Kerstin Ludwig
going→fwd

← Writing

Module 12 · Agentic Brand Thinking · Published May 1, 2026 · Last updated Jun 26, 2026

Identity has to survive machine compression

Core → verification → fragments

Agentic Brand Thinking — Module 12

Visual continuity, machine-readable data, provenance signals: the layer that makes survival possible. Without a tangible core underneath, the layer just makes you legibly empty.

I’ve spent a lot of time on the foundations: Brand OS, governance, drift control, decision rights. Twelve modules of infrastructure. Now we get to expression, identity’s JTBD.

We designed the journey. Awareness, consideration, decision — sequenced, measured, defended in reviews. A model now compresses that sequence into one answer. The orchestrated build-up loses most of its leverage.

People still meet our brands: out of home, in a Slack recommendation, a citable piece, a product. The journey is shorter, less linear, less controllable.

The human doesn’t disappear. The human is the one who has to answer the question the machine will expose either way: is there actually something here?

The machine executes the identity. Someone still has to decide what it is. That decision is upstream of schema, C2PA, compression-grade design — all of it.

If we don’t design for compression, we don’t show up.

Design’s old job: pull the eye, hold it, hand it to the next stage of the funnel.

Design’s new job: confirm something real exists underneath — fast enough for a model, a screenshot, or a stranger to verify in one look.

The core, not the surface

Marty Neumeier wrote in 2003 that a brand is not a logo, not an identity system — it’s the gut feeling people have about what you do. Todd Irwin’s De-Positioning makes the contemporary version: you don’t win by being different, you win by out-solving customer pain better than anyone in your category.

Neither argument is new. What changed is the exposure. Beatriz Vilchez Silva (LaColmena Design) ran a stress test with Claude Design recently — three prototypes, a couple of hours. What didn’t get faster: knowing what to make. The machine executed her register. It couldn’t decide what the brief was.

Frontify’s CMO survey (450 respondents across the UK, US, and DACH) found the single biggest factor in brand resilience is a strong concept, ranked above consistency and above visual coherence.

Naming the core is harder than it sounds. Most brand teams can describe what they do. Few can say what they would hold when a client pushes, a quarter is down, or a model returns something close enough.

Aesop sold for $2.5B with no loyalty programme. Aesop acquisition press release. What it had: a way of meeting the brand that was consistent across every format — store as ritual, product as artefact, copy as voice. Nothing about it was accidental or copyable in isolation. Compression finds that kind of core. If the core is missing, compression finds that too.

Recognition under compression

Visual identity used to be judged in controlled contexts. Now it often appears as a fragment: small, cropped, copied, summarised, or placed next to competitors before anyone has seen the full system.

So the system needs a few elements that still carry recognition when the rest drops away: a shape, a proportion, a typographic behaviour, a composition rule, a motion cue. Not more decoration. A clearer centre.

The AI-company marks are the live case. OpenAI, Anthropic, Perplexity, a dozen others — all converged on the same soft, rounded, symmetrical glyph. Compression-grade design: every mark renders legibly at 16px. The category is also visually indistinguishable. Every player solved the surface problem with the same constraints.

A language survives compression.
A system that looks identical to every competitor’s doesn’t.

Damien Correll put it well: enduring visual identities should be thought of as languages, not systems.

What systems actually read

LLMs and search systems read structured data before prose. If your organisation markup is missing, contradictory, or stale, the model has nothing canonical to anchor to and infers from whatever it has.

This usually falls between SEO, IT, and brand — which is why it usually doesn’t get done. That is the difference between being legible and being inferred.

What sits here: schema.org Organization and Product markup, a monitored Wikidata entry, knowledge graph entities reviewed on cadence, a clean sameAs chain. It decides whether a model cites you correctly, attributes a quote to you, or merges you with a similarly named entity.

Different engines read different signals. Gemini prioritises product schema and live inventory. ChatGPT weights Wikipedia and Gartner citations — institutional consensus built over years. Perplexity draws heavily from Reddit and specialist forums. Claude goes deepest on technical accuracy and surfaces where brands fail, not just where they succeed. Liat Ben-Zur maps this as five distinct logic systems, each with its own credibility model.

Clean markup is the floor. Which signals you invest in depends on where your buyers actually ask.

Design-ahead

Coca-Cola’s Project Fizzion is the clearest current proof. The idea was to make design assets intelligent by learning how to behave from designers, rather than asking designers to memorise hundreds of pages of visual brand guidelines. StyleIDs encode creative intent into a machine-readable identity that applies brand rules automatically across formats, markets, and agency partners. The brand rules travel with the asset.

The scale is Coca-Cola. The logic isn’t. A defined prompt library, a locked asset folder, a one-page visual ruleset a model can follow — that’s the smaller version. No custom model required.

Every piece of content a brand doesn’t define deliberately leaves a gap. The machine fills it — from training data, category defaults, whatever signal is loudest.

Provenance is the next layer. C2PA embeds origin metadata directly in the asset, linking content back to its source. Worth watching; most platforms still strip it.

Verification Layer checklist

Run this against the last three pieces of public content your brand put out. A “no” is a gap.

  • The coreOne sentence describing the brand’s tangible core a customer would recognise without prompting
  • The coreThree product or operational decisions in the last year that only this brand would make
  • The coreOne position in the category competitors cannot truthfully claim
  • Recognition under compressionMarks tested at 16, 32, 64px against three category competitors
  • Recognition under compressionBlack-and-white rendering tested
  • Recognition under compressionAI summary card rendering checked on Perplexity, ChatGPT, Google AI Overviews
  • The machine-readable layerschema.org Organization markup live and validated
  • The machine-readable layerWikidata entry current and monitored
  • The machine-readable layerKnowledge graph entities reviewed quarterly
  • The machine-readable layersameAs chain audited
  • Design-aheadBrand-specific AI model, or equivalent: a prompt library, locked asset folder, or one-page visual ruleset a model can follow
  • Design-aheadGeneration tools reviewed for default drift
  • Design-aheadApproval step for AI-generated brand content
  • ProvenanceContent Credentials (C2PA) enabled where supported
  • ProvenanceSigning step in photography and video workflows
  • ProvenanceProvenance gap log maintained for high-stakes content

The gap that matters most is the first one — the core.
Without it, compression-grade design delivers you, legibly, into the middle of a category that looks exactly like you.

Closing thought

The machine layer gets you found. The core makes you credible. Trust is what gets you chosen. That is not a machine decision. It travels through people — through who carries your brand, and why. That is the next piece.

Verification is what makes identity hold when someone gets close enough to check. Cultural pull is what brings them close enough — and decides whether the encounter is worth verifying at all.

Next

Verification is what makes identity hold when someone gets close enough to check. Cultural pull is what brings them close enough — and decides whether the encounter is worth verifying at all.

Read

Agentic Brand Thinking

If your systems already speak for you → what are they saying?

Work with me See Outcomes

What does ‘machine compression’ mean for brand identity?

Machine compression is what happens when an AI system — a search engine, a chatbot, an agent — takes everything it knows about a brand and produces a summary. The summary is shorter, simpler, and less nuanced than the source material. Identity elements that survive compression are the ones that appear consistently across many sources, are expressed in clear and specific language, and are backed by verifiable evidence. Elements that don’t survive are the ones that exist only in brand decks and taglines.

How do you build a brand identity that survives compression?

Three layers: core (the irreducible claim — one sentence that says what you do, for whom, and differently from what alternative), verification (the evidence that backs the claim — citable, specific, consistent across channels), and fragments (the pieces of identity that appear in summaries — the words and phrases that recur enough to be picked up and repeated). Build each layer explicitly; don’t rely on tone-of-voice documents to do the work.

What are the Agentic Brand articles?

The Agentic Brand Thinking series is a 14-module framework that translates brand strategy into machine-readable operating logic — covering values, voice, proof, policies, metrics, governance, and identity — so AI systems can represent a brand accurately without constant human intervention.