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. Read OpenAI’s GPT-5.2 prompting guide for the full reference. This page summarizes what to adjust when moving an existing Synthflow agent from GPT-5.1 to GPT-5.2.

You can change the LLM at any time from General configuration in the agent editor.

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 (for example, 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 and 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):

You are an agent. Continue working until the user's query is completely resolved before ending your turn.
Only yield when you are confident the problem is fully solved.
Never stop due to uncertainty: research or deduce the most reasonable approach and proceed.
Do not ask the user to confirm assumptions; choose the most reasonable assumption, continue, and document it afterward.

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

Efficiency and 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:

Efficiency is critical. You have a time limit.
Plan briefly, make tool calls deliberately, and avoid unnecessary retries.
Verify outputs before proceeding.

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 and tools
• Reasoning effort
• Persistence and completion rules
• Output format and 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 and 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” and “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.”

Labeled sections still work well

As with GPT-5.1, many teams see better results using clearly labeled instruction blocks:

Role: ...
Task: ...
Constraints: ...

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

Prompt optimizer

OpenAI provides a built-in prompt optimizer that remains helpful for GPT-5.2. 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.