AI doesn’t laugh – and other reasons why great content needs humans

Just to be clear from the outset, I’m not here to tell you AI has no place in content creation. Every day, I work with smart marketers and technologists at companies that are expanding the boundaries of AI’s usefulness. But while they rightly evangelise about the amazing things AI can do, they’re very clear about what it can’t – or shouldn’t – do.

Generative AI, for example, is excellent at analysing large amounts of data, identifying patterns, and predicting what comes next in a sequence. That makes it great at holding natural-feeling conversations with human users, summarising unstructured data like writing, and producing credible-looking writing of its own. However, like so much in life, there’s a trade-off.

Productivity vs effectiveness

This year’s CMI B2B Content & Marketing Trends survey found that the vast majority (89%) of B2B marketers are using AI for some aspect of content creation. Of that group, 87% have seen productivity increase, which is great. But 12% say content quality has decreased as a result of using AI, which isn’t so great.

Interestingly, 40% of respondents say they still struggle with ‘Creating content that prompts desired action’. And when marketers focus on increasing outputs (making lots of content) rather than improving outcomes (making content that gets results), that challenge will remain, no matter how much AI they throw at it.

AI can be an important part of the solution to this problem, but it’s unlikely to move the content effectiveness needle on its own, for a few important reasons.

AI doesn’t laugh

AI knows the setup-punchline structure of a joke. It’s read every joke ever written, and it knows which words are most statistically likely to appear next in that structure. But it doesn’t understand why something is funny.

Humour is intrinsically tied to shared human experience and emotions. It’s the reason watching sitcoms alone is no fun, and why we can watch our favourites again and again and still laugh out loud.

Now obviously, humour isn’t especially relevant in a white paper about headless CMS integration challenges, but emotion is very relevant. Great writing, even on bland topics, creates some sort of emotional connection between the reader and the author. That requires empathy to show (even if it’s hiding in the subtext) you have a shared understanding of why the CMS integration challenge is so difficult and the emotions attached to that experience.

A statistical algorithm can’t empathise with the human experience, so it can’t form an empathetic connection with the reader. That might be fine for a run-of-the-mill SEO blog that just needs to exist to generate traffic, but it’s unlikely to be effective for thought leadership content that needs to get engagement.

AI doesn’t understand what it writes

The label ‘artificial intelligence’ is quite misleading, especially in the case of large language models (LLMs), as there’s no meaningful ‘intelligence’ at play at all. There’s a lot of data crunching and statistical modelling at orders of magnitude beyond what a human could do, but there’s no real thinking.

AI can make highly educated guesses that certain words should appear in a certain order in response to a prompt, but it doesn’t understand the words’ meaning. Because AI outputs are a string of ‘tokens’ – connected by their likelihood to appear together rather than their meaning, context or impact – LLMs tend to produce facsimiles of meaning. Facsimiles that get fainter each time they’re reproduced.

And so, there’s always a danger that AI-generated content will substitute familiar tropes, hackneyed phrasing, and superficial readability for real substance, critical analysis, and original thought. But these qualities are vital for creating credible, engaging and, most importantly, compelling content.

At a macro level, as AI-generated content proliferates, there’s also the danger of model collapse, where each generation of LLMs learns from the outputs of previous generations, so errors are amplified over time until the outputs are too unreliable to be useful.

This isn’t to say that AI can’t be a helpful writing companion for marketers. Many find it incredibly useful for proposing initial ideas or approaches and generating early drafts for more straightforward assets. It’s just worth keeping in mind that AI doesn’t understand what it’s saying, so it might not always say the things that will engage the right audience and compel them to take action.

AI has no skin in the game

If there’s nothing at stake, then there’s no incentive to produce accurate, authoritative content that generates meaningful engagement. For marketers, there’s plenty at stake: brand equity, product awareness, lead generation, ROI, personal reputation…

But AI has no horse in this race; its only ‘incentive’ is that individual users respond positively to its outputs. This is a particular issue for LLMs, which are mostly trained through reinforcement learning based on user feedback. And as researchers at Anthropic established a couple of years ago, this leads to a phenomenon called AI sycophancy, where models will produce outputs that users like, regardless of whether those outputs are factually correct or ethically sound.

With no consequences for its actions, AI doesn’t need to check its sources, or even have any sources in the first place. This was highlighted recently when Deloitte agreed to pay the Australian government a partial refund on a report that contained AI-generated errors, including references to non-existent research and a fabricated ‘quote’ from a court judgment.

As I said earlier, AI can be a very useful tool for marketers, helping accelerate various elements of the content creation process. But it’s important to use it for what it’s good at and not rely on it for things it struggles to comprehend.

Great content – the assets and campaigns that create connections and start valuable conversations – demands human empathy, critical thinking, and fresh perspectives. And the best way to get that is to fill your team with people who challenge assumptions based on their experiences and expertise.

Oh, and you should probably find a specialist copywriting partner who will do the same, and put their skin in the game alongside you.