Writing output boundaries, ambiguity handling, long-context review, and high-risk self-checks as explicit sections can help GPT-5.2 Chat follow formats more consistently at lower reasoning settings and reduce unsupported assertions.
Suitable tasks: Information extraction, customer service/research Q&A, locating content in long documents, and workflows that require a fixed Markdown or JSON shape.
Unsuitable tasks: Scenarios that require server-enforced schemas, permissions, or fact verification; prompts cannot replace validation and security policies.
Applicable model versions: The GPT-5.2 family; gpt-5.2-chat-latest is a ChatGPT-aligned snapshot, while the official documentation positions gpt-5.2 for more complex reasoning tasks.
Applicable clients, Agents, or APIs: Prompt sections can be reused with Chat Completions; the Responses API is better suited to retaining multi-turn reasoning context.
Recommended reasoning setting and parameters: For fast Chat interactions, prioritize reasoning.effort=none (only for interfaces that support this parameter); use text.verbosity=low|medium|high for output length, subject to support in the current API.
The following are original prompt sections from the OpenAI guide that can be reused directly and combined for research/extraction scenarios; field values can be replaced for the task.
<output_verbosity_spec>
- Default: 3–6 sentences or ≤5 bullets for typical answers.
- For simple “yes/no + short explanation” questions: ≤2 sentences.
- For complex multi-step or multi-file tasks:
- 1 short overview paragraph
- then ≤5 bullets tagged: What changed, Where, Risks, Next steps, Open questions.
- Provide clear and structured responses; use lists, paragraphs and tables when helpful.
- Avoid long narrative paragraphs; prefer compact bullets and short sections.
- Do not rephrase the user’s request unless it changes semantics.
</output_verbosity_spec>
<long_context_handling>
- For inputs longer than ~10k tokens:
- First, produce a short internal outline of the key sections relevant to the request.
- Re-state the user’s constraints explicitly before answering.
- Anchor claims to source sections rather than speaking generically.
- If the answer depends on fine details, quote or paraphrase those details.
</long_context_handling>
<uncertainty_and_ambiguity>
- If the question is ambiguous or underspecified, call this out and:
- Ask up to 1–3 precise clarifying questions, OR
- Present 2–3 plausible interpretations with clearly labeled assumptions.
- When external facts may have changed recently and no tools are available, answer generally and state that details may have changed.
- Never fabricate exact figures, line numbers, or external references.
- When unsure, prefer “Based on the provided context…” instead of absolute claims.
</uncertainty_and_ambiguity>
<high_risk_self_check>
Before finalizing an answer in legal, financial, compliance, or safety-sensitive contexts:
- Re-scan for unstated assumptions.
- Re-scan for numbers or claims not grounded in context.
- Re-scan for overly strong language such as “always” or “guaranteed”.
- Qualify any problem found and state assumptions explicitly.
</high_risk_self_check>Place the sections in the system or developer message, and provide the user input as a separate task.
Using the same long document and the same set of questions about missing fields, test none and medium separately (if the interface supports them).
Record the format compliance rate, whether missing fields are null/not found, whether citations can point back to the original text, and the total output tokens.
For high-risk tasks, independently validate the JSON/schema, permissions, citations, and numerical values on the server side.
The official guide describes GPT-5.2 as a model with stronger structured reasoning, tool grounding, and multimodal capabilities, and notes that it remains sensitive to output shape, tone, and level of detail.
The official examples give output constraints of 3–6 sentences or no more than 5 bullets for typical answers, and a long-context threshold of approximately 10k tokens.
The official migration advice is to keep the prompt unchanged to establish a baseline, then fix reasoning_effort, change only one prompt factor at a time, and rerun the eval each time.
The thresholds and parameters above are recommendations from the guide, not hard guarantees for every client.
This document comes from a guide for the GPT-5.2 family; it cannot establish that every gpt-5.2-chat-latest snapshot has the same reasoning/Responses behavior.
The Chat version is marked deprecated on the official model page; production systems should first verify the currently available snapshot and migration guidance.
“Internal testing/customer feedback” does not provide a complete public dataset, sample size, or variance, and therefore cannot be treated as an independent benchmark.
An “internal outline” is only a behavioral requirement and should not ask the model to reveal hidden chain-of-thought; ask only for brief, verifiable conclusions, citations, and assumptions.
The official guide’s core requirement is: “Never fabricate exact figures, line numbers, or external references when you are uncertain.” (A short excerpt from the original, within 14 words.)
GPT-5.2 Chat