGPT can make mistakes. Check important info”

As to whether Google penalizes AI content — in a way. Google’s guidelines greenlight AI-generated copy. But it has to be “good.” In other words, it has to demonstrate expertise, experience, authority, and trust (E-E-A-T).

Therein lies the tale.

According to Google, the question isn’t whether an article was written by AI. The search giant says 10 years ago there was a glut of ghastly human-generated content. Their job back then was to separate the literary wheat from the electronic chaff. That’s still their job today.

And that’s where AI checkers come in.

Most content strategists will tell you AI checkers plain don’t work. For instance, here’s a screenshot from an online tool that flags a healthcare article as AI-generated.

The problem? The article was written by a human in 2021, when ChatGPT was just a gleam in OpenAI’s eye.

Google, “How do AI checkers work?” and you’ll find they’re AI-powered tools themselves. They compare massive numbers of AI-written samples to human-written samples. Through this training, they learn to “know it when they see it.”

So — what’s different about AI-generated text?

It’s generic.

The problem is the way AI is trained. Those millions of samples are kind of like looking at a million dogs, then sitting down with a charcoal pencil and a sheet of foolscap and drawing one generic Fido.

See the problem?

That’s not how humans work. They’re quirky. Obstinate. They draw on their unique experience. They don’t aggregate and genericize. Creative minds rebel against the commonplace. They learn the laws so they can break them. Twist them. Make them fresh.

It’s the very “sameness” of machine-made content that AI checkers learn to spot. They flag cookie-cutter writing that uses the same words, phrases, and cliches that “everybody else” does.

Put another way, AI checkers don’t call out the original or the iconoclastic. Content that bucks the status quo slips through their digital fingers.

That’s why, if you run 100 blog posts from 2020 through an AI checker, most will trip the AI-generated trigger. It’s not because some early AI model wrote them, but because they’re stiff and unoriginal. Google didn’t rank that kind of writing then, by humans, and it won’t rank it now, by AI writing tools.

So, yes, AI checkers work. But they don’t spot AI-generated content. They spot bad writing.

The solution is to put a human expert in the loop. Someone who can “twist the straight line,” to coin a phrase from comedy writing. Someone with a unique background, full of all the foibles of a human life. Someone who sees the world as no one else can. A living, breathing writer who doesn’t approximate and reduce, but who works to occupy the interesting edges around the fractal fringe of an idea. Someone who can show us something new. It’s that kind of writing that engages audiences. So far, only human writers know the way to make it tick.

Businesses hit AI like a tarpon on a jerkbait, as more than half of all companies struck the flashing lure of AI-generated content. AI use surged 47% to $66 billion. ChatGPT became the fastest-growing app in history. And 56% of users in a blind study said they liked AI-written articles better than their human-written counterparts.

Then flesh-and-blood reality came calling.

CNET was publicly lambasted for using AI to write a stream of plagiarism. The site lost credibility and sold for half its previous worth. Consumers shied away from openly AI-assisted publications, lamenting the loss of the human touch. Quality and accuracy issues reared their carping heads, and the term “AI slop” crashed the lexicon. Businesses have since called a collective takeback, fleeing AI-generated content use in droves. The flood tide of AI confidence has been followed by a matching ebb of digital regret.

So — is the party over for AI-written content?

Enter HITL — Human in the Loop.

With a human in the process, AI takes its rightful place in content creation — as a tool. Rather than leading the charge, it follows the direction of seasoned human content pros. In the same way that a hammer and a skilsaw can’t build a house without a contractor to wield them, a GPT can’t write an engaging thought-leadership article on risk management, or a traffic-getting blog on how to manage diabetes, without human help.

Only a human can make articles that resonate with humans. Why? Because only humans know what matters most to *us*.

What AI *can* do is make human writers faster and more effective. It can do lightning-fast research on the pain points around homeownership or retirement or PPOs, helping the writer to build empathy for the reader in minutes instead of hours. It can find and verify better sources.

Write suggested outlines. Draft paragraphs studded with the right keywords, worked into the text in natural ways. Verify accuracy. Call out omissions, missed opportunities, and awkward phrasings. And guided correctly, it can craft human-sounding content that engages us.

In other words, AI can help human writers do what they do better.

The rush to AI content as a low-cost panacea was a misstep. Today’s top businesses are course-correcting, putting human experts in the loop, and reaping the rewards in greater monthly traffic and conversions. With AI tools, living, breathing, feeling writers are creating written work with higher levels of accuracy and creativity, faster, at a lower cost. They’re reaching their potential in a way they never could before AI.

That’s why HITL isn’t just a new buzzword in AI-generated content. It’s the way forward.