PTN—06

AI identity

The AI’s name, avatar, and visual treatment as an expectation-setting system — recognizably synthetic, consistent everywhere, never a fake human.

Before the AI says a word, its identity has already made claims. The name, the avatar, the color of its messages — all of it tells the person what they are dealing with and how much to trust it. Design that identity as a system and it sets expectations honestly; leave it to whoever picked the avatar last and it sets them anyway, just wrong. The stock-photo face with a human first name promises empathy and judgment the system does not have, and every subsequent failure is measured against that promise.

The first job of the identity is distinction: AI content must be recognizable as AI content at a glance, before reading a label. A distinct avatar shape, a different message treatment, ghost text instead of typed text — the channel itself carries the disclosure, so the person’s calibration is never a paragraph behind their reading. Self-identification says “I am an AI” once; identity says it continuously, ambiently, in every pixel the system touches.

The second job is consistency. The same AI appearing as a sparkle icon in one surface, a chat bubble in another, and an unnamed “assistant” in a third fragments the person’s mental model — they cannot accumulate trust in a thing they cannot recognize across encounters. One identity, one treatment, everywhere the system speaks: in the composer, in the sidebar, in the email it drafted, in the notification it triggered. Recognition is what lets a relationship form at all.

And the identity must refuse human mimicry, structurally. No fabricated portrait, no fake typing indicator performing effort that is not happening, no first-person anecdotes about a life the system does not have. This is not aesthetic austerity — synthetic identity can be warm, characterful, even beloved. It is honesty about category: the person should never spend attention wondering whether they are talking to a machine, because the design answered before they asked. The introduction completes the contract in one line — what this thing is and what it is for — stated plainly, without apology or theater.

Anatomy

Fig. 01 — AI identityLive specimen

Assistant · identity system

Titan AssistantAI

I’m Titan Assistant — I draft, summarize, and schedule from your shop’s data. I can be wrong, and you approve anything that sends.

Which memberships lapse this month?

Seven Gold memberships lapse in July — $1,323 in renewals. Want the list?

The same mark on every surface it speaks from

ChatInline suggestEmail draft

A geometric mark, not a portrait — and no theatrical typing dots. Replies stream as they’re made.

An AI identity doing its job: visually distinct from human content, consistent across surfaces, honest about what it is.

1Distinct at a glance. AI content reads as AI before any label is read — the visual treatment itself is the disclosure.
2Consistent everywhere. One name, one mark, one treatment on every surface the system speaks from, so trust can accumulate.
3No human mimicry. No fabricated portrait, no fake typing indicator — nothing that performs a humanity the system lacks.
4Honest introduction. One line stating what it is and what it can do — capability named plainly, without theater.

When to use

AI output appears alongside human content — chat, comments, documents, support threads.
The system speaks on multiple surfaces and needs to be recognizable as one thing across all of them.
A conversational product is being given a name, an avatar, or a personality at all.
Brand pressure is pushing toward a human persona and the case for honesty needs a design answer.

Design considerations

01Design the avatar as recognizably synthetic — a mark, not a face; abstract geometry ages better than a mascot pretending to be a colleague.
02Give AI messages their own visual channel — background, border, or type treatment — so origin is legible before the first word.
03Audit every surface the AI touches and unify the identity across them; the sidebar, the email, and the notification are the same speaker.
04Strip the mimicry props: typing indicators that fake keystrokes, “thinking” theater unmoored from real work, first-person biography.
05Write the introduction as one honest line of capability — what it does and for whom — not a personality performance.
06Personality is permitted; deception is not — character should express itself in voice and craft, never in a claim to be human.

Pitfalls

✕The fake colleague: human name, stock-photo portrait, “typing…” indicator — a stack of small lies that converts the first failure into a betrayal.
✕Identity fragmentation — three names and four treatments across the product, so users cannot form a model of what the system is or how far to trust it.
✕Overcorrecting into sterility: an identity so anonymous and apologetic that people cannot tell the product has a capable system behind it at all.
✕The grandiose introduction — an identity that opens by overclaiming, spending in one line the credibility the visual honesty earned.

In the wild

Claude: One name, one abstract starburst mark, and the same identity across app, API, and integrations — recognizably synthetic and consistent everywhere it speaks.
GitHub Copilot: Machine-authored code renders as ghost text and Copilot speaks under its own mark in chat and pull requests — the treatment separates its work from yours at a glance.
Intercom Fin: Presents as a named AI agent with a non-human avatar inside human support threads, so the bot-to-human handoff is visible as a change of speaker.