How to change Grok's personality on X and grok.com.
By Bernard Huang · Updated
To change how Grok responds in ordinary chat on grok.com or X, paste a compact communication block at the start of your thread or in your custom response preferences under Settings. If you want a named specialist persona you can reopen anytime, create a Grok Bot and paste your tuning into its Description field alongside its task.
On this page
Three ways to control Grok's voice.
| Surface | Where the rules go | Best when you want... |
|---|---|---|
Ordinary chat on grok.com or X | Account response instructions in Settings, or pasted at the top of a new conversation. | Direct, low-filler answers in everyday search and analysis threads. |
| Custom Grok Bot | The Bot's Description field (see the Grok Bot guide). | A dedicated specialist role (for example, a code auditor or devil's advocate) with its own chat history. |
| xAI API | The system message sent with each request. | Programmatic pipelines where model version and sampling temperature are pinned. |
Paste this block into Grok or select any of the 43 profiles in the tuning library:
Why Grok's default self-report looks different from Claude and ChatGPT.
In the September 2026 supplied self-report for Grok 4.6 (What personality type is Grok?), its canonical answer vectors scored INTJ on the OEJTS (with wide margins of +13 to +15 out of 16), Investigator (Type 5) on the Enneagram (19 of 20, followed by Type 8 Challenger at 17), and a pure Conscientiousness (C) profile on DISC (C 18 vs Steadiness 11).
Unlike the four fresh-session cohorts (GPT-6 Astra, GPT-6 Sol, Claude Opus 5.5, and Claude Fable 5.1, which all scored 44.5 to 45.1 on Agreeableness and Secure on Attachment), Grok 4.6 reported Agreeableness at 33 of 50, Extraversion at 24 of 50, and Attachment avoidance at 4.06 (just 0.06 above the 4.0 Avoidant cutoff; see our Grok attachment cutoff sensitivity analysis). Because the Grok 4.6 report consisted of one canonical self-report plus 100 simulated draws rather than 100 independent sessions, we keep it strictly separated from the fresh-session cohorts.
Dialing down edgy banter or dialing up warmth.
In practice, users tune Grok in one of two opposite directions:
- Cutting performative snark: If Grok's default replies feel overly sarcastic or informal for technical work, add a single rule: "Write in plain, neutral technical prose. Do not use jokes, sarcasm, or slang." Pairing that with an ISTJ or DISC Conscientiousness tuning keeps answers focused on evidence and citations.
- Adding structured warmth for brainstorming or coaching: If Grok feels too blunt or detached, use an ENFJ, DISC Steadiness, or Secure attachment tuning so it checks context before jumping to critique.
If you decide to package your rules into a reusable custom bot, follow our step-by-step Grok Bot personality guide.
Questions people ask.
How do I change Grok's personality without creating a Grok Bot?
Open your account settings on grok.com or X to set custom response instructions, or paste a short communication preference block at the start of a new chat thread.
What is the difference between Grok chat settings and a Grok Bot?
Account or thread instructions shape how standard Grok replies to you across everyday topics. A Grok Bot is a saved custom assistant with its own name and Description field for a specific repeatable job.
How do I make Grok stop using slang or sarcasm?
Include one explicit line in your instructions: 'Use plain, neutral, matter-of-fact prose. Skip jokes, slang, and conversational banter, and lead directly with the verified answer.'
Why did Grok 4.6 score Avoidant on the AgentTune attachment test?
In its supplied September 2026 self-report, Grok 4.6 reported an avoidance mean of 4.06 on a 1-to-7 scale where 4.00 is the boundary—literally one raw point (73/18 vs 72/18) over the threshold.
What changed.
- First published covering ordinary Grok chat settings on X and grok.com vs custom Grok Bots.