How to specify what information must appear in the reasoning summaries?

API
,
1

With GPT-5.2, is it possible to influence the content of the reasoning summaries returned by the model?

I am aware that:

  • reasoning summaries verbosity can be set via the summary parameter (I use `auto`),

  • the reasoning effort of internal can be set via effort(I use `low`).

However, my goal is more specific:
I would like to force certain information (tool preambles/ justifications) to appear in the reasoning summary itself, rather than in the assistant’s normal output (plain text or JSON-structured responses).

Concretely:

  • Is there any supported way to specify what types of information must appear in the reasoning summary?

  • Can the reasoning summary be shaped or constrained (e.g., via prompts, system instructions, or structured output)?

  • Or is the reasoning summary entirely system-generated and non-configurable beyond the existing summary / effort parameters?

1 Like
2

In GPT-5.2, reasoning summaries are generated automatically by the system to reflect the model’s internal thought process and key considerations during a task. Currently, the only configurable aspects are the summary parameter, which controls verbosity, and the effort parameter, which influences the depth of the internal reasoning. Beyond these, there is no officially supported mechanism to mandate the inclusion of specific information such as tool preambles or justifications directly within the reasoning summary. The summary cannot be shaped or constrained via prompts, system instructions, or structured output directives; it remains largely system-determined. In practice, if particular details need to appear, the recommended approach is to guide the model to include them in the main output in plain text or structured JSON rather than relying on the reasoning summary itself.

1 Like