How a picture of gold can seem worth more than the gold itself

What's the base value of content? It's a market subject to strange variations over which publishers seem to be ceding control.

by Rob Corbidge

Published: 11:39, 26 April 2024

Last updated: 23:15, 08 June 2024

I kick off today's edition with a huge thank you to the various CEOs and owners who joined us in London last night at a closed-door event for publishing leaders to talk freely about things like AI content theft and gen AI search targeting their traffic.

It was a free-flowing discussion with fantastic insight from all viewpoints, including from those negotiating with tech giants, insight from the US and Asian market perspectives, and from the legal community.

Of the many topics discussed was one which I have been pondering more than normal recently: what is content worth? And perhaps more interestingly, what is it worth to the AI companies?

One of the most valuable emails to regularly drop into my inbox contains links to a site which can be described as a sort of technical marvel. I say "sort of" in the same way Stonehenge is a sort of technical marvel.

It's a hyper-local news site, sustained by a combination of reader donations, a smidgeon of local advertising, and some reporting support from a UK scheme called the Local Democracy Reporting Service, funded by the BBC.

From a technical perspective, the best thing that can be said for it is that at some point it was crudely "optimised" for mobile. Other than that, it's generally a 2001-vintage web experience.

Being lucky enough to walk among the gilded halls of publishers with the funds and understanding to make sites of technical excellence, the contrast to be found on my valued hyper-local site is actually quite hard to square in my reporter's brain. But the cognitive gear-grinding is because - despite the tooth-cracking experience of the site - I find the content really useful and hence of value to me.

Everything from local crime reports, planning applications, open days at public organisations, local elections, clubs and societies in the area, and even wildlife features is covered on a shoestring budget that would leave most of us looking for better shoestrings.

In strict news terms, it's the lowest rung of original editorial content - by which I mean first, not low quality - but original editorial content being what it is it still costs money to produce.

How are we in such a period of confusion of value?

We hear daily of the gold-rush of activity around AI and the content which is needed to assist its learning, with truly vast sums of money flowing around it.

So what is the value of content to this AI Hydra?

A Reuters report from earlier this month has some interesting detail via Daniela Braga, CEO of Defined.ai, an org which aims to create the world's largest marketplace of ethically-sourced LLM training data.

Rates vary by buyer and content type, but Braga revealed AI companies are generally ponying up $1 to $2 per image, $2 to $4 per short-form video, and $100 to $300 per hour of longer films. Disappointingly, she revealed the market rate for text as a meagre $0.001 per word, shattering my justification for going any longer than strictly necessary.

The Reuters report is 1,631 words long. So, on those numbers it's worth a giddy $1.63 in the AI market. A total of five named editorial staff worked on the report, and $1.63 isn't gong to buy them a lot of hand-caught clams - especially at today's prices.

Obviously this is a simplification of the Reuters business model, and a simplification of what kind of money AI companies are willing to pay depending on the data they are getting, but it's illustrative of what value could be placed on something that took proper money to produce.

From the Reuters sharp end of the reporting spear, to the more gentle paddle around my local area, the common factor is value. Different value, but essentially made of the same stuff: time and effort and expertise

Nevertheless, confusion over the value of content is widespread in our currently challenged industry, and what would on the face of it appear to be a content seller's market is actually one driven by the buyers. 

They favour an argument around apples and oranges, while we we seem to be struggling to work out the average price of fruit, without coherent market data.