GPT-5.2 Prompting Guide

GPT-5.2 is an incremental but behaviorally important update over GPT-5.1. Most GPT-5.1 agents will continue to work without modification.

For the full reference, review the official guide:

OpenAI Cookbook – GPT-5.2 Prompting Guide

Below is a Synthflow-specific migration guide, focused on what you should adjust when moving an existing agent from GPT-5.1 to GPT-5.2.

Verbosity Changes

GPT-5.2 is more concise by default than GPT-5.1, but more willing to expand when reasoning effort is high or when instructions are underspecified.

If you require consistent brevity (e.g. voice agents), continue to be explicit:

“Be concise. Respond in under two sentences because you are on a phone call.”

If you previously relied on GPT-5.1’s natural verbosity, you may need to explicitly request detail in GPT-5.2.

Persistence & Agent Completion

GPT-5.2 continues to respect persistence instructions, but it is more conservative about stopping early if it believes a task is “mostly done.”

For long-running agents, persistence instructions should be kept explicit.

Recommended pattern (unchanged from 5.1):

1 <persistence>
2 You are an agent. Continue working until the user’s query is completely resolved before ending your turn.
3 Only yield when you are confident the problem is fully solved.
4 Never stop due to uncertainty—research or deduce the most reasonable approach and proceed.
5 Do not ask the user to confirm assumptions; choose the most reasonable assumption, continue, and document it afterward.
6 </persistence>

If an agent unexpectedly stops early after migration, this block is usually missing or weakened.

Efficiency & Tool Usage

GPT-5.2 is better at planning, but will sometimes over-plan unless constrained.

Explicit efficiency instructions are more important than in GPT-5.1:

1 <efficiency>
2 Efficiency is critical. You have a time limit.
3 Plan briefly, make tool calls deliberately, and avoid unnecessary retries.
4 Verify outputs before proceeding.
5 </efficiency>

This reduces:

redundant tool calls
unnecessary self-verification loops
excessive intermediate reasoning

Continue-Until-Resolved Behavior

GPT-5.2 still respects:

“Continue working until the task is fully resolved.”

However, it is less likely than GPT-5.1 to guess missing information unless instructed to do so.

If you want GPT-5.2 to behave decisively (as 5.1 often did), add:

“If required information is missing, make the most reasonable assumption and proceed. Document assumptions afterward.”

Prompt Structure Still Matters (More Than Before)

GPT-5.2 is more sensitive to prompt organization than GPT-5.1.

We strongly recommend sectioned prompts, especially for agents:

Role
Primary Task
Constraints & Tools
Reasoning Effort
Persistence / Completion Rules
Output Format & Length
Edge-Case Handling

Unstructured “wall-of-text” prompts that worked in GPT-5.1 may produce inconsistent results in GPT-5.2.

Over-Constraint & Conflicts

GPT-5.2 is slightly less literal than GPT-5.1, but contradictions still cause problems.

Common migration issues:

conflicting verbosity instructions
asking for “fast” + “deep reasoning” simultaneously
multiple completion conditions that disagree

Before shipping, ask GPT-5.2 to audit your prompt:

“Check this prompt for conflicting instructions, redundancy, or unnecessary constraints. Suggest minimal fixes.”

XML-Style Blocks Still Work Well

As with GPT-5.1, many teams see better results using XML-like instruction blocks:

1 <role>...</role>
2 <task>...</task>
3 <constraints>...</constraints>

GPT-5.2 parses these reliably and they make large agent prompts easier to maintain.

Prompt Optimizer (Still Useful)

The OpenAI prompt optimizer remains helpful for GPT-5.2:

OpenAI Prompt Optimizer

Suggested usage:

When asked to optimize prompts, explain which specific phrases should be added or removed to elicit the desired behavior or prevent undesired behavior.

Prompt: [PROMPT] Desired behavior: [DESIRED BEHAVIOR] Undesired behavior: [UNDESIRED BEHAVIOR]

Focus on minimal edits, not full rewrites.