Insight
Why AI output looks off-brand, and how to fix it
By Brett Raven, APAC Lead & Board Member, AIREPublished
AI output looks off-brand because the assistant doesn't know your brand. With no context, it falls back on the most common choices, which is why so much of it reads and looks the same. Switching tools won't fix that. Write your brand down once, put it where the assistant reads it automatically, give it an owner, and every person's output starts from the same place.
What changes
The complaint we hear most in workshops isn't that AI gets things wrong. It's that the output doesn't sound like the business or look like it. Staff fix it by hand, then decide the tool isn't worth it.
- Generic output is what an assistant produces when it has no brand to follow.
- Claude and ChatGPT both let you store brand rules once so every chat starts with them.
- Claude now checks designs against your design system before you see them (Anthropic, 2026).
- The fix is one owner and one approved version, not a better prompt in each person's head.
It isn't the tool
Ask any assistant for a landing page or a client email with no other context and you'll get the safest, most common answer: familiar fonts, a gradient, a big number with a small label, the same handful of phrases. That's not a fault. It's what a model does when you haven't told it what makes you different.
Anthropic says as much. Its own frontend design skill exists to make Claude's pages avoid "generic AI aesthetics", and tells Claude to make choices that don't read as templated defaults (Anthropic, 2026). If the company that makes the model needs a written brief to get past the default, so does your team.
So switching from Copilot to ChatGPT to Claude rarely fixes off-brand output. The new tool has the same gap: it doesn't know your brand either.
Why prompting harder doesn't scale
The usual workaround is a good prompt that one person keeps in a note and pastes in. It works for that person. Everyone else writes their own, the versions drift, and the brand ends up as varied as the people using it. The brand lives in individual habits rather than anywhere the tool can read.
Where your brand should live
The major assistants each have a place to keep context that applies automatically. Use it, rather than relying on people to paste it in.
| Where | What it holds | Who it reaches |
|---|---|---|
| Claude design system | Colours, type, components and layout patterns. Claude builds with them and checks designs against them. | Everyone in the organisation once it's set as the default. A beta on paid plans. |
| Claude skill | Written rules, examples and templates Claude loads when a task needs them, such as a brand guidelines skill. | Every Claude plan, with code execution on. |
| ChatGPT project | Instructions and files that apply to every chat inside the project. They override personal custom instructions. | People working in that project. |
The visual side and the written side are different jobs. A design system keeps layouts and colours on brand. Voice, banned words and how to structure a proposal belong in written rules: a skill or project instructions (Anthropic, 2026; OpenAI, 2026).
How to set your brand once
- Collect real examples. Your best finished work: a landing page, a deck, 3 emails that sound exactly right. Anthropic's own advice for design systems is to include real examples, not just specifications.
- Write the rules down. Colours, type, logo use, tone, words you never use, and which template suits which job. One page is enough to start.
- Put them where the tool reads them. A design system for visuals, a skill or project instructions for writing.
- Test with a plain prompt. Ask for something ordinary, such as a one-page flyer, and compare it with your real work. Fix the rules, not the output.
- Make it the default. Share it with the whole organisation, so new work starts on brand without anyone asking.
For the Claude steps in detail, including how to move design systems before claude.ai/design closes on 14 December 2026, see our Claude Design guide.
Give it an owner
Brand context is a governance job as much as a design one. Without an owner, three things go wrong.
- Copies multiply. People tweak their own versions. Claude lets an admin approve one standard design system and lock edits, which is worth doing in any team of more than a handful.
- Nobody switches it on. On Claude Enterprise, design systems are off until an owner turns them on. Plenty of teams have the feature and don't know it.
- It quietly disappears. Where an organisation deletes artifacts after a retention period, a design system nobody opens can be deleted for good, even while designs that use it are in daily use (Anthropic, 2026).
Name one person. Their job is to keep the rules current, approve changes and open the system now and then.
Write that ownership into your AI policy alongside who may use which tools. Our AI policy template has a place for it.
What to do next
- This week: ask 3 people to share the prompt they use for client-facing work. The differences are your brand drift.
- This month: write one page of brand rules and load it into the tool your team actually uses.
- Within 90 days: set it as the default, name an owner, and check output against it each month.
The rules get people most of the way. Knowing how to brief the tool and judge what comes back is the rest, which our AI training for teams covers. If your team runs on Claude, our Claude training includes setting up design systems and skills.
Brand rules are one part of being ready to use AI well. For the rest, see what AI readiness actually means.
Tired of fixing AI output by hand? We'll help you get your brand into the tools your team already uses.
Frequently asked questions
Why does AI output look generic?
Because the assistant has no context about your business, so it falls back on the most common choices in fonts, layouts and phrasing. Anthropic's own frontend design skill exists to steer Claude away from these generic defaults. Giving the assistant your brand rules is what changes the result.
Will switching AI tools fix off-brand output?
Rarely. Copilot, ChatGPT and Claude all start without knowing your brand. The fix is to store your brand rules where the tool reads them automatically, such as a Claude design system or skill, or ChatGPT project instructions, and to make that the default for the whole team.
What is a brand guidelines skill in Claude?
A skill is a folder of instructions, examples and resources that Claude loads when a task needs it. A brand guidelines skill holds your colours, type, logo use, tone and wording rules, so Claude applies them without anyone pasting them into a chat. Skills work on every Claude plan with code execution on.
Who should own our AI brand rules?
One named person, usually whoever owns the brand or marketing. Their job is to keep the rules current, approve changes, make the approved version the default and make sure it isn't deleted under a retention policy. Without an owner, copies multiply and the brand drifts.
Sources
We read each source on the date shown beside it. These products change often, so check the vendor's page before you rely on a detail.
- Anthropic, Claude Design now stays on brand for daily work (17 June 2026, updated September 2026). Read 5 November 2026.
- Anthropic, Set up your design system in Claude Design. Read 5 November 2026.
- Anthropic, Move from standalone Claude Design to Claude. Read 5 November 2026.
- Anthropic, What are skills?. Read 5 November 2026.
- Anthropic, Frontend Design plugin (claude-plugins-official on GitHub). Read 5 November 2026.
- OpenAI, Projects in ChatGPT. Read 5 November 2026.