Keep the Grok thread
getting started grok · Step 5
Two weeks later I couldnt find the chat where we finally got the change fee right. I found four chats named some auto title about United, one of them was a joke thread, and I re-asked the question like I had never done the work.
That is a waste. The point of a good Grok week is not a pile of clever replies. It is one thread you can reopen when the same class of problem comes back.
If you are arriving from a scare, when Grok is wrong is the previous step. The start of the series is getting started.
What is worth keeping
Not the whole performance. The locked facts. The URLs that actually contained the claim. The version of the email you sent, not the second draft you almost sent. The one-sentence problem, because next time you will forget what you were even deciding.
I copy those into a note I control. A running file, a phone note, whatever you will open when you are annoyed again. Grok has history, on grok.com it is the history control in the corner, on iOS they say you tap your profile picture. History is a library. It is a bad filing cabinet. Titles drift. You will not search for the right phrase.
I also paste a short "what I believe now" block back into the thread itself, at the end, in my words.
As of Aug 29 2026 I believe same-day change on this fare is $0 if I do it in the app, I am not relying on a travel blog for that, page is this URL, I sent the email at 9:40pm, I did not mention the joke about Sunday.
Future-me can scroll to that and skip the bit. Future Grok can be told "start from the block I wrote at the bottom, do not re-litigate 2024."
They advertise memory across chats on the public page. Treat it as a bonus, not a filing system. I have had a preference stick. I have also had a new chat act like we never met. If a fact matters, write it where you can see it.
How I name and reuse a thread
One problem, one chat, named by me the moment it becomes useful. "United return change Aug 2026" not "New conversation." If the product comparison starts in the same box I cut it out. I have said this already and I still mess it up when I am on the couch.
When the same problem returns, I do not start clean out of superstition. I open the old thread and paste the new wrinkle. "Same fare class, they emailed me a different number, here is the email, does the page we used still match." That is a cheaper question than rebuilding the context. If the old thread is polluted with jokes and a tablet debate, I start new and paste the locked block only.
Share links are handy for a coworker who wants the trail. On grok.com they say Share is in the top right, or the share icon under the conversation. Remember it is a public link if you make it one. I revoke leftovers at grok.com/share-links when I remember. I do not share a thread that has a record locator in it. I make a clean recap and share that.
Private chat is the opposite tool. Ghost icon, conversation not sitting in your history, gone from their systems after a delay. Use it for the question you do not want in the sidebar. Do not use it for the question you will need on Thursday. I have done that. Thursday me was irritated.
If you connect an X account later, do it on purpose in Settings, Account. Same person on the phone and the laptop. A split login looks like amnesia.
What I pin in the note, really
Four lines, not a system.
The problem in one sentence, from you dont need another tool energy. "I need to know if Tuesday is free to change to."
The source that survived a click. Full URL. Date I opened it.
The output that left the house. Email, post, the two-line decision.
The failure if there was one. "Grok cited a 2025 roundup, I almost used the number." That line trains you faster than a prompt library.
I do not pin prompts that sound clever. I pin prompts that produced a checkable answer. "Search this month, primary pages only, facts then take" is a keeper. "Act as a world-class travel agent" is not. I have never been served by a world-class travel agent in a chat box.
Custom instructions exist if you want a standing tone. I keep mine short when I use them. No jokes in drafts I will send. Ask me for a URL on any number. Do not call me chief. That is enough. A novel in the instruction box and every chat sounds like a brand.
When I jump
I jump to Cursor when the next action is in files. A script, a site, a repo, a thing that has to run. Grok can sketch the function. Cursor can see the rest of the project and I can read a diff. If I catch myself pasting file after file into grok.com I am already late.
I jump to Claude when the next action is a long sit with messy notes and a rewrite I will put my name on. Landlord email that has to stay kind. A page that has to stay accurate for a month, not an hour. Grok can start that. Claude is where I go when I am tired of the wit and I want the boring middle.
I stay in Grok when the clock on the information is the point. What changed. What are people saying today. Which of two public pages is less stale. Then I write the short thing in the same thread and I leave.
Which tool when is the fuller version of that overlap. There is no clean map. There is the problem sentence and whichever tab will not invent a clause.
I do not jump because a model is "better." I jump because the job changed. Search-plus-a-take is a Grok job. Twelve files and a failing test is not. A 2,000 word rewrite of meeting notes is not. You can force those through grok.com. You will spend the time you saved on catching tone and missing files.
A small loop I will actually do
Open the named thread or start one with the one-sentence problem. Make it search. Click the links. Lock a short fact block in my words. Draft the email or the decision in the same chat. Copy the sent version into my note. Close the laptop.
Next time the airline emails me a new number I am not a beginner again. I am a person with a URL and a paragraph.
That is the whole series. Step 2 if you want the search habit again, step 3 if you keep opening a blank email like I do.
I still open Grok for dumb current questions. I still click the page. I still sound like myself in the thing I send. The thread is just there so I do not have to pretend I have a great memory.