GPT‑5.3 Instant Responses Include `:::writing` Wrapper Around Generated Content

Summary

While testing the new GPT‑5.3 Instant model, generated writing outputs are wrapped in :::writing and ::: markers. This appears to be intended as a structured signal for “copy‑ready” content, but these markers are returned directly in the API response and may surface to end users unless explicitly handled.

Description

When the model is asked to produce a complete piece of writing (for example a blog post, email, or announcement), the response is wrapped with:

:::writing
...content...
:::

The model itself explains that this wrapper is used to mark finished artefacts intended to be copied and used outside the chat. However, this behaviour may be unexpected for API consumers and front‑end applications, as the markers appear as literal text in the response.

Steps to Reproduce

1. Use the GPT‑5.3 Instant model.

2. Prompt the model to generate a complete piece of writing (e.g. “Write a short blog post about X”).

3. Observe that the response content is wrapped in :::writing and :::.

Actual Result

The returned text includes the :::writing wrapper markers directly in the response body.

Expected Result

One of the following behaviours would likely be preferable:

• The markers are stripped from the API response.

• The markers are returned in a structured field (metadata) rather than inline text.

• The platform automatically hides or renders them with specific formatting.

Impact

Applications displaying model output directly to users may unintentionally show these markers, which can appear confusing or like formatting artefacts.

Additional Notes

The model indicates these markers are meant to signal “finished artefacts” such as emails, cover letters, blog posts, announcements, speeches, scripts, or messages intended to be copied verbatim. If this behaviour is intentional, it may be helpful to document it or provide guidance for handling it in client applications.