A forty page brand book and a folder of approved campaigns. That is what most teams have when they decide to make a model work in their brand. The catch is that a brand book is written for people who can sense what is meant by clean and confident. A model cannot sense anything. Everything has to be spelled out or shown.
Start with the rules you can actually check
Go through the brand book with one question in mind: could a freelancer who started this morning follow this line without asking anything? Keep the rules that pass. Rewrite the ones that do not.
Confident and human becomes short sentences, active voice, no more than eighteen words per sentence, no exclamation marks. Premium look becomes one product per frame, dark background, one light source from the left. You will usually find that forty pages contain about twenty five rules that survive this test, and those twenty five do almost all the work.
Choose the examples that will shape the model
Examples teach faster than rules, so pick them carefully. A reasonable starting set looks like this:
- Fifteen to twenty pieces of published copy across the formats you make most: a landing page, five social posts, two emails, a product description.
- Twenty to thirty approved images that represent your normal standard, not your award entries.
- Ten minutes of clean audio per voice you use, recorded on one microphone in a quiet room.
- Two or three finished videos that show your pacing, your titling and how you cut.
- Your logo files, colour codes and typefaces in the formats your team really uses day to day.
Quality beats volume every time. Twenty pieces you are proud of will produce better results than two hundred where half are three years old and off the current standard.
Write down what is never allowed
Most teams document the ideal and skip the opposite, which is a mistake. The fastest improvement comes from a clear list of what must not happen. No stock photo smiles. Never the accent colour as a background. Never the word solution. No sentence longer than twenty five words. No music under a spoken testimonial.
Add ten rejected examples with a one line explanation each. A model that has seen what fails, and why, corrects itself sooner. This is also the part your team can build fastest, because everybody remembers the things that got sent back last month.
Prove it works before you trust it
Build a test set of ten briefs that are typical for your brand, including two you know are hard. Run them, then score each result on three things: is it factually correct, does it follow the rules, would you publish it as is. Anything below eight out of ten on the third question means your rules or your examples are still too vague.
Keep the same test set forever. Its value is that it does not change, so when you add examples or rewrite a rule you can see whether the result really improved or just felt different.
Keep the model alive
The strongest input is not the brand book at all. It is the stream of small decisions your team makes every day: this one goes out, this one does not, this line needs to be shorter. When those approvals and corrections feed back in, the model gets closer to your brand every week instead of drifting away from it.
Set a rhythm you can hold. Once a quarter, spend an hour removing examples you would not publish today and adding the five best pieces from the last three months. Once a year, check the rules themselves against what your team is actually doing, because in practice the standard moves before the document does.
One concrete way to begin: take the last twenty things your team published and sort them into yes, no and almost. Write one sentence per item explaining the call. That single afternoon gives you sharper input than the whole brand book, and you can hand it to a model tomorrow.
Further reading
Want to see what Content-Agent makes for your brand?
We load your brand rules and styling once. After that you make images, video, audio and text in minutes, always on brand.