Most brand guidelines were written for a designer or a writer who reads them and uses judgement. A growing share of proposals, decks and campaign variants is now drafted by AI tools, which follow rules literally and fill every gap with something generic. These are the parts a brand system needs once AI is producing some of the work, taken from the one Pivitt runs on.
A tool cannot read tone from a mood board. Voice has to be written as rules it can apply, with examples of right and wrong.
Where the guidelines are silent, a tool writes something plausible and generic, so the gaps have to be closed in writing.
Without one approved source, numbers, client names and claims change a little every time they are retyped.
Sentence length, words to use and avoid, and how to open and close, with paired examples of right and wrong.
Every figure and client name with its exact wording and its source.
The words and sentence shapes the brand never uses, kept as a list a tool can check against.
How products, clients and numbers are written, the same way every time.
Every draft tested against the rules and marked ready, ready after fixes, or held.
The voice and the check travel into every AI tool the team uses.
Every figure and client name can be traced to its source.
Mistakes are stopped before a client ever sees them.
Part 1 of 3
The voice and the check travel into every AI tool the team uses.
The voice rules and the draft check are packaged as two skills sharing one rules file, so any tool drafts on brand and checks its own work.
22 banned words · 29 approved figures
Swipe to see the whole model
Part 2 of 3
Every figure and client name can be traced to its source.
Each claim is held with its exact wording and where it came from. If a claim is not on the register, it does not ship.
Swipe to see the whole model
Part 3 of 3
Mistakes are stopped before a client ever sees them.
Every draft passes four checks, voice, approved claims, banned words, and names and figures, and comes out ready, ready after fixes, or held.
15 mistakes stopped in the first five days · 73% approved first time, June to July 2026
Swipe to see the whole model
Brand guidelines written so AI tools can follow them: voice as explicit rules with examples, an approved list of claims and figures, banned words and constructions, and a check that every draft passes before it is used.
Ordinary guidelines rely on a person reading them and using judgement. AI brand guidelines turn that judgement into written rules and checks, because a tool applies what it is given literally and fills any gap with something generic.
Any tool that accepts written instructions or a rules file. At Pivitt the voice and the check are packaged so every tool the team uses drafts on brand and checks its own work.
With the brand's existing guidelines and a sample of recent work. The gaps show up quickly: the claims nobody approved, the words that keep recurring, and the places where the tone drifts.