AI wins 6 out of 6 copywriting battles, as judged by 700 humans

https://searchengineland.com/human-vs-ai-generated-content-survey-437062

11 points · 4 comments · view on lemmy.world

4 Comments

nyan@lemmy.cafe · 20 pts · 2y (2 replies)

Advertising copy is likely overrepresented in the general corpus of texts from the Internet that most LLMs are trained on. Plus, it isn't a genre where truth matters all that much. It's intentionally vague and cliché-riddled even when humans are writing it. So it's something that I'd expect LLMs to be pretty good at creating.

"Garbage in, garbage out" is just fine if garbage was your desired output to begin with.

bionicjoey@lemmy.ca · 12 pts · 2y (1 reply)

If you asked me "what job is most likely to be replaced by a computer program which generates plausible and convincing text and media, regardless of their agreement with reality", my number 1 answer probably would be "marketing".

funkless_eck@sh.itjust.works · 2 pts · 2y

Copywriting is a mostly junior / entry level position and it's only a benefit to most marketing teams to automate the busy work.

There are a lot of processes in Marketing that you'd think you'd want to automate until you're faced with an irritated SDR team who are wasting their time calling people that don't exist, or aren't even MQL, let alone ready to talk to sales

lobelia581@lemmy.dbzer0.com · 6 pts · 2y

That's a very misleading title, since "battles" here is referring to survey results and not actual legal battles. The results make sense though. AI will probably generate the most popular kind of post because that's what has the most representation in its training data.

The main issue here is what is "popular" changes over time, and is directly related to what is available to the public. So if AI floods the internet with the same style of posts because it's currently the most popular, that style will quickly become boring, and using AI to get clicks will essentially lead to it writing itself into obsolescence. Until it gets trained or fine tuned on a new dataset which includes its own results, which leads to a separate issue where the training data is objectively bad.