Most people who've studied brand identity theory even briefly can recite five of Kapferer's six facets without much effort — physique, personality, culture, relationship, self-image. The sixth one, reflection, tends to get treated as the throwaway facet, the one people skim past on the way to something that feels more immediately actionable in a client meeting. I think that's a mistake, and I think AI-generated brand content is making it a much more expensive mistake than it used to be.
What "reflection" actually means, and why it's not the same as "target audience"
Jean-Noël Kapferer's Brand Identity Prism describes six interdependent facets that together constitute a brand's identity: physique (the tangible, physical traits), personality (the character the brand projects), culture (the values and codes it draws on), relationship (the kind of exchange it implies between brand and customer), reflection, and self-image. Reflection is the subtlest of the six, and the one I think gets most consistently flattened into something simpler than it actually is. It's not your target audience, in the market-research sense of a demographic segment. It's the stereotype of a customer that the brand's own communication seems to be addressing — the mirror image a piece of marketing holds up, whether or not that image matches who's actually buying. Self-image, the sixth facet, is the flip side: how the customer sees themselves, internally, as a result of using the brand.
The distinction matters because a brand can target a broad, varied audience in the market-research sense while still projecting a narrow, specific reflection in its actual communication — and when those two things mismatch, the customer notices something is slightly off, even if they couldn't name the mechanism.
Why AI-generated content defaults to a narrow reflection unless someone actively steers it
Here's the finding I think deserves far more attention than it's currently getting in branding circles specifically, as opposed to the broader AI ethics conversation where it already lives. In 2023, Bloomberg conducted a large-scale analysis of the image generator Stable Diffusion, producing thousands of images across a range of occupations and prompts. The results skewed dramatically: prompts for high-paying professions like CEO, lawyer, judge, and politician overwhelmingly generated images of lighter-skinned men, while prompts for lower-paying roles like housekeeper or fast-food worker skewed toward darker-skinned subjects and, disproportionately, women. A separate University of Washington study, presented the same year, documented similar patterns of gender and racial stereotyping baked into the same category of tool.
Translate that finding into Kapferer's terms and something clicks into place that the AI ethics framing alone doesn't quite capture. These models aren't just exhibiting demographic bias in some abstract, statistical sense. They're defaulting to a specific, narrow reflection — a mental image of who a "CEO" or a "customer" looks like — and every single time a brand asks one of these tools to generate imagery or write copy without explicitly countering that default, it gets that stereotype back, dressed up as a neutral creative choice. A brand that hasn't deliberately specified its own reflection is not getting a neutral, blank-slate output. It's getting whichever narrow stereotype the model's training data happened to statistically favor, presented with exactly the same fluent confidence as a genuinely considered creative choice would have.
The quiet mismatch this produces
Here's where this becomes a brand problem rather than a purely technical one. A brand that has spent real effort building a broad, inclusive positioning — the kind of positioning I've written about in the context of not niching an audience down unnecessarily — can still end up projecting a startlingly narrow reflection through its AI-assisted content, simply because nobody explicitly told the tool otherwise. The self-image facet suffers as a direct consequence: a customer who doesn't see anyone resembling themselves reflected back in a brand's communication has a much harder time forming the "this is for someone like me" self-image Kapferer's model says is central to genuine brand attachment, even if the brand's actual stated values and actual customer base are far broader than what its AI-generated content happens to be depicting.
This is, I'd argue, a subtler and more insidious version of the same mechanism behind Levi's 2023 backlash over AI-generated diversity in its product imagery — not identical, because Levi's case involved an explicit, named initiative rather than an unnoticed default, but structurally related. Both cases involve a brand's actual, stated identity commitments colliding with what an AI system, left unsteered, quietly does by default. Levi's mismatch was loud and immediate because it was announced publicly. The mismatch I'm describing here is quieter, because it's not announced at all — it's simply what shows up, unremarked, in the ordinary flow of AI-assisted content a brand produces every week, month after month, without anyone stopping to check whether the reflection in that content still matches the brand's actual intended identity, or ever matched it in the first place.
What checking for this actually looks like
I'd suggest a specific, concrete exercise here, because "be aware of bias" is advice too vague to actually run. Pull a sample of your brand's AI-generated visual or written content from the last month specifically for this check — not the general integrity check I've recommended elsewhere, but one narrowly focused on reflection. Ask, plainly: who does this content seem to be picturing as the customer, across gender, age, and background, and does that match who you actually serve or who you actually want to serve five years from now, not just today? If there's a consistent, narrow pattern showing up that nobody explicitly chose, that's not a values violation in the dramatic sense Levi's faced. It's a quieter version of the same underlying gap — an unexamined default quietly standing in for a deliberate identity choice, facet by facet, without anyone ever having actually decided it should.
Why the other five facets don't catch this on their own
It's worth being explicit about why this gap can survive even a brand that's genuinely done the work on its other five facets. A brand can have a meticulously documented physique — exact colors, exact typography, exact logo usage — and a clearly articulated personality and culture, and still have never once written down, anywhere, what its reflection is supposed to look like, because most brand guideline documents I've reviewed over the years simply don't have a section for it. Physique and personality get checked constantly, because they're visually obvious the moment they're wrong — a logo in the wrong color jumps out immediately. Reflection is quieter. Nobody flags a piece of AI-generated content for looking "off-brand" because of who it pictures, in the way they'd flag it for using the wrong shade of blue, even though the effect on a customer's sense of belonging may be considerably larger.
A concrete way to write reflection down, finally
If your brand guidelines don't currently have a reflection section, here's what I'd actually put in one. A short, explicit description of who your communication should appear to be picturing — not your full target market in all its breadth, but the specific mirror image your content projects back at whoever's looking at it. A handful of reference examples, images or copy you'd hold up and say "yes, this is the reflection we intend," so anyone briefing an AI tool, or reviewing its output, has something concrete to check against rather than a vague instinct. And a standing instruction, given explicitly to whoever operates your AI content tools, to actively counter the tool's statistical defaults rather than assuming a neutral prompt produces a neutral result — because Bloomberg's data makes clear it very much does not.
Why this facet deserves to stop being the one people skim
I think reflection gets skipped in most practical brand work because it's the hardest of the six facets to operationalize into a checklist — physique and personality translate easily into a style guide; reflection requires actually looking at what a brand's communication implies about who it's picturing, which is a slower, more interpretive kind of check. That difficulty is exactly why it needs to be named and checked deliberately now, rather than left to intuition, because the tools producing an increasing share of brand content have their own very strong, very specific, very unexamined intuitions about who they're picturing — and those intuitions were never yours to begin with unless you actively made them so, one deliberate reference example at a time, rather than inheriting them by default from whatever the training data happened to contain.
Wondering whether your own AI-generated content is quietly reflecting a narrower customer than the one you actually serve? Get Your Free Brand Audit →
References
Kapferer, J.-N. (2008). The New Strategic Brand Management: Creating and Sustaining Brand Equity Long Term. Kogan Page.
Nicoletti, L., & Bass, D. (2023, June 8). Humans are biased. Generative AI is even worse. Bloomberg.
University of Washington. (2023). Gender and racial stereotypes in text-to-image models presented at EMNLP 2023.