AI has made targeting smarter, production cheaper and content easier to create. That sounds like progress. But when everyone has access to the same machinery, the machinery stops being an advantage. That shifts to something much harder to automate: having a distinctive idea that people actually care about.
What marketers used to fight over has quietly become marketing infrastructure as we enter autumn/fall 2026. The audience segmentation, the retargeting logic, the endless testing of creative variants against slightly different cohorts: every one of those capabilities is now available, cheaply and automatically, to a competitor with a fraction of your budget. AI has started to do the same thing to basic production. A simple brand film that once needed a shoot, an edit suite and a fortnight now needs a prompt and sometimes only an afternoon.
This isn't a prediction about where the industry is heading. It has already happened, and the consequence is uncomfortable for anyone who built a career on the machinery rather than the message: the machinery no longer differentiates anybody.
Meta’s Andromeda is evidence of this shift. When Meta’s engineering team detailed the system in December 2024, what they described was a transformation of the retrieval stage of the advertising system: much more sophisticated machine learning narrowing tens of millions of eligible ads to the few thousand most relevant candidates for an individual, with later ranking models then determining what actually gets shown. The detail that matters is what it moved away from.
Meta’s own account contrasts Andromeda with “conventional retrieval models that rely on expert-engineered features”, in other words, with systems that depended on humans specifying in advance what signals mattered. In November 2025 they went further, publishing details of GEM, a foundation model it describes as the largest of its kind for recommendation systems, “trained at the scale of large language models” and, in Meta’s words, “trained on ad content and user engagement data from both ads and organic interactions.”
GEM doesn’t serve ads itself; it’s a teacher, distilling what it learns into the hundreds of smaller production models that do. Neither system replaced targeting. What both illustrate is a steady, deliberate migration of judgement away from the person configuring the campaign and into the platform.
For a long time, marketing discipline revolved around targeting. Decks were built on personas, interests layered on top of behaviours, audiences narrowed until they felt reassuringly precise. There was comfort in the settings panel: if performance dipped, you adjusted the audience and told yourself you were optimising.
So, that precision is now largely a commodity if everyone can press the same buttons, and the platforms are increasingly capable of inferring relevance from behavioural signal rather than waiting to be told who to look for. The practical question has shifted from “who do you want to reach” to “what is most relevant to this person right now”, and that shift moves the centre of gravity away from the settings panel and onto the thing being shown.
When the infrastructure is the same for everyone, what used to be a specialist skill becomes a baseline expectation, and the competition moves upstream to whatever the infrastructure can’t manufacture on your behalf: the idea itself.
Organic social has always worked this way, incidentally. The algorithm never cared about the audience you intended, only the reaction you earned.
Paid media is gradually catching up to a logic that content creators have lived inside for a decade.
What makes the current moment genuinely different from the one the industry was navigating two years ago is that AI is pervasive. Research from IAB and Sonata Insights found that 83% of ad executives now say their company has deployed AI somewhere in the creative process, up from 60% in the equivalent 2024 study. A single marketer with the right tools can now produce in an afternoon what used to require a production company and a six-figure budget.
That sounds like good news, and in one sense it is. It’s also the trap.
If everyone can produce more, at lower cost, with fewer constraints, then volume stops being scarce and it stops being an advantage. Publishing more than your competitors used to be a viable strategy because output was expensive and therefore rationed; when output is cheap for everybody, being prolific is simply the new baseline, and a feed full of content generated from broadly similar tools, trained on broadly similar patterns, converges on a house style that nobody chose and nobody remembers.
AI has made content production cheaper at the moment that distinctiveness has become more valuable. That paradox is the real story of 2026, far more than any single platform update. As platforms get better at matching content to the right person, the harder question is no longer whether it will arrive. It’s whether it deserves the attention it gets when it does.
Now we have to optimise interest, and interest is harder because it requires decisions: a clear point of view, a sharper expression of it, a willingness to make something that won’t appeal to everybody. You can reach exactly the right person and still be invisible.
Why this is an employer-brand problem, not a social-media one
The old (overly simple we think) model of employer brand communications ran in a straight line: message, target audience, media, reach. You wrote the EVP, you identified who needed to hear it, you bought the media, and reach was the metric that told you whether it had worked. That model assumed the scarce resource was distribution, and for most of the last two decades it was.
The model now looks more like this: distinctive idea, audience response, algorithmic learning, distribution.
Response has become an input rather an outcome. How people engage with your content informs how much further it travels, and that feedback loop is most obvious in organic distribution, but the underlying principle increasingly shapes paid performance too. Reach is less something you simply buy up front and more something the response either unlocks or quietly caps.
If your employer brand promises belonging, ambition, stability, progression, impact and flexibility, you are, in all likelihood, promising exactly what every competitor in your category is promising, because those words appear in many (nearly every?) EVP written in the last ten years. Naming those attributes has never made an organisation distinctive. What makes them distinctive is the particular truth, behaviour, tension or lived experience that makes belonging recognisably yours rather than a synonym for “we’re a nice place to work.” A logistics company’s version of stability looks nothing like a scale-up’s version of stability, and an EVP that fails to locate that difference isn’t really saying anything at all; it’s just occupying the category.
The uncomfortable implication is that greater automation doesn’t fix a generic EVP. It just distributes it more efficiently. If the underlying idea is the same as everyone else’s, all that sophisticated targeting and AI-assisted production buys you is a faster, cheaper way of being ignored at scale. AI can multiply an idea. It cannot make an undifferentiated idea distinctive.
This is why employee advocacy deserves to be treated as a structural response to what’s happening, not a nice-to-have social media tactic bolted onto the end of a campaign plan. As the open internet fills with generated and increasingly polished content, audiences have more reason than ever to value things that visibly could not have come from a model: a specific memory, an awkward but true moment, a voice that sounds like exactly one person and nobody else.
Reddit is the clearest evidence of that instinct playing out at commercial scale.
Announcing second-quarter results in July 2026, with daily active uniques up 18% year on year to 130.3 million, chief executive Steve Huffman put the company’s performance down to precisely this dynamic: “In an increasingly automated web, the value of real human perspective has never been higher. Reddit’s commercial momentum reflects that.” It’s a self-interested reading, obviously, but it isn’t only Reddit making it. Dr Yusuf Oc of Bayes Business School, speaking to the BBC, described the same behaviour from the demand side: as content feels increasingly automated, “people look for signals of lived experience, disagreement and nuance.”
There’s signal closer to the ground, too, though it needs handling carefully. Sprout Social’s 2026 research found that 40% of consumers discover new products or services through employee-generated content each month, rising to 62% among Gen Z. That’s consumer research rather than employer-brand research, but the behavioural signal is interesting: individual employee voices are becoming an increasingly important route into organisations.
Worth noting the same study’s more awkward finding, though. Asked who they trust for product recommendations, respondents put brand employees last, on 8%, well behind everyday users on 57%. Read together, those two results suggest something useful about what employee content is actually good for. Employees are not as persuasive as endorsers. They are credible as evidence, which happens to be exactly the job employer brand needs them to do.
Because employees possess something a model, however capable, cannot generate on demand: specificity born of experience. Not a polished summary of what it’s like to work somewhere, but the particular, unrepeatable texture of actually having worked there. That’s the argument for building advocacy properly rather than performatively: fewer scripts and more guardrails, more emphasis on observable behaviour and less on corporate claims, and a tolerance for content that feels experienced rather than approved, because believability collapses the moment something sounds like it went through eleven rounds of sign-off.
None of this means the discipline of employer brand has become easier, and it certainly doesn’t mean the answer is to publish more, faster, everywhere. It means the industry has lost its favourite excuse. For a long time it was possible to blame a weak campaign on the wrong audience, the wrong platform, the wrong budget. That excuse is thinning, because the systems distributing content are increasingly sophisticated at identifying who is most likely to respond. What they can’t do is manufacture a reason for that person to care.
The real question for any employer brand leader going into the next planning cycle isn’t which platform to prioritise or how much budget to shift toward AI production tools. It’s whether the idea at the centre of your EVP would survive being said out loud by an actual employee, unscripted, to someone who already works there. If it wouldn’t, no amount of automated distribution is going to rescue it. If it would, you have something worth being disciplined and unfashionable enough to build a strategy around, rather than a slogan to multiply.
If your employer brand could belong to any employer, better targeting and more content won’t fix it. We help organisations find the distinctive truth at the heart of their employee experience, then turn it into ideas people actually notice, believe and respond to.
If that sounds like a challenge worth solving, talk to us.