Megaphones and Mirrors

Last May I wrote that Unilever's all-in influencer bet was wrong. I stand by most of it. But I need to correct something I got wrong, and the correction matters more than the original critique.

I dismissed influencer marketing as a format problem. It isn't. It's an infrastructure problem. And the difference between those two things is about to reshape how every CMO allocates budget.

Here's what changed my mind.

When I wrote the Unilever piece, I framed influencer content as borrowed credibility, where brands hide behind paid advocates rather than doing the hard work of earning trust directly. I pointed out that Unilever CEO Fernando Fernandez's logic was flawed: if consumers are "by default suspicious" of big brands, they're arguably more suspicious of people paid to speak for those brands.

I still believe that. But I was looking at the format through the wrong lens.

At Simbioniq, we build AI that simulates human psychology so that brands can stay closer to their consumers. We spend our days studying what makes human communication persuasive at the individual level, such as parasocial trust, narrative transportation and semiotic resonance. And the data forced me to reckon with something uncomfortable: influencer content isn't just another advertising format. It's the most psychologically powerful commercial communication vehicle we've ever created.

The engagement rates aren't a fluke. When someone watches a creator they follow talk about a product, their brain processes it closer to a friend's recommendation than an advertisement. Persuasion resistance drops. Attention is voluntary, not coerced. The content competes with entertainment, not interrupting it.

The format isn't the problem. The format is extraordinary.

The problem is that we're using it like a megaphone when we should be building mirrors.

I wrote recently about the corner store paradox and how every step from corner store to main street shop to big box retailer to national eCommerce chain moves you further from your consumers. Not from data about them (you're drowning in that) but from actually understanding them. CDPs capture the WHAT. The corner store knew the WHY.

Influencer marketing was supposed to solve this. Creators as the new corner store, with trusted faces with authentic relationships, speaking to communities rather than demographics. That's why it works. That's why Fernandez bet billions on it.

The problem is that we've taken a format built on intimacy and deployed it as broadcast. Every other major marketing channel has been personalised to the individual. Display ads are assembled dynamically by decision engines reading hundreds of signals about you specifically. Email is personalised. CTV is personalised. Even out-of-home is heading there.

All except influencer content. The format we keep calling "authentic" still operates on a one-to-many model. A creator speaks to a demographic. Not to a person. A fitness influencer is incidentally more relevant to fitness purchasers, but that's contextual adjacency, not personalisation.

This is the gap. Not more influencers. Not louder megaphones. The gap is the missing mirror, where content with the emotional power of human storytelling, adapted to reflect back each individual viewer's context, psychology, and needs.

Imagine a creator you follow posts a product recommendation. The version you see shows the product in a kitchen where you feel happy, perhaps suburban, warm lighting, family context. The soundtrack is relaxed. The creator speaks at a measured pace with a slightly premium register. Your neighbour sees the same creator, same product, same authentic voice, but the setting is an urban apartment, the music has more energy, the use occasion is different, the cultural references shift subtly.

Neither of you perceives the adaptation. You both just feel the content "speaks to you."

This isn't "Hi Jane!" personalisation, the overt name-insertion that signals surveillance and triggers rejection. 54% of consumers say personalised ads creep them out. But Netflix's recommendation engine doesn't creep anyone out. The difference is semiotic adaptation versus explicit personalisation. Netflix doesn't tell you why it showed you something. It just shows you something that feels right.

I know what you're thinking. This sounds compelling in theory, but AI video still looks like AI video. People can tell. The uncanny valley protects us.

Except it doesn't. Not anymore.

Three things happened since I wrote about Unilever last year. First, AI video quality crossed a threshold: models powering platforms like HeyGen and Synthesia now produce output where trained observers struggle to distinguish AI from human in controlled settings. Second, consumer calibration shifted. As AI-generated content became ambient across social feeds, the binary of "real versus fake" started dissolving. It's being replaced by a different filter: "relevant versus irrelevant." Audiences increasingly don't care how content was made. They care whether it's worth their attention. Third, the infrastructure for licensed digital twins of real human creators now exists at commercial scale.

The moment AI video becomes indistinguishable from human video, the strategic question inverts completely. It stops being "can AI replicate a human influencer?" and becomes "can we make the AI-rendered version more relevant to each individual viewer than the generic human version?" That inversion is the unlock. That's when personalised influencer content moves from experimental to inevitable.

What can already be adapted without losing authenticity: language and accent, product variant, background environment, voiceover register, soundtrack, caption style. What cannot yet be done at consumer scale: real-time rendering of a photorealistic licensed creator, personalised per viewer, at programmatic speed and cost. The technology exists in fragments. The integrated pipeline does not. That's the race.

When I critiqued Unilever last May, I focused on volume. Too many influencers. Too little control. Brand dilution risk. And Fernandez has proven those concerns real since Unilever now works with close to 300,000 creators, and creator rates have jumped roughly 30% industry-wide since the announcement. The measurement challenge alone is staggering.

But the deeper problem isn't that Unilever hired too many influencers. It's that they're solving a reach problem when the real opportunity is a relevance problem. They've built the biggest chain store in influencer marketing. What they need is 300,000 corner stores with each one recognising who just walked in.

Three hundred thousand megaphones. Zero mirrors.

Here's what I mean: personalised video generates 16x higher click-to-open rates than generic video. Personalised product videos drive a 35% higher conversion rate. McKinsey's analysis shows effective personalisation can reduce acquisition costs by up to 50% and lift revenues 5-15%.

Combine those two effects. Authentic influencer-style content that's dynamically personalised at the individual level. That's not a cost efficiency play. That's a conversion revolution.

Building this requires five integrated layers: identity and data infrastructure, licensed creator digital twins, a personalisation decision engine, a dynamic video rendering layer, and programmatic distribution. No single company has all five today. But of the major holdcos, Publicis is best positioned. They have the industry's most sophisticated first-party data infrastructure in Epsilon. They've since spent over $750 million acquiring Influential, Captiv8, and BR Media, giving them 15 million+ integrated creators sitting on top of Epsilon's identity graph. The missing piece is a rendering engine. If Publicis acquires or builds that, they have the complete stack.

The alternatives: TikTok Symphony is the most complete deployed attempt but still faces US regulatory uncertainty. Meta has the data depth. Adobe invested in Synthesia and explored a $3 billion acquisition. Startups like Tavus are building the exact rendering layer an incumbent could acquire. The smart money says this gets assembled through acquisition, and the window for first-mover advantage is open now.

If you're a CMO watching the influencer spending arms race and wondering whether to follow Unilever's volume play, pause. The question isn't how many influencers you need. The question is whether your infrastructure can make every piece of influencer content individually relevant.

Three things to do now:

1. Audit your personalisation readiness. Do you have robust, consented first-party data with behavioural and psychographic richness? If you have DCO capability today, extending it to AI video is an upgrade path. If you don't, you're two steps behind.

2. Start building creator licensing frameworks. The SAG-AFTRA/Narrativ model (creators licensing digital replicas with consent, compensation, and governance) is the template. California and New York already require explicit written consent for digital replicas in commercial use. The brands that establish ethical infrastructure now own the trust premium when this category normalises.

3. Test semiotic personalisation at segment level. You don't need individual-level rendering to start. Generate multiple variants of the same influencer-style content, such as different settings, soundtracks, cultural registers, and test which semiotic cues drive conversion for which segments. Build the muscle before the technology demands it.

Last May I wrote that Unilever should "embrace its scale and heritage as strengths" rather than hiding behind an army of paid advocates. I still believe that. But what I underestimated was the format itself, the raw power of human voice and human face, processed by human brains that evolved over millennia to trust faces like theirs telling stories that feel relevant.

The industry doesn't need to choose between influencers and AI. It needs to fuse them. Not by replacing creators with avatars, that's the wrong conversation. But by taking the most psychologically persuasive format we have and giving it the personalisation layer that every other channel already possesses.

Megaphones got us here. Mirrors are where we're going.

But this is also a maturity choice, the same one I wrote about with Dario Amodei's "technological adolescence." AI that personalises influencer content can be built to manipulate, to exploit psychological vulnerabilities with surgical precision. Or it can be built to restore the closeness that scale stole, to make every person feel known rather than targeted.

The corner store didn't manipulate its regulars. It understood them. The brands that build the mirror with that same intent will own the most powerful commercial communication capability in the history of marketing. The ones still buying megaphones will wonder why their 300,000 voices aren't being heard.

Brandflow is written by Justin Billingsley, who has spent his career on all three sides of the industry's table: senior client, global agency leader, technology founder. First published 24 February 2026 in the Brandflow newsletter on LinkedIn.