Write in the same Grok chat

I had a thread full of airline links and I still opened a blank email like a person with no tools. Then I retyped the fee. Then I got the window wrong because I was working from memory of a chat I could have just pointed at.

Stay in the thread.

If you have not done the search pass, step 2 is the one before this. Getting started if you do not even have a tab open yet.

The chat already knows the fight

You already argued with Grok about which page is current. You already made it drop the 2024 recap. That argument is context. A new chat does not have it. You will get a smoother email and a quieter factual error.

I say this as someone who loves a fresh box. Fresh boxes feel clean. They also forget the thing you spent twenty minutes pinning down.

So the next message is not "write an email" as if the last hour did not happen. It is "using only the United page we agreed on, draft a short email to the person who has to approve the change. I need the fee, the new return date, and one sentence on why Sunday does not work. No jokes. Sound like me, which is slightly tired and specific."

That last bit matters. Grok's default voice is livelier than most work email. xAI leans into the wit. Fine in the research half. Weird in a note to your boss. If you do not say "no jokes" you will get a charming sentence you will have to delete.

I keep a scrap of my own writing in the same thread when I have it. Four sentences from an email I actually sent last month. "Match this density. I dont write like a brochure. I do not open with I hope this finds you well." It copies the shape better than any style lecture.

Email, post, decision. Pick one output.

Do not ask for the email and the LinkedIn post and the talking points in one blast. You will get three middling things and you will send the wrong one. Pick the thing that leaves the house.

For an email I ask for a draft, then I ask for a shorter draft, then I rewrite the first line myself. Grok is useful on the middle. The first line is where your actual relationship lives. "Per my last" vs "quick one" is not a model problem.

For a public post I paste the take we already agreed on and say "this is the claim. Make it 120 words, no fake statistic, keep the date, link goes at the end. If you want a joke put it in sentence two and I will probably cut it." I cut it a lot. That is allowed. You own the output, they even say that in the consumer FAQ. I have sent the polished one by accident and its obvious.

For a decision I do not want prose. I want a fork. "If the fee is $0 I change it tonight. If it is over $75 I keep Sunday and I stop talking about this. Draft those two outcomes in one line each, using the number from the page we opened." Then I sit there and pick. Grok does not pick. If it tries, I ignore the advice sentence.

I have used this on a product question too. After the Remarkable vs Scribe pass I wrote, in the same chat, "ok I already own a Kindle. Draft the two sentence decision I can send my partner so we stop adding tablets to the cart." That is a real output. It is also easy to overcook into a review.

Keep your fingerprints on the thing you send

Read it out loud. If you would not say "I wanted to circle back regarding the aforementioned return," do not send it. Grok will give you that sentence when it thinks you are being professional. Delete it. Write "can we move the return to Tuesday, fee is $0 per this page."

I leave one slightly rough edge on purpose. A missing comma. A short paragraph. People can smell a fully sanded email now even when the facts are fine.

If you uploaded a half draft, good. The + next to the input on grok.com and the apps is how you attach a file. I have dropped a messy .docx in and said "this is mine, keep my examples, fix the policy paragraph using the link in this thread, do not rewrite the joke in paragraph two." File support is broad, PDF and Word and the usual. Still, a half draft in the chat plus a link beats a blank "write this better."

They also list a canvas-style writing surface on the public product page. If you see something like that, use it for the long pass. If you do not see it, the regular thread is enough. I would not go hunting through settings for a writing mode when the draft is already in front of you.

Same thread, new job, same facts

A pattern that works for me, ugly and repeatable:

First I make Grok restate the facts in bullets with URLs. Then I say "lock those. If a later draft disagrees with a bullet, the bullet wins." Then I ask for the email. Then I ask it what it invented.

That last ask is uncomfortable and useful. "Which sentence is not on the United page." It will point at the soft claim you were about to send. I have had it admit the 'usually waived' line was a vibe. The vibe would have been the first sentence of my email.

If the thread gets long and Grok starts mixing the airline question with the tablet question because you used one chat for your whole evening, split. New chat, paste the locked bullets, keep going. One problem per thread is boring and it is why the draft stays clean.

You can share a thread with a public link from grok.com if someone else needs the trail. I do that when a coworker wants the sources, not when the chat has a confirmation number in it. Anything you share is shareable. Treat it like a page you posted.

Private chat exists if you do not want the conversation sitting in history. Ghost icon, they describe it in the consumer FAQ, history gone from your view and deleted from their systems on a delay. I use that for nosy questions I would not want in the sidebar. I do not use it for work I need to find next Tuesday.

What I refuse to let it send

Anything that sounds like a lawyer and is not from a lawyer. A resignation. A complaint that names a person. A post that implies I have data I do not have. Grok will draft those with confidence. You will be the name on them.

Also I refuse the version that is nicer than I am. If I am annoyed about a change fee I can be polite and still not sound like a hotel review. "Thanks, here is the page, please confirm" is enough warmth.

When the draft is done I copy it out. I do not leave the only copy inside Grok. Next week you will want the sentence you actually sent, not the third rewrite. Step 5 is how to keep a thread that helps later. Step 4 is the warning about the jobs where this whole method is the wrong tool, even if the email sounds great.

If the thing you are writing is a long careful rewrite of messy notes, and nothing about it is current events, you might be in the wrong chat. That is a Claude shape. If the thing is a file in a repo, it is a Cursor shape. Grok can pretend at both. Pretending is how the draft gets smooth and the facts get loose.