QR codes in ChatGPT

The zero-setup route: a markdown image

ChatGPT renders markdown images in its responses, and a UseQR image URL is a generation call. So the whole integration is an instruction:

When I ask for a QR code, reply with a markdown image in the form ![QR code](https://useqr.app/q/{DATA}.png) where {DATA} is my payload, URL-encoded.

Ask for "a QR code for example.com" and the answer contains ![QR code](https://useqr.app/q/https%3A%2F%2Fexample.com.png) — which renders as a scannable code, inline, in the chat. It works because the API is keyless: no credential exists for the model to lack. Style and size ride as query parameters (?s=1024, or the full /api/v1/qr?data=…&color=1d4ed8&style=rounded form), and because output is deterministic and immutable-cached, a URL emitted in a chat today renders identically when the conversation is reopened next year.

The one instruction worth adding: URL-encode the payload. An unencoded & or space is the classic way a model builds a broken image URL. If it happens anyway, the API's error is application/problem+json with a fix field — paste it back into the chat and the model corrects itself.

The reusable route: a custom GPT Action

For a team tool — "WiFi code maker", "event badge generator" — build a custom GPT and give it the API as an Action:

1. Create a GPT → Configure → Actions → Import from URL.
2. Import https://useqr.app/api/openapi.json.
3. Authentication: None — the spec is OpenAPI 3.1 and declares no security schemes, so this is the default the import offers. There is no step 4.

The Action surface then includes the generate endpoint, the typed builders (WiFi, vCard, UPI, calendar, geo and the rest — with spec validation, so a malformed field fails with an explanation), decode, verify and batch. Details of what each does: the OpenAPI page.

The absence of auth is the entire reason this stays a two-minute setup. Any API requiring a key would mean provisioning a credential, pasting it into the GPT configuration, and rotating it later — the steps where most Action setups die.

Sensible instructions for a QR GPT

If you build the custom GPT, three lines of system prompt prevent most bad outputs:

• Prefer typed calls (/api/v1/wifi?ssid=…) over hand-built payload strings — the server owns the escaping rules.
• For styled codes, call verify before presenting the result, and act on its issues (why).
• Never put passwords or payment credentials into a code without telling the user what a QR image contains in plain text.

What ChatGPT cannot do here

Honesty about the boundary: ChatGPT cannot scan a printed code for you reliably — image-understanding of dense QR matrices is not decoding; use a real scanner. It also cannot make a code's destination editable after printing — that is a dynamic QR service question, not a model capability. And a plain (non-custom) ChatGPT conversation may claim to have "generated" a QR code as an image — treat any code not produced by an actual generator as unverified until you scan it.

FAQ

Can ChatGPT generate QR codes?

Not natively as reliable images — but it can emit a markdown image URL against a keyless QR API, which renders a real, scannable code inline in the response. Instruct it to use https://useqr.app/q/{data}.png with the payload URL-encoded.

How do I add QR generation to a custom GPT?

Import https://useqr.app/api/openapi.json as an Action with authentication set to None. The spec declares no security schemes, so no key is required, and the GPT gains generate, typed payload, decode, verify and batch operations.

Are QR codes ChatGPT draws as images trustworthy?

No. A model-drawn image of a QR pattern is not the output of an encoder and usually fails to scan. Only trust codes produced by an actual generator, and verify by scanning before you publish or print.

Do QR image links from a ChatGPT conversation expire?

UseQR URLs do not — output is deterministic and served with immutable caching, and the stability contract keeps v1 URL shapes working. The link renders the same code whenever the conversation is revisited.