AI-Generated Content: When Credibility Gets Lost in Translation

Let’s talk about AI. We’ll start by looking at a pair of stats.

69% of people trust AI-generated content less than human-generated content.

41% of LinkedIn posts that are over 250 words are generated by AI.

Taken together, those numbers don’t bode well for the future of the platform. Especially if they keep going up, which seems inevitable. Of course, LinkedIn knows this. And if things continue moving in the current direction, there’s a good chance it won’t be LinkedIn that gets burned, but those who’ve been posting LLM-generated content. But we’ll come back to that. First…

The case for AI-generated content.

I get it. I do. Writing well comes naturally to a relatively small number of people. For most, it takes a lot of work and practice. Hell, I consider myself to have some modest amount of talent for it, and it still takes a lot of work and practice.

A lot of people would love to share their opinions or expertise with colleagues, customers or the world at large, but have held back because they’re not confident in their writing ability. Now we have a tool where anyone can put their thoughts down with little regard to readability and magically have them translated into what appears to be well-written prose. Of course a ton of people jump at that opportunity. Like I said, I get it. This blog is a warning, not a judgment.

Here’s the thing about letting AI translate your thoughts into words...

Using AI-generated content risks your credibility.

Consider three people who all want to post the same basic message that demonstrates their status as a subject matter expert. None of them are exceptional writers.

Albert writes his own post. It takes him a fair amount of time and has some grammar and spelling errors.

Bill works with a professional writer. He provides his expertise via an interview, the writer drafts a well-written version of the relevant concept, and then Bill reviews and edits it before posting.

Carl types his thoughts into an LLM and asks it to make them sound good. He reviews the output for accuracy and posts it.

Now consider what an audience thinks of all three posts, assuming they know how each was created, but don’t know anything about Albert’s, Bill’s and Carl’s actual levels of expertise.

People may have a bit of trouble reading Albert’s post, but his expertise is seen as genuine and authentic.

Bill may have used a professional writer to achieve a higher level of polish, but the expertise in his post had to come from somewhere, and it’s much more likely to have originated with him than the writer. People get that.

Unfortunately for Carl, no one has any way to estimate his level of expertise. If LLMs were merely ‘writing optimization’ tools, he’d be ok. But they’re not. While Carl entered a long prompt that clearly demonstrated his expertise, another user might simply prompt, “write a 300-word post about (insert subject)”. The results someone gets from such a prompt can bear a remarkable resemblance to the results Carl got.

To add insult to injury, because of the way LLMs work, most of the originality in Carl’s viewpoint likely got sanded off, as LLM models are built off and trend towards recognized consensus. People close to Carl probably give him the benefit of the doubt, but for those he’s likely trying to reach? Those who don’t know him that well or at all? He’s earned little to no credibility with that audience. And for some, he’s generated outright skepticism.

Now’s the point where all the Carls reading this think, ‘Well, no one can tell that I’m using AI…”

People know you’re using AI.

Ok, maybe some don’t, but most do. Or they at least suspect it. Especially if your social media activity has experienced a significant uptick that’s coincidentally coincided with the rise of LLMs.

Also, there are already exceptional tools for identifying content written by AI. Pangram has a false positive rate of just 0.01%. In other words, when that tool analyzes 10,000 pieces of content written by humans, it will only mistakenly identify 1 of them as being written by AI.

We’re likely at the point where some Carls are thinking they’re immune to all this, that people in their audience won’t instantly catch their use of AI and are unlikely to use something like Pangram to check it…

Social platforms will integrate something like Pangram or die.

Let’s circle back to what I said about LinkedIn not being the one that’s going to get burned here. As platforms get flooded with AI-generated content and more people view such content as untrustworthy, inauthentic or useless, the platforms will face a choice. They can either lose their user base and become irrelevant or they can incorporate AI-detection tools. Does anyone believe any tech company will choose the former?

Whenever this happens, the tools likely won’t differentiate between new and old content. One day the Carls may wake up and find all of their LLM-generated posts hidden or deleted. Or they might just all carry a disclaimer that they were generated by AI.

Now, like I said back near the beginning…

This blog is a warning, not a judgment.

You may read the above and just not care. Maybe LLM-generated content requires so little of your effort that you figure there’s no potential downside. Maybe you recognize the risk and find it acceptable. Or maybe you need a social presence for your brand and simply don’t have the budget to hire or contract with a professional.

It may well be that AI-generated content yields you a benefit in the short term. Most things in business and in life involve some level of trade off. Nothing is perfect and few things are perfectly awful. It’s important to consider both the benefits and drawbacks of any tool, and my hope is this blog opens the eyes of those who have focused solely on the former when it comes to using LLMs to do their writing.

A quick note to wrap up.

If you came to this page from somewhere other than LinkedIn, you may have missed the announcement that Fact House has officially closed its doors. If you’re looking for top-tier B2B (or should I say B4B) marketing services, I hope you’ll reach out to me at my new/old home, dgs Marketing Engineers.