AI is making everything sound the same, so how can brands stand out?

I asked ChatGPT to come up with a powerful first line for this article.

It suggested this:

It’s becoming harder to tell whether something was written by a person—or assembled by a system trained on everything people have already said.

Aside from the fact it uses the infamous AI em-dash, it’s ok, if not exactly punchy. And to be fair to ChatGPT, a human might also have come up with it, though probably not a Brit, given the em-dash. And it does sum up the problem. Large language models (LLMs) have learned to imitate humans by learning from “everything people have already said” – except it isn’t everything.

Currently, LLMs are mostly learning from what’s already out there in the public domain. As a result, most AI-written copy is boringly generic and includes a generous amount of marketing jargon. This is partly our own fault. AI has only learned phrases like “ever-evolving landscape” and “drive meaningful results” because someone has written them first. The B2B sector in particular has some explaining to do.

Big brands spend gazillions deciding which cool friend/mother figure/potential lover that they want their drink/baking product/cat food to represent. Tone of voice matters because it makes the audience feel as though they are hearing from a real person. So how can smaller brands – even those in B2B – achieve the same effect?

Some AI evangelists predict that LLMs will be able to mimic individual humans accurately before too long, so we’ll just be able to issue a few instructions and AI will do the rest. Maybe they are right. To do this, however, AI will need to learn from a much wider, high quality base of data. The big tech companies are doing their best to help themselves to this but an increasing number of writers are objecting to having their work stolen in order to train a computer. Personally I like the theory that with so much AI-generated copy “flooding the internet” (to borrow a phrase from Martin Sorrell) LLMs will continue to learn mainly from themselves and will eventually implode.

"Large language models (LLMs) have learned to imitate humans by learning from “everything people have already said” - except it isn’t everything."

Xanthe Vaughan Williams director, Fourth Day

Brands, B2B included, need to hold on to the fact that they are talking to humans and that personality matters. Some brands are brilliant at this. Nobody thinks that Compare the Market is run by meerkats, but they can recognise human characteristics in the anthropomorphised animals. Those characteristics are then attributed to the brand as a whole.

For B2B tech brands, heavily accented furry animals may not be the right choice to deliver messages. But if we’re expecting anyone to understand our words, we must make it clear that people, rather than AI, are behind them. Even if it’s a reassuringly unglamorous expert on the processes involved in insurance. There’s a reason why on LinkedIn, the least exciting of all social media platforms, individual videos do well. It’s because they show real people talking. (Except when it’s an AI deepfake of course, but that’s misinformation, which is a different blog.) In either case it’s more engaging than a post saying that someone is very excited to be at Excel, which somehow never quite resonates with me.

AI does pose a real danger of reducing all B2B copy to generic mush, but it’s a danger we were already facing. I’m not the first to say this, but B2B brands need to be bolder. Even a by-lined article designed for a trade magazine can risk showing a bit of personality. Maybe one day AI will be able to interview a technical expert and produce an article that brilliantly captures their tone of voice and engages readers worldwide, but it certainly isn’t there yet. In the meantime, we B2B folk need to remember that “a business audience” is made up of individual people.