Colour and Typography: Brand Guidelines for AI Products
By Ramesh Emandi, Co-Founder and CTO · Last updated · 9 min read

In short: Good colour and typography make a brand recognisable and a product usable. For AI interfaces and enterprise products, start with a modular type scale, two typefaces plus a monospace, and semantic colour tokens. Meet WCAG 2.2 AA contrast: 4.5:1 for body text, 3:1 for large text and interface components. Then design for streaming text, code and confidence states.
Most advice on fonts and colours was written for logos and marketing sites. Enterprise software and AI products put far more pressure on type and colour: dense data, long sessions, dark mode, text that appears word by word, and agents that act on a user's behalf. This guide sets out practical rules we use when designing interfaces, with exact numbers and primary sources.
Why do typography and colour matter more in AI and enterprise products?
In a marketing site, type and colour mostly carry personality. In a product, they carry meaning. They tell people what is a heading, what is clickable, what failed, what the system is unsure about and what an agent is about to do. When those signals are inconsistent, people lose trust, and in AI products, trust is the whole game.
There is also a new reader. AI agents increasingly read and operate interfaces, and coding agents build them. A colour and type system defined as named, machine-readable tokens is easier for both people and agents to apply correctly. At Quantana we call this AX, or agent experience, and we design for it alongside UI and UX.
How do you build a type scale for a product interface?
A type scale is a small, fixed set of sizes generated from a base size and a ratio. It stops "just one more size" creeping in and gives every screen the same rhythm.
- Base size: 16px for product UI body text, 17 to 18px for long-form reading. Set sizes in
remso they respect the user's browser setting. - Ratio: use a tighter ratio such as 1.2 (minor third) for dense dashboards and tools, and a wider one such as 1.25 (major third) or 1.333 (perfect fourth) for marketing pages, where headings need more drama.
- Example scale at 1.2 from 16px: 11, 13, 16, 19, 23, 28, 33px. Round to whole pixels, and cap the number of steps at six or seven.
- Line height: about 1.5 for body text and 1.1 to 1.3 for headings. Long line lengths need more line height.
- Line length: aim for roughly 45 to 75 characters per line for reading text, including AI chat responses.
- Weights: two or three weights is plenty (for example 400, 500 and 600). Use weight and size for hierarchy before you reach for colour.
Your scale must also survive user adjustments. WCAG 2.2 requires that no content is lost when text is resized to 200% (success criterion 1.4.4) and when users override spacing to a line height of 1.5 times the font size, paragraph spacing of 2 times, letter spacing of 0.12 times and word spacing of 0.16 times (1.4.12). [1]
What are the rules for pairing fonts in brand guidelines?
- Two families, plus a monospace. One for headings, one for text and interface, or a single versatile family for both. AI and developer products also need a monospace for code, IDs and logs.
- Pair by contrast of structure, not by mood. A geometric sans with a humanist serif works because they differ clearly. Two similar sans serifs look like a mistake.
- Match x-heights. Faces with similar x-heights sit comfortably on the same line, which matters when a monospace appears inside a sentence.
- Check the details that products need: tabular (fixed-width) numerals for tables and metrics, clearly distinct I, l and 1 and O and 0, and a wide enough character set for the languages you support.
- Prefer variable fonts where available. One file covers many weights, which helps performance.
- Write licensing into the guidelines. State where each font may be used: web, desktop, app embedding, and documents. Brand guidelines that ignore licensing cause problems later.
Good brand guidelines for fonts specify the families, the scale, weights, line heights, fallback stacks, and when to use each style, with examples of correct and incorrect use.
What contrast ratios does WCAG 2.2 AA require?
WCAG 2.2 Level AA sets these minimum contrast ratios: [1] [2]
- Body text (1.4.3): at least 4.5:1 against its background.
- Large text (1.4.3): at least 3:1. Large means at least 18pt (about 24px) regular, or 14pt (about 18.5px) bold.
- Interface components and graphics (1.4.11): at least 3:1 for the visual boundaries and states of controls, such as input borders, toggles and focus indicators, and for parts of charts and icons needed to understand them.
- Enhanced, Level AAA (1.4.6): 7:1 for body text and 4.5:1 for large text. Worth targeting for long reading sessions.
Two details catch teams out. First, ratios are not rounded: 4.499:1 fails. A mid grey of #767676 on white is 4.54:1 and passes, while #777777 is 4.48:1 and fails. [2] Second, the exceptions are narrow. Disabled controls, pure decoration and logotypes are exempt, but placeholder text that users need to read is not. A common light grey such as #9CA3AF on white is only 2.54:1, which fails for any text.
Colour must also never be the only signal (1.4.1). A red border on an invalid field needs an icon or message as well. WCAG 2.2 added a new AA criterion, Focus Not Obscured (2.4.11), which means sticky headers, chat composers and cookie banners must not completely hide the focused element. [1]
Should you use APCA instead of WCAG contrast ratios?
Use WCAG 2.2 for compliance, and consider APCA as an extra design check. APCA (the Accessible Perceptual Contrast Algorithm) is the candidate contrast method for the future WCAG 3. It scores contrast as a lightness contrast value (Lc) that accounts for font size and weight, and its authors say it predicts readability better, particularly in dark mode. [3]
APCA's basic guidance is: Lc 90 preferred for body text, Lc 75 minimum for body text, Lc 60 minimum for other readable content text, Lc 45 minimum for large headlines, Lc 30 absolute minimum for placeholder or disabled text, and Lc 15 minimum for non-text elements such as dividers. [3] WCAG 3 is still a draft, so APCA scores do not make a product conformant today. Designing to pass both is the safest approach.
How should you structure a colour system with design tokens?
Design tokens are named values for colour, type and spacing that design tools and code share. The W3C Design Tokens Community Group published the first stable version of its format specification, 2025.10, in October 2025, with a dedicated colour module. [4] A three-tier structure works well:
- Primitive tokens hold the raw palette:
blue-600,grey-900. Build each hue as a tonal ramp from light to dark. - Semantic tokens describe purpose:
color.text.primary,color.text.muted,color.surface.raised,color.border.focus,color.feedback.danger. Components only ever use these. - Component tokens handle the exceptions:
button.primary.background.
Semantic tokens are where accessibility lives. Test every foreground and background pair that the system allows, in light and dark modes, and record the ratio next to the token. Then a designer, developer or coding agent picking color.text.muted on color.surface.raised gets a pairing that is already known to pass. Keep brand colour mostly for actions and key moments; enterprise products work best with calm neutrals and a small set of strong semantic colours.
How do AI chat and agent interfaces change typography?
AI interfaces introduce states that traditional brand guidelines never covered. These are the ones we design for first.
Streaming text
- Keep the response container stable as text arrives. Avoid layout jumps, auto-scroll that fights the reader, and per-word fade animations that hurt readability.
- Render Markdown progressively but conservatively, so half-finished lists and tables do not flicker between styles.
- Announce progress to screen readers without reading every token. Status messages must be available to assistive technology without moving focus (WCAG 4.1.3), so announce "Generating" and "Response ready" rather than the stream itself. [1]
- Respect reduced-motion preferences for typing indicators and cursors.
Code and structured output
- Set code in your monospace, slightly smaller than body text, with a distinct background that still passes 4.5:1 for every syntax colour.
- Show diffs with symbols (+ and -) as well as green and red, to satisfy the use-of-colour rule.
- Use tabular numerals for costs, token counts, latency and other figures that users compare.
Confidence, provenance and agent actions
- Distinguish user messages, model responses, tool output and citations through layout and labels, not colour alone.
- Show uncertainty and sources in words, such as "Unverified" or "From 3 sources", supported by an icon and a semantic colour token.
- Give agent states their own tokens and styles: thinking, running a tool, waiting for approval, completed and failed. An approval request is a high-stakes moment, so it deserves the strongest hierarchy on screen, with the action, its target and its consequences set out in plain text.
What should your brand guidelines say about fonts and colour?
A practical checklist for a product-ready brand system:
- Typefaces, weights, fallback stacks and licence terms.
- The type scale, line heights and line-length targets, in
rem. - Primitive, semantic and component colour tokens, in a machine-readable format.
- Approved foreground and background pairs with their contrast ratios, for light and dark modes.
- Rules for states: hover, focus, disabled, error, success, warning and loading.
- AI-specific patterns: streaming responses, code, citations, confidence and agent approvals.
- Do and don't examples, including how the brand behaves inside a dense data table.
We bring design thinking to these decisions and design across UI, UX and AX, so the system works for the people using it and for the agents working alongside them. Talk to us about your design system, or get an AI roadmap.
Frequently asked questions
How many fonts should a brand use?
Two is usually enough: one for headings and one for body and interface text, or a single versatile family for both. Software and AI products should add a monospace for code and data. More than that tends to dilute the brand and slow pages down.
What is the WCAG 2.2 AA contrast ratio for text?
At least 4.5:1 for normal text and 3:1 for large text, meaning at least 24px regular or about 18.5px bold. Interface components and meaningful graphics need at least 3:1. Ratios are not rounded, so 4.49:1 fails. [1] [2]
Is APCA required for accessibility compliance?
No. APCA is the candidate contrast method for the future WCAG 3, which is still a draft. Today's legal and procurement benchmarks reference WCAG 2.x contrast ratios. Many teams use APCA as an extra check, especially for dark mode and thin fonts. [3]
What is a good type scale ratio for a dashboard?
A tighter ratio such as 1.2 suits dense dashboards and enterprise tools, because it keeps headings close to body size and saves space. Marketing pages can use 1.25 to 1.333 for more contrast between headings and body text.
How should colour be used in AI chat interfaces?
Use calm neutrals for conversation, semantic colours for states such as errors, warnings and approvals, and never rely on colour alone. Pair every colour signal with a label or icon, and keep code, citations and agent actions visually distinct through layout and type.
What are design tokens?
Design tokens are named, reusable values for colour, typography and spacing, such as color.text.primary , shared between design tools and code. The W3C Design Tokens Community Group published a stable format specification in October 2025. [4]





