ChatGPT/prompt
Open with a prompt
Opens ChatGPT with the composer already filled in.
Format
https://chatgpt.com/?q={prompt}Try it
https://chatgpt.com/?q=Explain%20universal%20links%20in%20one%20paragraphOpenWhat you supply
{prompt}The prompt to put in the composer.
URL-encode it. Spaces become %20, and an unencoded
&will silently truncate the prompt at that point.
It fills the box. It does not press send. The prompt lands in the composer and waits for the person to submit it, which is the right default for a link somebody else wrote — but it does mean that a link promising "ask ChatGPT this" is really promising "offer ChatGPT this". Do not build a flow that assumes the answer is already being generated.
Do not carry that assumption to other assistants, either: Grok's ?q= looks
identical and submits on load.
?prompt= works identically; the interface rewrites ?q= into it. ?q= is the older and more
widely-linked form, so it is the one documented here.
It only works on the main chat. Hanging ?q= off a custom GPT — chatgpt.com/g/g-abc123?q=…
— is the obvious next thing to try, and it fails silently: the GPT opens with an empty composer and
the prompt is dropped. Launch context for custom GPTs is an open feature request, not a feature, so
there is no link that will hand one a prompt. Conversation starters, configured on the GPT itself,
are the only mechanism that exists.
&model= is documented in enough places to be worth warning about: passing ?q= alongside it
reportedly drops the model back to the default, so a link that appears to select a model may not
have. Treat the model as unset and let the person choose.
Two variants of this link have pages of their own, because they change what the answer is rather than how it is phrased: with web search, and in a temporary chat.