Brand identity: how to keep yours when everyone's using the same tools

Everyone has access to the same creative shortcuts

AI has democratised creative production. A junior marketer with a laptop can now generate concepts, headlines, product shots and full campaign decks in an afternoon, work that used to take a studio a week. That's a genuine gain. It's also the reason so much brand content now reads like it came from the same machine, because in a lot of cases, it did.

The brands that avoided this weren't the ones that stayed away from AI. They used it and still came out sounding like nobody else. Nike's “Never Done Evolving”, built with AKQA for the brand's 50th anniversary, used generative tools to imagine an AI-rendered tennis rally, and it still felt like a Nike film: motion, tension, a single athlete carrying the whole idea. Under Armour's “Forever Is Made Now”, produced with Tool of North America and fronted by Anthony Joshua, leaned on AI-driven production and came out as unmistakably Under Armour: physical, unglamorous, built around effort rather than polish. Designer Marco Simonetti's AI-generated Jacquemus x Nike pop-up concept went viral in 2022 for the same reason: it read as a genuine collision of two real aesthetics, not a generic mockup with two logos pasted on.

Three different tools, three different outputs, one shared trait: each one is recognisable with the logo removed. That's the bar. And it's worth holding every AI-assisted campaign to it, including the ones that don't clear it.

AI doesn't kill brand identity. Defaults do.

Coca-Cola's AI-generated Christmas ad is the case study nobody asked for, but it's an honest one. The company used AI to remake its own holiday imagery, the red trucks, the polar bears, the warm-glow Santa, and the result landed with the opposite reaction Nike and Under Armour got. Coverage from Creative Bloq, NBC News and Forbes described it as flat, uncanny and generic, despite using assets that are about as brand-owned as advertising gets. The backlash was strong enough that Coca-Cola faced the same criticism again the following year.

The lesson isn't “don't use AI for Christmas ads.” It's that the same brand, using the same category of tool, can end up sounding like everyone else if the process behind it defaults to whatever the model produces on the first pass. AI isn't inherently generic. Generic prompts, default settings and accepting draft one are. Nike and Coca-Cola had access to comparable AI capability. The difference was what each fed the model, how many rounds it went through, and who was allowed to say no. The problem was never using the same tools as everyone else. It's using them the same way as everyone else.

Your brand identity needs to exist before AI can reproduce it

Before you ask a model to “sound like us,” you need an answer to what “us” actually means, written down somewhere more specific than a slide with three adjectives on it. That means clear voice pillars, an actual vocabulary the brand uses and one it avoids, behaviours the brand exhibits (short sentences, no exclamation points, always names the customer's problem before the solution), and a list of things the brand would never say, which is often more useful than the list of things it would. That's just as true for how your brand looks: in a photo, a poster, a graphic, an ad, or a video.

This isn't a branding exercise for its own sake, it's the raw material an AI system runs on. Under the hood, “sounding like us” (and looking like us) becomes context the model reads before it generates anything: a system prompt built from your brand guide, or a retrieval layer that pulls from a library of your own approved copy and imagery before generating a draft. If that context only contains adjectives like bold, innovative and human, the model has nothing concrete to pattern-match against, so it falls back on the median of everything else it's seen, which is exactly the sameness problem in miniature. Specificity is what makes the input usable. “Bold” gives the model nothing. “We open with the customer's problem in the first sentence and never end on a question mark” gives it something to actually follow.

Teach AI with examples, not adjectives

“Be bold and playful” is not direction, it's a vibe. Adjectives alone leave too much room for interpretation. Examples give AI a clear pattern it can try to match. Instead of describing your brand in the abstract, give the model actual examples inside its working context and let it infer the pattern directly, which produces far more consistent output than a one-line style instruction ever will.

In practice, that means skipping “be bold and playful” entirely and showing the model the difference instead. On the website, we sound like this: “We build AI systems that turn your vision into tangible results, from design to execution.” On social, we sound like this: “How to introduce AI in a company: identify repetitive work, map where the data already exists, start with one automation, measure it, then scale what works.” Same brand, two different rhythms, both real, both usable. A line that's too safe, a line that's trying too hard, a line that sounds like a competitor: each is more instructive than another adjective.

The same fix works for imagery. “Dynamic angles” and “natural lighting” are adjectives with a camera pointed at them, and a model reads them just as generically as it reads “bold.” Replace the brief with an actual moodboard: five to ten real reference shots, each tagged with why it's on-brand, sitting next to a handful of images that look close but wrong, tagged with why they miss. Feed the model what you look like, and just as importantly, what you don't. Over time this becomes a working library of brand references, words and images both, that the model (and any new hire) can draw from, which is a far more durable asset than a one-page tone document nobody rereads after week one.

Feed AI your brand, not generic prompts

Most teams open a chat window and start typing from zero every time, which means every session relearns the brand from scratch and forgets it the moment the window closes. The fix is building the brand into the system once: uploading actual brand guidelines, past campaigns, product detail, audience research and existing content as context the model can reference, rather than re-explaining the brand in every prompt.

That context needs structure, not just volume. Clear pillars give the model principles it can apply consistently across a hundred different requests, rather than a pile of documents it has to guess the priority order of. And the most durable version of this isn't a prompt you retype, it's custom instructions: brand knowledge turned into reusable context that loads automatically, so every new request starts from the brand's actual voice instead of a blank page and a generic system default. This is the layer most teams skip, and it's the one that separates a one-off AI-generated asset from a system that keeps producing on-brand work at volume.

Push back on the output

Most AI models have a documented tendency towards broadly agreeable answers, a behaviour known as sycophancy, explored by Sharma et al. in Towards Understanding Sycophancy in Language Models (ICLR 2024). That's a reasonable default for a general-purpose assistant. It's a bad default for creative work, where the interesting option is often the wild card: the one that gets a reaction, starts a conversation, or divides the room. Creative work that everyone agrees with is usually creative work no one remembers.

So here's how you get around that. Treat the first output as a starting point, not a finished draft. Ask for alternatives that go further in an unexpected direction. Ask the model to argue against its own answer, or to generate the version a competitor would never approve. Request options with more friction in them, the ones with an edge the safe draft sanded off. None of this replaces human judgment. The human is the one who decides the first draft isn't good enough, argues for the angle the model played safe on, and brings the reference, the joke, or the risk the model was never going to generate on its own.

Don't let AI remove the interesting bits

Left to its own devices, AI will often smooth an idea until nothing distinctive is left, because “smoother” and “safer” tend to be the same direction it optimises toward. That instinct is worth resisting deliberately. The slightly odd phrasing. The unexpected visual reference. The cultural joke that won't land for everyone but lands hard for the people it's aimed at. The sentence the grammar checker keeps flagging. These are frequently the parts of a piece of work that make it recognisably yours, and they're also the first things a model will quietly polish away if nobody stops it.

This is the same failure mode behind the Coca-Cola backlash: technically competent output with every rough edge sanded off, which is also what made it forgettable. The fix isn't avoiding AI polish altogether, it's knowing which specific choices in a draft are load-bearing for the brand's identity, and protecting those on purpose while letting the model clean up everything else.

Bring back the creative brain

AI can generate a hundred variations in the time it used to take to make one. But more doesn’t mean better. Everyone has taste and an opinion. A creative brain is trained through years of making things, getting them wrong and learning what works. It knows which ideas are worth pushing, how to make a visual stronger, when a sentence needs a different rhythm, which references will land, what feels current or already dated and, more importantly, what to cut.

AI can give you an entire campaign in minutes. And it might look good. The copy works. The visuals work. Everything is technically there. But without a creative brain behind it, it can still feel empty: no point of view, no tension, no personality, no edge. AI can make a perfectly good ad. A creative brain makes the one you remember.

Same tools. Different eye.

Access to AI is no longer a competitive advantage. Every serious competitor in your category has the same models, the same generation speed and, increasingly, the same default settings. What's left to differentiate on is what you feed those models and what you're willing to reject from what they hand back: the brand context, the specific examples, the standards, and the trained eye to say a technically fine draft isn't good enough.

Start with one document, and don't make it only words. Before your next AI-assisted brief goes out, write down five things your brand would never say, five real lines that sound unmistakably like you, and one recent draft you rejected along with the reason why. Then do the same with images: five shots that look like you and five that look close but wrong, each tagged with why. Turn that into the context every AI session starts from, instead of a blank prompt window. That single file, words and images both, will do more for your brand identity than any adjective-laden style guide sitting unread in a shared drive.

Sources