Brand Constitutions

A Brand Constitution is not a brand manual. A manual describes how a brand looks and leaves the rest to the reader's interpretation. A Constitution is written to be executed: numbered rules, facts with their source and date, and an evaluation set that lets you audit whether a system consuming it behaves as it should.

The difference is in who reads it. A person interprets a manual. A Constitution is executed by people and also by systems: the team's internal projects and assistants, content-generation tools, the agencies and partners that work for the brand, and the artificial intelligences that answer questions about it without ever having been invited to.

More and more purchase decisions start with a question asked to a language model. That model does not read the whole site: it retrieves a loose fragment, one or two paragraphs, and compares it with other sources. Whatever contradicts itself or cannot be verified gets discarded.

That is where the problem almost no brand has measured shows up. If the website says one thing, the sales deck says another and LinkedIn a third, the model does not pick the correct version: it distrusts all three and answers with whatever it finds elsewhere. Inconsistency is not punished with a bad description. It is punished with absence.

A Constitution solves that the least glamorous and most effective way: it fixes one canonical fact for everything the brand claims, with its source and its review date, and it forces every surface to say the same thing in the same words.

A Constitution keeps a brand consistent even when nobody is watching.

Brands become inconsistent without anyone deciding it. New people join, the agency changes, someone rounds a number so it looks better on a slide, someone else translates a claim and shifts its meaning, a partner describes the client in their own words. None of those moves is a serious mistake on its own. Added up over three years, they produce a brand that says five different things about itself.

The document cuts that off at the source: if a piece of material contradicts the Constitution, the Constitution wins and the material gets corrected. And whatever is still undefined gets marked as pending, instead of being resolved a different way every time.

The work starts with the diagnosis, not the writing. Brand Echo Tracker, Mashin's tool that monitors how people talk about a brand and how AI tools recommend it, shows what version of the brand is circulating today in the models.

From there comes what is most uncomfortable and most useful: the inventory of contradictions between the client's own sources. The Constitution is then drafted, resolving those contradictions one by one and marking what the team still has to decide.

The last stage is publishing in machine-readable format, so an outside agent takes the declared data instead of inferring it from the site's prose.

Artificial intelligence assistants no longer just answer questions: they compare options, filter candidates and in some cases complete the purchase. When that happens, the brand's interlocutor stops being a person who can be won over with a good piece of creative work.

An agent does not get excited about a campaign. It verifies. It compares what the brand says with what other sources say, and discards what does not add up.

Preparing for that conversation is not a positioning tactic: it is putting in order what the brand claims about itself so it can be understood by whoever has no context and no goodwill.

Mashin wrote its own Brand Constitution before offering the service. The process found twelve contradictions between the website, the credentials deck and the agency's internal materials: from the number of countries it operates in to how it refers to itself.

They are documented, with their origin and their resolution status. It is proof that the method works, and also of what it finds: no brand comes out unscathed from looking at itself closely.