Since its take off around 2006, social media has gone through several noticeable phases, and many of these relate to content creation. It’s worth running through these briefly to give context to how generative AI will affect social media.
Phase 1: Photo sharing and friend connection
The first phase, beginning with MySpace (founded in 2003) but then popularised through the general release of Facebook (in 2007), was connecting with friends largely through photo sharing and instant messaging.
Phase 2: Mobile social
The second phase was the proliferation of mobile accessibility and broadcasting, beginning around 2009, through new mobile first apps like Twitter and later Instagram, which at the time did very specific things.
Twitter was a broadcasting SMS service (with 140 character limits), while Instagram was for photo uploads. Things began to change with the rise of mobile first app Snapchat, which began to add in video messaging, stories and augmented reality.
Phase 3: Broadcast media
The third phase was social media as broadcast, beginning around 2015. By this time Facebook and Twitter had major influence on the news industry, and there were concerns that its algorithms, and their manipulation, had the potential to influence real world events.
Simultaneously, YouTube was rising as a monetizable broadcasting channel, and influencers on the platform, often combined with reach on Instagram, were growing in popularity.
From around 2018, YouTube began to eat into traditional television market share on new smart TVs, while TikTok presented a new form of vertical video broadcasting that got its uses hooked and propelled some to stardom. Facebook, Instagram and YouTube all made a play for vertical video.
What this brief history so far indicates is that social media is, by 2024, really quite different from its initial stated aims of ‘connecting people’.
Friendships tend to continue digitally via instant messaging Whatsapp over specific social networks, and increasingly the networks themselves have become broadcast mediums. YouTube has hoovered up a huge 25% of viewing on connected TV sets in the US, for example.
Meanwhile, does anyone use Instagram or TikTok for connecting with friends? Didn’t think so. They’ve become marketing machines for ‘personal’ brands where the key metrics are followers, likes and watch time.
Phase 4: Artificial Intelligence x Social Media
While we are going deeper into the phase of social media as broadcast, AI generated (sometimes labelled) ‘synthetic’ media is likely to have a marked effect, and possibly change the game again. But how exactly will it do this?
Firstly, generative AI effects a wide range of digital production processes. Images, audio and now video are all much easier to create through using generative AI.
If we cast back to 2018, getting editorially cleared images would likely cost around $5 each from a stock library, while now you can make hundreds of generations on whatever comes to your imagination via Midjourney.
To get a voiceover, you would either need to hire a voice over artist, or get a good microphone to record yourself. Both options are relatively expensive and time intensive when compared to using applications like Eleven Labs and Uber Duck.
The third part of this is that generative video is going to get a whole lot better over the next few years. What we saw released in 2024 via tools like Luma, Kling and Runway, is a fairly experimental race for users. It’s difficult to say that the outputs are currently commercially viable.
For example, you would not be able to put them in a feature film or broadcastable TV output yet. However, they are certainly viable for small screen usage – social media. It’s also quite clear, given the massive leaps we’ve seen in image generative AI, that we will reach 4k broadcastable generative video relatively soon – perhaps 12-24 months.
A related point to generative video is that the concept of digital twinning or AI avatars has not yet become normalised.
Many people are still quite weirded out by the idea of an AI representation of themselves doing their presenting for them (you can also combine this with a voice clone using Eleven Labs). But rather than saying this will never happen, we believe it will be increasingly normalised. This has the potential to effect online presentation in a variety of ways – to be explained further below.
The AI enhanced production process
There are really four forces converging on social content at once – the enhanced ability to create images, audio, video and even presenters, at marginal cost. A further major leap is presented by AI tools that enhance the production process.
For instance, generative AI enables eye tracking correction (both offered by Descript), green screen removal (Descript and Runway) or automatic silence and transcript based editing (AutoPod, Firecut and Descript). Further common enhancements come from using generative AI for subtitles and converting longer clips into shorter, as well as changing aspect ratios.
The proliferation of social content
All of these advances simply enable more people to produce video, which is increasingly what social platforms demand for followings to grow.
Again, if we cast our minds back to 2018, video production was far more difficult and generally required more highly skilled editors (often using After Effects) to do what a less skilled editor (for example me) can do with AI.
Added to this, phone cameras simply weren’t as good – but now I can get very crisp footage from an iPhone 15 Pro, and all smartphones down to around the generation 12 are capable against DSLRs.
What this means is there will be far more competition as the playing field inevitably levels. A person armed with a few AI tools and the right knowledge could be as productive as a 3-4 person production team in 2020, considering tool usage through the process. For that, see our article the 10 Best Tools for Video Editors, but also add in ChatGPT (or another LLM) for research, ideation and writing.
The rise of the faceless channel
Used correctly, the use of AI tools enables the creation of storytelling that wouldn’t otherwise be possible for many, simply because of the time and opportunity costs. Many faceless YouTube channels have sprung up, as well as plenty of longform, and some of them are clearly partially assisted, if not majority assisted by AI.
The reality is creating objective content that sustains viewer interest, when it is majority AI assisted, is very difficult. YouTube is increasingly gearing to prioritising higher watch times (and thus length of content) as it eats into TV market share. Many AI first channels will be opting for shorter content (via TikTok, Instagram or YouTube shorts) to build an audience.
However, while the opportunities to build large but low value audiences are currently significant, the ability to transition such audiences (or followings) into commercially viable concerns is difficult, and I think unlikely in the long term.
There is certainly a lot of hype around building faceless channels and gaining tens of thousands of followers. Unfortunately, in a world of significantly increased competition, followings of hundreds of thousands no longer represent a liveable income.
‘Making hay while the sun shines,’ is how one (very significant) social media expert defined the issue to me on a recent call. I agree in an audience sense, but in a revenue sense the sun isn’t shining now – and the audience might will likely be eclipsed soon.
The AI labelling issue
The key issue for channels majority assisted by AI is that YouTube and Meta both require the channel/creator to label their content as AI generated. If they don’t there is the possibility of algorithmic penalty. This was done relatively haphazardly by social platforms due to the 2024 US Presidential Election, but its policing will improve over time.
Social networks, particularly Facebook, experienced a blizzard of negative PR after the 2016 election and the coronavirus crisis due to the proliferation of fake news. AI content has a high probability of being objectively incorrect, and, worse still manipulated by bad actors to influence people. Thus logically, a social network should depriortise it as such. Their reputation is significantly on the line.
Thus ‘making hay while the sun shines’ may only be a 2024 phenonemon even in an audience sense. Social media is now a mature industry, under significant threat of strong regulation, and huge players like Meta and Google won’t want to take an unnecessary risk once they can quantify it. The US election is the opportunity for them to do so, and in the aftermath I anticipate AI labelled content to be downgraded and more closely policed.
The blended personality approach
Most social media broadcasting (or publishing) is not AI first or faceless. The most successful channels are personality led – the influencer or creator – such as Mr Beast. Adam Mosseri, the CEO of Instagram, has explicitly said that Instagram will favour individual creators over publishing brands.
Faceless channels and particularly AI generated ones, fall under this category, but have further problems of little brand equity or AI labelling. My prediction, related to the question of authorship on Google, is that the identifiable individual, or expert human, will have to be prioritised for the good of the platform overall. The CEO of Instagram appears to agree.
Of course, Mr Beast’s output is not completely ‘real’. His thumbnails go through a significant amount of doctoring and split testing via a significant supporting creative team, often to the point where only his face is the only ‘real’ element within them. With AI, the costs of this output is markedly lower.
As noted though, new generative AI tools enable an enhancement of the the video production process. Individual creators that embrace these tools are simply going to be more competitive, in a world of increased competition.
The biggest time consumption (and thus cost) of the creator production process is generally in the editing and distribution end. When these are alleviated by new tools, new talent has more time to shine.
Are digital avatars faceless?
An interesting unknown in this equation is just how far AI avatars will be embraced or can go within the next two years.
Both TikTok and Meta have embraced the ‘digital twinning’ of some degree, without a significant impact. But these experiments leave us with some intriguing questions, bordering on the philosophical representation of the self:
- What happens when the original creator cannot be defined against the digital representation of themselves?
- Must the creator label that representation of AI, even though that would be themselves in reality?
- Does a creator thus manage ‘faceless’ channels, where they are represented by digital equivalents of themselves at far lower production costs?
This is a quandary where social media becomes more meta than itself (or indeed Meta’s name). We thus enter a metaverse. And perhaps I should stop writing now, before I disappear up my own arse.
