Reality check · 6 min read
Grokbot's limits — the honest edges of a very good tool
This isn't a takedown. Grokbot is one of the most capable assistants you can buy, and for a lot of people it does the job of several tools at once. But every tool has edges, and knowing where they are is the difference between trusting it well and trusting it badly.
1. Usage caps, not capability caps
The free tier is genuinely usable, but it runs out. Almost everything you're buying as you climb the plans is headroom — more messages, more generation, priority when the service is busy. The model doesn't get smarter between the mid tiers; you just get more of it. That's an unusually honest pricing structure, and it means the right question is "how much do I use it?" rather than "how good do I need it to be?"
2. Confidence outruns accuracy on details
Its blunt register is a real advantage — you get an answer instead of a hedge. The cost is that a wrong answer sounds exactly like a right one. It's at its best quoting something it just pulled from the live web, and at its weakest reconstructing precise numbers, dates and niche specifics from memory. Ask for the working, not the conclusion.
3. Long, careful sessions still favour a specialist
For extended writing and large refactors, a slower and more deliberate assistant still holds a thread better. Grokbot gets you to a strong draft faster than almost anything; the last 20% of a very long piece of work is where people tend to switch. See Grokbot vs Claude for where that line actually falls.
4. It knows your public self, not your private one
If you post, it starts with a real head start on your interests and your voice. But your public persona isn't your life. It doesn't know the project you haven't announced, the decision you made in a private document, or the constraint that shapes everything you're doing this quarter — unless you tell it again, every time.
5. The limit that never improves
Rate limits rise. Models get better. Refusals get tuned. Every item above is on a trajectory. The one that isn't: whatever Grokbot learns about you stays with xAI. Change plans, change assistants, or watch a model you'd shaped yourself around get retired, and the accumulated version of you doesn't come along.
The better Grokbot gets, the more that matters — because the more of your thinking runs through it, the more there is to lose. Here's what's actually happening with its memory.