← All insights

Average, at scale

Producing a social post used to cost something. An hour, an idea, a small argument about whether it was any good. That cost is now roughly zero, and companies have done what anyone does when a thing becomes free: had far more of it. Feeds are fuller than they’ve ever been and harder to tell apart than they’ve ever been. This month TikTok banned AI-generated voices from shopping livestreams, and it’s worth asking why the platform where selling is most direct was the first to draw a line.

The constraint was never volume

Companies whose social media produces nothing almost always diagnose it as an output problem. Not posting enough. Not on enough platforms. Not fast enough. So when AI turned up offering unlimited output at no cost, it got adopted at speed to fix a problem most of them didn’t have.

If four posts a week produced no enquiries, twelve will produce no enquiries faster. Volume was never what stood between a feed and a pipeline. Being worth reading was. The companies now publishing every day and still waiting for the phone to ring have just industrialised the thing that wasn’t working.

A prompt is an averaging machine

Here’s the mechanical problem, and it isn’t a matter of taste. A language model is built to produce the most probable next sentence. Probable means common. Common means what a great many people have already written. Ask it for a post about your category and you’ll get something competent, grammatical, faintly warm, and publishable by any of your competitors without changing a word. Which is a strange thing to pay for.

Marketing has known for a long time that brands aren’t chosen because they argued better. They’re chosen because they came to mind, and they came to mind because they were recognisable: a voice, a look, a repeated way of seeing things. That recognisability is the asset doing the work. Being well-written isn’t. Generated content hands you the fluency and quietly removes the recognition, and it’s the second half that costs you.

The platform that sells hardest drew the line first

TikTok’s rule change is the tell. Shopping livestreams can no longer use AI-generated voices, recorded narration or radio-style audio. Sellers have to speak to the room in real time, and violations now feed the platform’s new Account Health Rating, the score governing access to campaigns and commission programmes. This isn’t an ethics position. It’s a commercial one. In the moment money changes hands, a synthetic voice tells the buyer that nobody could be bothered to turn up, and that doesn’t sell.

The audience got there first anyway. The share of people who’d rather see AI-generated creator content than the human kind has fallen from 60% in 2023 to 26%. Nobody’s objecting to the technology. They’re objecting to being sold to by something that isn’t there.

Automate the plumbing, not the point of view

None of this is an argument for doing everything by hand out of principle. AI is genuinely good at the unglamorous middle of the job: sorting research, pulling the three usable minutes out of an hour-long client call, reshaping one idea for four platforms, drafting the eleventh version of a caption when the first ten were wrong. Put it there and it buys back the hours the real work needs.

Put it on the thinking and you’ve automated the only part that was doing anything. The opinion. The proof. The number from a real project. None of that can come out of a prompt, because none of it is in the training data. It’s sitting in your business, and getting it out is the actual job.

The test

Take your last ten posts. Cover the logo. Hand them to someone who knows your category and ask which company published them. If they can’t tell, you’ve spent a quarter paying to produce content that would serve your competitors just as well. That isn’t an efficiency gain. It’s a slow, well-organised disappearance.

The fix isn’t more output, and it certainly isn’t less software. It’s a channel built on a position worth holding, assets that make you recognisable in a scroll, proof instead of adjectives, and every post doing one defined job. That’s the whole of the Social Strategy Sprint, and the reason it survives contact with a feed where everything else now sounds the same.

Quick answers.

Should I use AI to write my social media posts?

Use it on the machinery, not the thinking. Sorting research, pulling usable quotes out of a call, reshaping one idea for four platforms: all fair game. The opinion, the proof and the specific numbers have to come from you, because they aren’t in the training data. They’re in your business.

Does AI-generated content hurt my reach?

The bigger risk isn’t reach, it’s recognition. Generated content reads as fluent and anonymous, so it gets scrolled past rather than penalised. Platforms are starting to draw harder lines too: TikTok now bans AI voices in shopping livestreams outright.

Why isn’t posting more often producing more enquiries?

Because volume was never the constraint. If four posts a week produced nothing, twelve will produce nothing faster. Enquiries come from a recognisable point of view, visible proof, and a clear path from post to conversation, not from output.