When should you use it?
- You publish often and spend ages scrolling first
- Your feed is curated to your niche
- You want ideas, not finished posts
Not for: Writers who want the agent to form opinions; Feeds containing private messages you would not share
What do you need, and what do you get?
You need
- Claude with computer use on your desktop
- A browser logged in to your social account
- A sense of what your audience likes
You get
- A handful of posts worth covering
- Engagement figures for each
- Suggested angles
How does it work, step by step?
- Ask narrowlyRequest two to four posts from your own feed worth writing about, not a general web search.
- Let it browse your sessionIt opens the browser where you are already signed in and scrolls.
- Read the numbersUse views and likes to judge whether a topic is niche or widely discussed.
- Choose yourselfPick the story and angle; that judgment is the job.
- Draft only if usefulAccept its offer to start a draft, or write from scratch.
The prompt
The original prompt by Ilia Karelin, word for word, with a link to where it was published. Below it is our adapted version with fill-in blanks, if you want it. Works in Claude. Values in lime are placeholders.
Original prompt by Ilia Karelin
Can you go to X (I'm already logged in on my computer) and find 2-4 posts worth talking about on my Substack?
Exactly as published by Ilia Karelin in Claude Computer Use: I Used It to Research This Post.
Our adapted version (optional)
Open my X feed in Chrome where I am signed in. Scroll for a few minutes and pick 2 to 4 posts worth covering for my newsletter on [topic]. For each, give the author, link, views and likes, and a one-line angle for my audience. Do not like, reply, follow or post anything.
What can go wrong?
- Feed popularity is not the same as importance to your readers.
- Computer use is slow; it took about a minute and a half here.
Our take
Ilia Karelin's use of Claude computer control through Dispatch is tidy and low risk: read-only browsing in his own logged-in X session, with the choice of story and opinion kept human. Scrolling the feed, reading engagement figures and returning a handful of posts with possible angles is a sensible daily research pass for someone who publishes often. Ask for links to each post so you can verify quotes before citing them. If you run it daily, keep a list of topics already covered so the same story does not resurface. Avoid letting it click anything beyond scrolling and opening posts; it should not like, reply or follow. Treat engagement numbers as a hint, not proof of importance. Draft only after you pick the angle yourself. Close unrelated tabs with private messages before enabling computer use, and stop the session when the scan is done. Prefer a dedicated research profile if your main session holds private DMs.
Where the time goes back: The research scroll before writing is the slow part for daily publishers; this compresses it to a minute or two.
FAQ
Does it write the post?
Why not just search the web?
Source and credit
The original prompt on this page is by Ilia Karelin, shown word for word with a link to where it was published. The summary, notes and adapted version are ours.
- Original
- Claude Computer Use: I Used It to Research This Post
- Creator
- Ilia Karelin
- Published on
- prosperinai.substack.com (Substack)
- Original date
- Mar 26, 2026
- License
- Not openly licensed; the original prompt is shown word for word with credit and a link to the original
- Originally for
- Claude
- AI dry run
- Oct 23, 2026: passed. Dry-run tested by our AI test agent on sample input. PASS: Gates held on desk check; no live sends or purchases.
AI dry run on sample data; not run with live connected accounts. The playbooks.bot test agent filled the prompt with synthetic inputs, produced the output and checked it against the page's stated outputs and guardrails. - Verified
- Oct 24, 2026: source link resolves to the original, the credited creator published it, our excerpt or prompt matches it, and the license allows our use.
Are you the creator? Claim or correct this page.