The brand kit an AI can actually follow

Every AI content tool asks you to “describe your brand”. Almost all of them accept a paragraph. And a paragraph is exactly what makes the model invent the rest.

This is the list of fields worth filling in, and — more usefully — the kind of bad piece each one prevents. If a field prevents nothing, it should not be in the form.

Why a paragraph is not enough

A language model fills gaps. That is literally what it does. If you do not say who the audience is, it picks one. If you do not give the price, it writes “special price”. If you do not list the words you never use, it will use the most common ones in your sector — which are exactly the ones you spent ten years avoiding.

The difference between a good kit and a paragraph is not quantity of text. It is specificity a model can obey. “We are a welcoming bakery” cannot be obeyed. “Never write ‘flash sale’” can.

Identification — prevents the piece that belongs to another company

Legal name, industry, website, city and state. It looks like paperwork, and it is what stops two annoying errors.

The first is geographic: without a city, the model writes as if you were national, and the piece loses the “here in the neighbourhood” that makes a local account work. The second is sector: the same “write a tip post” request becomes very different things in a clinic and in a parts shop.

Personas — prevents text written for nobody

Describe two to four real people from your audience: rough age, what they do, when they buy, what worries them, what they have already tried.

A well-written persona changes the structure of the text, not just the vocabulary. “Mother, 30 to 45, buys in the morning on the way to work” produces a caption about time and convenience. “Retired, 65+, comes in the afternoon and likes to chat” produces a completely different one, about the same bread.

The common mistake is writing the persona you would like to have instead of the one you have. The model will obey whichever is written, and the piece will speak to an audience that is not yours.

Voice — prevents generic AI text

This is the block that changes the output most, and the worst filled in.

Formality and complexity on a scale. Two sliders, not adjectives. “Informal” means different things to different people; a value on a scale does not.

Required terms. The words you always use. If your product is a “membership” and never a “plan”, write that down. Models like synonyms, and a synonym is brand noise.

Banned terms. The highest-return field in the entire kit. Every sector has burnt-out words — “unmissable”, “revolutionary”, “transform your life”. A list of ten banned words improves the text more than any tone instruction.

Good and bad examples. Two captions you would write and two you would never write. This is worth more than the rest of the block put together: models learn far better from examples than from rules. If you only have time to fill in one field of the kit, fill in this one.

Offers — prevents “contact us for pricing”

Name, description, price and link. The price is the field most people skip, and the one that generates the worst kind of piece: the one that promises without saying anything.

With the price in the kit, the caption can say “from $4.20”. Without it, the model writes “special price” or invents a number — and inventing a price is the most expensive mistake a system like this can make. With the link, the CTA becomes a pointer to the right place instead of a vague sentence.

People who can appear in videos — prevents the face that changes every week

This one is video-specific and worth explaining. If you register real people who can appear, the first on the list becomes the default, and pieces gain continuity — the same person shows up in the videos, like a real brand.

If the list is empty, the person is AI-generated, different in every piece. That is not a bug: it is the only honest output when nobody is registered. But it is worth knowing that is the choice you are making by leaving the field blank.

Visual — prevents the feed that looks like three brands

Five colours (primary, secondary, accent, background, text), heading font, body font, logos.

Those colours are not panel decoration: they are injected as CSS variables into the template at render time. That is why the same template comes out with each client brand — the template asks for “the primary colour”, and the kit answers what it is.

If you do not know your colour codes, there is the scan: paste your company website address and EverFeed proposes the colours, the Google fonts the page loads and candidates for logo and icon. Everything is a suggestion — nothing enters the kit until someone accepts it. For most companies that already have a website, that fills the entire visual block in one click.

Compliance — prevents the problem that is not a marketing problem

Per post type, the disclaimer that has to appear.

Anyone working in health, law, finance, food or education knows exactly what this is about: a professional registration number, a “results not guaranteed” notice, nutritional information, “consult a specialist”. People outside those sectors tend to think they do not need it — until the first piece about a children product.

The design point here is that the disclaimer is per post type, not global. The notice that has to be on an offer post is not the same as on an educational one, and cramming all of them into every piece is the fastest way to make sure nobody reads any of them.

How to fill this in without losing an afternoon

The order that works, in our experience:

1. Run the website scan. Solves the visual block and part of the identification in one click.
2. Write the banned terms. Ten minutes, highest return in the whole form.
3. Paste two good examples and two bad ones. Take them from posts you have already published.
4. Register the offers with price and link. Only the ones you actually promote.
5. Write one persona. One. The main one. Add the others later.
6. Leave the rest for next week. The kit has autosave and is built to grow.

A kit like that already produces pieces that pass. And it improves on its own as you use it: every time a piece comes out wrong, the fix is usually a field that was missing — which turns every error into a permanent improvement instead of a regeneration.

The test for whether your kit is good

Ask for a piece on a subject you have never asked for before and read it without editing. If it sounds like your company, the kit is good. If it sounds like any company in your sector, what is missing is not a better prompt — it is a filled-in field.