Online community is dying, and generative AI won’t save us

When I was in my last year of secondary school (in 2009, if you want to know my age), I found myself with a crisis on my hands. I had a cold sore on my lip, and as a French Horn player with a summer concert rapidly approaching, this was potentially disastrous. Desperately seeking advice, I threw myself on the mercy of a LiveJournal community for French Horn players.

Within a few hours, the replies started rolling in with thoughtful advice on how to deal with the sore and practice safely. Back in the late 00s (and earlier), these kind of niche hobbyist communities were the backbone of the internet – and provided you had internet access (which was a privilege that not everyone enjoyed), you could benefit from their expertise and experience on any question you might have, with nothing asked in return.

I bring this up because in 2023, a decade and a half on from my cold sore crisis, we’re looking at a starkly different internet landscape. Nothing illustrates this more effectively than the impact of the ‘Reddit blackout’ – in which thousands of Reddit communities, niche interest-driven spaces like the French Horn community that helped me, have set themselves to private in protest of proposed charges for Reddit’s API that will make many third-party apps and tools untenable.

Setting aside the wisdom of this decision from Reddit’s end, let’s look at the impact of the protest on the wider internet and search quality. Adding “Reddit” to the end of a search has long been a popular hack to get search results that contain considered, human responses to niche questions, rather than an endless parade of opportunistic content marketing. And yes, I’m aware that I’m saying this to an audience of marketers!

The response to the Reddit blackout, therefore, has been to despair of the low quality of Google search results without Reddit’s community-driven advice – but is the problem the results served, or the dearth of high-quality advice to populate them? As I reflected to my editor in the conversation that sparked off this piece, you would never get a piece of content marketing that satisfies my cold sore question, because there’s no commercial benefit to answering it. Even for a musical instrument retailer, there’s nothing for them to sell off the back of the question unless they have a sideline in topical anti-viral medications.

And yes, many businesses have blogs or articles that contain more generalist advice without a direct sell, but businesses have an incentive to target the search queries that are likely to bring in a lot of traffic; and if you’re going out of your way to target the long tail, you want to make sure there’s some likelihood of repeat visits or business. A one-off question about cold sores isn’t likely to bring that about.

As much as I enjoy writing about the subject of search and the art of SEO, the SEO mindset is to blame here – because it promotes the idea that the only type of content worth creating is the type that gets traffic, and ideally also business.

As for the community discussions that used to fill the gap, they’ve been vanishing in the age of social media centralisation, with sites like Twitter not really offering an easily findable (anyone tried using Twitter’s search function lately?) space for like-minded users to gather – unless they have existing connections there. Facebook groups and pages have suffered from the overall decline of Facebook as a platform – certainly for younger users – and they aren’t indexed by search anyway. One of the last bastions is Reddit.

So, what is a searcher to do? Google and Bing want us to start using large language model AI generators like Bard and Bing Chat for all our hyper-specific question needs. To see how it stacked up against the advice of a community of human beings, I decided to pose a version of my French Horn question to Bard.

‘AI-powered’ search is off to a problematic start. Can Google and Bing fix it?

To give credit where credit’s due, Bard’s response contained some good advice such as cleaning the mouthpiece to promote hygiene, and taking breaks to save my lip. However, other parts of its response were clearly woodwind advice (“Use a softer mouthpiece or reed” – no such thing for a brass instrument!) or were mostly helpful in the abstract, like treating the cold sore early – not much help if you haven’t caught it early but need to know how to reduce the pain in the immediate short term.

More importantly, though, an AI generator can’t replicate the – yes, it’s become a bit of a buzzphrase – human touch: a group of real people telling you, “Don’t panic, I’ve been there, here’s what to do”.

The fact is that when people have a question, whether it’s about a hobby or a DIY project or a recommendation for products, what they want is to hear from other human beings – not to get generic, traffic-targeted advice, or worse, an AI response trained to regurgitate that advice without any genuine understanding of the topic. Unfortunately, we’re moving away from that version of the internet, not towards it. Want to ask a question on Quora? Well, you can now pose it to an integration of ChatGPT – or you can pay to read a response written by a human. Even content marketing is now being authored by AI.

Well, at least we’ll always have Reddit – oh, wait.