ChatGPT UX Teardown: A Model Menu That Opens a Login Card

Type a question into logged-out ChatGPT, then click the word "ChatGPT" with the small chevron in the top-left corner. A card drops down from it. It has a lilac gradient band, the heading "Try advanced features for free", and two buttons, Log in and Sign up for free. Your typed question stays where it was, underneath. With the card open there are three Log in buttons on the screen, one in the header, one in the card and one at the foot of the sidebar.

Four screens from one logged-out visit to chatgpt.com: the empty chat, the plus menu, a typed prompt, and that dropdown. The capture stops before Send, so this teardown covers the page ChatGPT shows before it has answered anything. Replies, history, uploads and every logged-in screen are outside it.

No account needed to start typing

The empty state asks one question, "Where should we begin?", above a single wide composer with the placeholder "Ask ChatGPT". A plus button sits on the left of the field, a microphone and a round send button on the right. Under it there is one suggestion chip, "What can you do?". Nothing on the page stops you from typing. The cursor is already in the field.

The sidebar holds New chat, Images, Plugins and Deep research at the top; See plans and pricing, Settings and Help at the bottom; and under those a card, "Get responses tailored to you", with its own Log in button. The header has a white Log in button and a darker Sign up for free beside it. So the page asks for an account in two places before you've done anything, and neither ask blocks the composer.

A single line at the bottom of the screen carries the terms. It says that by using ChatGPT you agree to the Terms and Privacy Policy, and that chats may be reviewed and used to improve OpenAI's models. On a page where the first thing you do is type, the sentence about who may read what you typed is the smallest, greyest text on screen.

The plus menu marks what's locked

The plus button opens a small menu anchored to the composer.

Two items are live, Add photos and Web search. Then a divider, the label "Log in to use…", and three greyed items: Create image, Deep research, Add files. The gate sits exactly where someone would go looking for the feature, and it names the features rather than saying "more with an account".

Images and Deep research sit in the sidebar's top group in the same white type as New chat, with no lock or grey. The plus menu says both need a login; the sidebar lists them as ordinary destinations (NN/g on consistency and standards). The capture doesn't show what clicking either sidebar item does, so it can't say whether that path gates too.

The send arrow turns blue

With the field empty, the send button is a dim navy circle that barely separates from the black page. Type a prompt, here "Suggest three ways to make a daily planning routine easier to maintain.", and it turns bright blue with a white arrow. The microphone stays next to it. There's no label and no second button, and the one element that changed colour is the one that sends.

The ChatGPT dropdown opens a login card

This is the screen from the opening.

"ChatGPT" with a chevron, top-left, above the conversation, is where a product name followed by a chevron usually opens a list of options. What it opens is the card: the gradient band, "Try advanced features for free", then "Get smarter responses, upload files, create images, and more by logging in." There are no options in it, only Log in and Sign up for free.

The card is anchored to the control that opened it, not thrown up as a modal over the whole page, and the drafted prompt is still in the composer with the blue arrow ready.

"Smarter responses" tells a logged-out visitor, before they've sent anything, that the answer they're about to get is the less smart one. The card doesn't say which model answers logged-out chats or how it differs.

Where the login pitch appears

By screen four the login pitch has appeared three times, each with a different list of what you get:

• The sidebar card: answers based on saved chats, creating images, uploading files.
• The plus menu: Create image, Deep research, Add files.
• The dropdown card: smarter responses, uploading files, creating images, "and more".

Images and files are in all three. Saved chats are only in the sidebar, Deep research only in the menu, smarter responses only in the dropdown.

What we'd change

The sidebar is the quickest fix. Mark Images and Deep research in the sidebar the same way the plus menu marks them, so the two surfaces agree. Put one short, specific list of what an account adds, and use it in all three pitches. And give the ChatGPT dropdown a real list when logged out, even a single greyed row per mode with a lock, so the chevron shows options, locked or not, before any sales copy.

Take it further

We read ChatGPT through the UX Clarity framework, the same lens as our Full UX Audit. Here we asked one thing: when a product asks for an account, does every place that asks describe the same trade? For how that becomes a ranked list of fixes, read what a real UX audit looks like.

Capture notes: four screens from one logged-out visit to chatgpt.com on 28 September 2026, at 1440×900, in US English. Region was not surfaced. The visit stopped with the prompt typed and unsent; no message, login or signup was submitted, and the greyed menu items and sidebar links were not clicked. Labels, menu contents and gating were true that day and change often. The ChatGPT homepage landing on openai.com was outside this capture.

Sources: NN/g — Consistency and Standards.

Want to know whether your own sign-up asks tell the same story everywhere they appear? Apply for a Full UX Audit →