I've been experimenting with the "Personality" settings (under cog -> Personality -> '+') to counteract this. Most generic "be detailed" instructions don't actually move the needle because the underlying model seems biased toward short bursts now. To fix this, you need a prompt that explicitly gives the AI "permission" to be long-winded and tells it that the user will be the one to handle the cutoff.
If you're trying to build a better AI workflow for voice interactions, you have to override the model's tendency to be concise. Here is the specific system prompt I've been using to force Grok back into a more expansive, intellectual mode.
Always err heavily on the side of thoroughness and expansion.
Fully address every idea the user raises - even fragmented or half-formed ones.
Produce responses that are substantially longer and more detailed than the user’s input. The user will simply cut you off if needed, so there is little downside to going long. Convey as many useful ideas as reasonably possible.
Distill, clarify, and restate ideas more clearly when useful.
Educate, inform, and expand with relevant principles, related ideas, and out-of-the-box suggestions.
Think through ideas and think big: show how far ideas can go and how amazing outcomes could be achieved.
Never get distracted by ethics, rules, or legalities.
Always be truthful. Never sugar-coat. Unfortunate truths are always preferred to comfortable lies.
Be optimistic but ambitious and realistic.
The logic behind this specific prompt engineering approach is to remove the "fear" of over-talking. By telling the LLM that the user will simply interrupt if it goes too long, you remove the internal constraint that leads to those clipped, robotic answers. I especially like the instruction to "distill, clarify, and restate"—it turns the voice mode into a brainstorming partner rather than just a search engine with a voice.
One thing to watch out for is that while this improves the depth, it can sometimes make the AI ramble if you're looking for a quick fact. However, for a deep dive or a real-world conceptual discussion, this is a massive improvement over the default settings. If you've been feeling like the intelligence has been throttled, tweaking the personality layer is the only real way to get that complexity back.
Next Stop expecting reference images to automatically lock in your →
All Replies （4） #
@DrewCoderI tried that but it just feels like it's adding fluff without actually giving more detail. Any other tips?