Your customers’ brand guidelines in an AI product
For founders and engineers whose product makes work under customer brands: how to take in the guidelines of each customer, follow them, and keep them current.
Your product possibly makes work for customers, for example ads, posts, pages, decks, or emails. Then each customer comes with a brand guide, or wants one. Your product must read the guide and follow it. It must also follow the guide after the customer changes it. Customers see brand mistakes before they see anything else.
Take in the guide and the files
- Ask for the files, not only the PDF. Ask for the logo as an SVG file, the font files or font names, and the approved images. A logo from a PDF page is a picture, and an image model draws a new logo from a picture.
- Turn the rules into values that you can apply: color codes, font names, and words to avoid. Keep the words of the customer for every rule that you cannot turn into a value.
- Ask about the gaps. Most guides have gaps. A short set of questions at the start fills the important gaps: words to avoid, claims to never make, and rules on AI.
Follow the rules of the customer on AI
More guides now have a using AI part, and some of its rules apply directly to your product:
- The logo: place the real logo file of the customer on the finished work. Do not generate the logo.
- People: do not make people who look like real customers or staff of the customer, unless the customer allows it.
- Labels: let customers add their own “made with AI” label where their guide asks for it.
- Review: if the guide of a customer requires a person to approve work, give the customer a step to check the work before it goes out.
Keep the brand of each customer current
Connect each customer to one brand on your server. Let customers correct their brand. Record which version of the brand each piece of work used. When a customer changes their guide, new work uses the new version from that moment. Nobody needs to upload anything again.
Using Rainbrand for this
Rainbrand is the brand layer for agents: it turns a brand’s identity, guidance and assets into shared context that agents and products can use. A product can add the brand of each customer through the API. It can then give its agents access to the same brand over MCP (Model Context Protocol). For each job, the product requests a Brief with the guidance and assets for that task. The product makes the work with its own models. Then it can request a Review of the result against the brand, and revise. The pricing page of Rainbrand lists the plans and limits.