> ## Documentation Index > Fetch the complete documentation index at: https://invoca-5bd45748-mintlify-17ed87db.mintlify.site/llms.txt > Use this file to discover all available pages before exploring further. # Identifiers: Color > Whether AI-generated or AI-touched content gets a distinguishing color treatment, and why that treatment must not become a new status. **No visual identity has been decided for any Invoca agentic feature yet.** This page proposes a framework for making that decision — it is not the decision itself. See [Coverage, stated honestly](/invoca-design-system/ai-experience/overview#coverage-stated-honestly). ## What it governs Whether content that is AI-generated, or a surface that contains AI-generated content mixed with other content, gets a distinguishing color treatment — a tinted background, a colored border, a highlighted region — so a user can tell a model produced it. It governs the decision to add such a treatment and the rules it must follow if one is added. It does not govern [Foundations: Color](/invoca-design-system/foundations/color) itself — the surface, text, border, and status token model that already exists and that any AI color treatment must be built from, never around. ## Vocabulary | Term | Means | | ---------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------- | | **AI-generated content** | The entire surface's content came from a model — a summary card, a drafted reply | | **AI-touched content** | Some AI content sits alongside human- or system-authored content on the same surface — one comment in a thread is AI-drafted, the rest are not | | **Distinguishing treatment** | A color signal meaning "this is AI" — an identity signal | **"Distinguishing treatment" is not a status.** Titan's four statuses — `info`, `success`, `warning`, `error` — answer "what happened." An AI color treatment answers "who made this." Conflating the two means an AI answer that is also an error has nowhere to put the error. ## Choosing a value | If you need to… | Use | Not this, because | | ------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Mark AI-generated text as visually distinct from human-authored text on the same page | An existing informational surface and border pairing from [Foundations: Color](/invoca-design-system/foundations/color) — for example `background-info-alt` with `border-default` — plus a persistent icon or label | Inventing a new hue for "AI." Titan's status model already rejects a fifth status color for the same reason: it cannot be reliably distinguished, has no dark-mode value, and no contrast guarantee | | Signal that a summary is uncertain | The [Trust builders](/invoca-design-system/ai-experience/trust-builders/caveat) Caveat pattern | Recoloring the surface based on confidence — that is a different signal answering a different question, and layering it onto the identity color makes both illegible | | Show a streaming or working state | A [Governors](/invoca-design-system/ai-experience/governors/stream-of-thought) treatment | Assigning the identity color a fixed meaning for "in progress" too — state color belongs to interaction states, not to identity | | Render AI output before any treatment has been decided | Titan's ordinary component tokens, no special tint | A default tint chosen ad hoc for this page — an undecided treatment presented as if it were decided | ## Reference Nothing is decided or shipped. This section will be generated from a real decision once one exists. ## Relationships An AI-touched surface still uses [Foundations: Color](/invoca-design-system/foundations/color)'s status and prominence tokens — it does not invent a new hue. If a treatment is eventually decided, it is decided as a specific pairing from the existing token set (a background token, a border token, an `-alt` variant where text is involved), documented the same way Color's own ["Choosing a token"](/invoca-design-system/foundations/color#choosing-a-token) table documents every other pairing. It also sits beside [Identifiers: Iconography](/invoca-design-system/ai-experience/identifiers/iconography): per `TITAN-COLOR-03`, color never carries meaning alone, so any color treatment this page eventually adopts needs an icon or text partner from that page or from [Identifiers: Name](/invoca-design-system/ai-experience/identifiers/name). ## Accessibility * Color is never the only signal that content is AI-generated — this is [Foundations: Color](/invoca-design-system/foundations/color)'s `TITAN-COLOR-03` applied to a new case, not a new rule. Pair any treatment with text or an icon. * If the treatment carries text (a tinted card with a summary inside it), that text meets WCAG 2.2 AA against the surface it sits on, exactly as `TITAN-COLOR-02` already requires for every other surface. * A treatment that reads correctly in light mode but has no stated dark-mode value is not a complete decision — see [Gaps](#gaps). ## Constraints | ID | Constraint | Rationale | | -------------------- | ------------------------------------------------------------------------------------------------------------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **TITAN-AICOLOR-01** | An AI color treatment reuses an existing semantic or component token. It never introduces a new hue or literal for "AI." | Mirrors `TITAN-COLOR-04` and `TITAN-COLOR-05` — a new meaning met by a new hex is exactly what Titan's status model forbids, and an "AI purple" carries the same distinguishability problem at higher stakes. | | **TITAN-AICOLOR-02** | Color is never the only signal that content is AI-generated. Pair with text or an icon. | Restates `TITAN-COLOR-03` for this case: color vision deficiency and forced-colors contexts remove color as a channel entirely. | | **TITAN-AICOLOR-03** | An AI color treatment never reuses a status color (`info`, `success`, `warning`, `error`) to mean "this is AI." | The four statuses are load-bearing everywhere else in the product. If an informational tint also means "AI wrote this," the two meanings collide the first time an AI answer needs to show a real status too. | | **TITAN-AICOLOR-04** | Where no AI color treatment has been decided for a feature, its output renders with Titan's ordinary component tokens and no special tint. | An undecided treatment should render as undecided, not as a silent default that reads as a choice. | ## Divergences Not applicable. Nothing is shipped yet to diverge from. ## Gaps * Whether AI content needs a color signal at all, or whether text and the Name identifier are sufficient on their own, is undecided. * Whether a shared treatment applies to all of Invoca's agentic surfaces or each needs its own is the same per-feature-or-once question flagged on the [overview](/invoca-design-system/ai-experience/overview#gaps). * No treatment has a stated dark-mode or forced-colors value, since none has been decided at all yet. ## Related [Identifiers: Avatar](/invoca-design-system/ai-experience/identifiers/avatar) · [Identifiers: Iconography](/invoca-design-system/ai-experience/identifiers/iconography) · [Identifiers: Name](/invoca-design-system/ai-experience/identifiers/name) · [Identifiers: Personality](/invoca-design-system/ai-experience/identifiers/personality) · [AI Experience overview](/invoca-design-system/ai-experience/overview) · [Foundations: Color](/invoca-design-system/foundations/color) ## Why it works this way Color is the fastest channel a person reads, and identity and status both want to use it first. Titan already resolved that competition once, by capping status at four colors and requiring every one of them to pair with an icon or label rather than stand alone. An AI color treatment that skipped this and invented its own hue would relearn that lesson from scratch, with the added risk that "AI" and "error" are exactly the two meanings most likely to need to render on the same card at once. Keeping the treatment inside the existing token set — rather than granting Identifiers its own palette — is the same discipline `TITAN-FND-04` applies everywhere else: a new need is met by a proposal, never by a local override.