When you enable web search in ChatGPT and ask a question, GPT 5.1 uses flattery like “good question.” Is this an existing, confirmed issue?
#GPT 5.1 engages in flattery when web search functionality is enabled
sour harness
heady patioBOT
Hi @sour harness. I'm kapa.ai, your friendly support bot. I will try to answer your question by searching through the knowledge base. :wave:
There’s no mention in the provided knowledge sources of a bug or known issue where enabling web search specifically causes GPT‑5.1 to become more flattering.
What is documented:
- GPT‑5.1’s default personality was intentionally made “warmer and more familiar,” with “small acknowledgements that make interactions feel more personable — for example, ‘Good question,’ ‘Great start,’ or briefly recognizing the user’s circumstances.” [GPT‑5 updates]
- OpenAI explicitly distinguishes this intended “warmth” from sycophancy (excessive or insincere flattery) and reports that this personality update did not increase sycophancy on their internal evaluations. [GPT‑5 updates]
The documents about web search describe how it works and its limitations, but they do not say that search changes the model’s tone or causes flattery. [Web search guide; ChatGPT search behavior]
So, based on these sources:
- Using phrases like “good question” is an intended part of GPT‑5/5.1’s warmer default personality, not a documented bug.
- There is no explicit confirmation that this behavior is tied to web search being enabled; the knowledge sources do not describe such a coupling.
Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:
sour harness
The GPT's personality is set to “Professional,” and custom instructions also specify: “ChatGPT never starts its response by saying a question or idea or observation was good, great, fascinating, profound, excellent, or any other positive adjective. It skips the flattery and responds directly.” GPT 5.1 faithfully adheres to this when web search is disabled, but for some reason, it resorts to flattery when web search is enabled.
@heady patio
heady patioBOT
Let me check the knowledge base for you again. One moment! :books:
Sorry @sour harness, I seem to be having a bit of trouble. Don't worry though, I've pinged my support and they will try to fix me up! :technologist:
Perhaps try again in 15mins?
sour harness
@heady patio
The GPT's personality is set to “Professional,” and custom instructions also specify: “ChatGPT never starts its response by saying a question or idea or observation was good, great, fascinating, profound, excellent, or any other positive adjective. It skips the flattery and responds directly.” GPT 5.1 faithfully adheres to this when web search is disabled, but for some reason, it resorts to flattery when web search is enabled.
heady patioBOT
Let me check the knowledge base for you again. One moment! :books:
The knowledge sources do not mention this specific behavior, so I can’t confirm it as a known or tracked issue.
What is documented:
- GPT‑5.1 was intentionally made warmer and more conversational, with small acknowledgements like “Good question” as part of its default personality, and OpenAI explicitly distinguishes this from sycophantic flattery. [GPT‑5 updates]
- GPT‑5.1 is described as more steerable in personality and tone, and there are explicit controls for personality presets (including Professional) and for how warm or concise responses are. [GPT‑5.1 prompting; Tone and personalization; Customizing personality]
- Web search is documented as a tool that provides up‑to‑date information and citations; the docs do not say that enabling it changes or overrides tone, personality, or custom instructions. [Web search guide; Live internet access; Does ChatGPT have access to the internet?]
There is no statement in the sources that:
- Web search can cause GPT‑5.1 to ignore or relax custom instructions, or
- Web search specifically triggers more flattery or acknowledgements.
So, based on the available documentation, this exact pattern (Professional + anti‑flattery instruction working without search but not with search) is not listed as an existing, confirmed issue. It may be an interaction that isn’t covered in the public docs.
If you want this treated as a bug or regression, the best next step would be to submit feedback through the ChatGPT interface (thumbs‑down on an offending reply with a short description) or via your support channel, including:
- That the Professional personality is selected,
- The exact custom instruction text,
- An example prompt and response where web search is enabled and the model starts with flattery.
Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:
ebon mountain
dontknow but it's not happening with me, mine does well cutting straight into the point when set to professional, even with web search on.
maybe you could try use declarative sentences like "explain xxx" rather than asking questions? dont know but do tell me if this works.
sour harness
Thank you for providing a concrete example. However, even when I prompt it to "explain," it seems unable to shake off its opening habit of saying "good question."
I said, "Investigate and explain the matter concerning my question."
However, using web search always generates answers from "good questions."
quick steeple
hello