Synthetic content temptations

Increasingly, the cost of machine-generated content drops ever closer to zero…

“Gemini, using our bespoke GPT write 750 words on how our reliance on China for manufactured good has exported deflation to the west for three decades”

Whether your trained GPT model has a valid opinion or not, LLMs churn out thousands and millions and billions of words on this or any other subject making “niche” entirely obsolete.

If you hold the keys to the entrance of the web, then this is a problem.

See, if you are Alphabet or Meta or even ByteDance then you act as the doorman to the interwebs and your value is highly correlated to the increasingly specificity yet decreasing quality of User Generated Content (UGC).

For good, or for evil

Used correctly and with skill, LLM can assist in the production of useful content. Used maliciously, it can produce derivative drivel that clutters up the web. How much of the content on the web today is junk? 50%? 75%? 95%?

So, until the day frontier-model LLMs produce content genuinely indistinguishable from human-written content, the gatekeepers are likely to blacklist synthetic content – especially content which is designed to game their attention/ad-display algorithms.

For example, how would you even apply Google’s Quality Guidelines around author expertise and experience when scoring LLM-derived content? Which begs the question … “if you rely on organic SERPS, would you run the risk of an organic position penalty because of the use of LLMs?”

There is a day, not very far in the future, when trained-GPTs with memories and improved reasoning will create, with guidance, genuinely original content – until that day, we have to expect to still mostly hand carve content if we want to be taken seriously by the gatekeepers.

Using LLMs to amplify and assist

For all the doom and gloom, a well-trained GPT will amplify human creativity by generating diverse ideas, providing unique perspectives, and suggesting innovative phrases. This expands the writer’s creative toolkit, fostering more dynamic and engaging content.

Good models also streamline the writing process by automating routine tasks like grammar checks, content summarisation, and research assistance. As well as tailoring suggestions to individual styles and needs. This allows writers to focus more on high-level thinking and creative aspects of their work.

Using generic GPT to increase the low quality content clutter on the interwebs? Don’t be surprised if one or more of the gatekeepers stops you coming in at all. That is a whole different kind of FOMO and a whole heap of pain.