Brand compliance checks: how to review every asset against your brand guidelines

A brand compliance check compares an asset with the written rules of a brand (logo placement, colours, typography, photography style, mandatory legal text) and records which rules it meets. Brand teams have always done this by eye, one file at a time, usually just before publication. This guide explains how to do it for every asset as it enters the library, which parts AI can judge and which parts should be measured exactly, how to write rules that a model answers consistently, and how to keep a reviewer accountable for the result.

The brand guidelines glossary entry covers what goes into the guidelines themselves. This page is about checking assets against them.

Why check assets in the asset library

Most off-brand material is not created on purpose. It comes from an agency that used last year's logo file, a dealer who recoloured a banner to match a showroom, a photographer who delivered images without the required clear space, or a colleague who exported a slide at the wrong size. By the time the brand team sees it, it is often already published.

Checking in the asset library moves the review to the point where files arrive: an upload, a new version of a file, or a batch delivered by an agency. Three things follow from that:

• The result travels with the asset. A compliance status on the asset can be filtered in search, shown on the thumbnail, and used to hide non-compliant files from partners or to block a workflow step.
• The rules are written once. Guidelines that live in a PDF are interpreted differently by every reviewer. Rules written as short requirements are applied the same way to every file.
• The history is kept. Each check records which rules it applied, what it found and who approved the outcome, which matters when someone asks later why an asset was published.

Which rules AI can check, and which it should not

Vision models are good at describing what an image shows: whether a logo is present, what it sits on, whether a person is smiling, whether a product is shown in use, what text is printed on it. They are poor at precise measurement: whether a colour is exactly #047DBA, whether an image is 1,200 pixels wide, whether the file is under 25 MB. A model asked these questions guesses, and its guess changes from one run to the next.

Split your rules accordingly:

A good brand compliance tool lets you mark the second group as exact checks so they are computed from the file and its record, never asked of the model. They cost nothing to run and they give the same answer every time.

How to write rules that get consistent answers

A model answers each rule on its own, looking at one asset. The wording of a rule decides most of the quality of the result. These principles follow from how vision models read a single image, and match the guidance DAM vendors publish for their own checks:

1. One requirement per rule. "The logo must sit on a light background and be at least 10% of the width" is two rules. When it fails, nobody knows which half failed.
2. Use "must" or "must not". "Should" and "ideally" invite the model to pass almost anything.
3. Add a condition when the rule only sometimes applies. Start with "If …:". "If a person is shown: they must be using the product." Assets without a person then get Not applicable instead of a failure, which keeps the result readable.
4. Give exact values. Name colours by hex value or by the names in your brand palette, sizes in pixels, proportions as ratios.
5. Describe what can be seen. The model sees one image, the text on it, its colours and size. It cannot know where the asset will be published, who approved the shoot, or whether a model release exists. Rules about those belong in metadata checks.
6. Avoid vague words. "On-brand", "premium", "modern" and "clean" mean something different to everyone, including the model. Say what they look like.
7. Show, when words are not enough. A reference image of the approved logo, or of an example to avoid, makes logo rules far more reliable than a description. A rule that shows the approved wordmark catches a catalogue cover printed with an older version of it: "The logo is displayed in lowercase letters and uses a regular o without the required diagonal slash." Without the reference, the model has no way to know which version is approved.

Start with ten to fifteen rules for your most common asset type, run them on fifty assets you already know are compliant or not, and adjust the wording until the answers match your own judgement. Then add rules.

Keep a person in the review

AI makes mistakes, and brand decisions often have exceptions: an approved monochrome logo for a co-branded event, a partner's colours on a joint campaign. The check should propose a status, and a person should approve it or set another one with a reason. The approved status is the one that goes on the asset; the AI's findings stay visible next to it, so a later reader sees both what the check found and why a reviewer decided otherwise.

Plan for the volume. A library that receives a few hundred images a week can review each proposal. A library that receives thousands needs a way to approve compliant proposals in bulk and to send only failures to a reviewer.

For regulated content (financial promotions, health claims, alcohol, packaging), treat the check as a first pass that catches obvious problems, and keep the legal review you already have.

What it costs to run

An AI check is one call to a vision model per asset and rule set, with the image, the rules and the text read from the image. With current multimodal models that costs a fraction of a US cent per check for a few dozen rules. Exact checks cost nothing. The cost grows with the number of rule sets an asset is checked against and with rechecks after rules change, so a daily limit on the number of checks is worth having from the start, with checks over the limit waiting for the next day rather than failing.

How vendors approach it

Several DAM vendors now offer brand compliance checking, often as an "agent" or an AI add-on. When you compare products, ask:

• Can exact values (colours, sizes, formats, metadata) be checked without the AI, or does the model judge everything?
• Is each finding explained with what the model saw, rule by rule, or is there only an overall score?
• Must a person approve the result, and is an override kept with the original findings?
• Can the result be used in search filters, access rules and approval workflows, or does it stay in a report?
• Which file types are checked: images only, or PDFs, presentations and video frames too?
• How many rules and characters fit in one rule set, and how many assets in one run? Bynder's Brand Compliance Agent, for example, documents 50 rules or 10,000 characters per agent, 1,000 assets per run and files up to 25 MB (support.bynder.com, read on 25 October 2026).
• How is cost shown and capped: credits per check, or the provider's actual cost?

How Freedam does it

In Freedam, brand compliance is part of the asset library on the Business and Tailored plans:

• Rule sets hold rules in plain language, grouped by category (logo, colour, typography, photography, copy, legal, format), with up to 50 rules and 10,000 characters per set. Each set applies to the assets its conditions select, using the same filters as search.
• Exact checks cover colours from a palette, minimum dimensions, aspect ratio, file format, file size and required metadata. They are computed from the file and its record, never by the AI.
• Every upload and every new file version is checked against the rule sets that apply to it, and anyone with permission can check an asset again.
• Each check records a finding per rule with an explanation, and proposes Compliant, Partially compliant, Non-compliant or Not applicable. Rules can be blocking or not.
• A reviewer approves the proposal or sets another status with a note. Only the approved status is stored on the asset, where it appears on gallery cards and in filters, access rules and saved searches.
• Every save of a rule set is a version, and every check records the version it used.
• A daily allowance of checks comes with the plan (500 a day on Business, 2,000 on Tailored). Checks over it wait and run the next day, and an activity page lists what was checked each day, with what result, and what is waiting or failed.

The Brand Compliance documentation explains how to set it up, and Brand compliance on assets shows the review on the asset page.

Frequently asked questions

What is brand compliance software?

Software that checks marketing assets against an organisation's brand guidelines and records whether each asset complies. In a DAM, it runs on assets as they are uploaded and stores the result on the asset, so it can be used in search, sharing and approvals.

Can AI check brand guidelines reliably?

It can judge visual rules (logo presence and placement, backgrounds, photography style, printed text) well when each rule states one requirement and a reference image shows what is meant. It should not be used for exact values such as hex colours, pixel sizes or file formats, which are better measured directly.

How many rules should a rule set have?

Start with ten to fifteen rules for one type of asset, test them on assets you have already judged yourself, then add rules. Several smaller rule sets (one per brand, market or channel) are easier to maintain than one large set.

Does an AI brand check replace the brand team?

No. It does the first reading of every file so the brand team reviews proposals and exceptions instead of every asset. A person should approve the result, especially for legal and regulatory rules.