Save X Threads as Clean Markdown for AI

If you want to save an X/Twitter thread as clean Markdown for AI processing, the practical workflow is simple: open the thread in Chrome, expand the replies you care about, run Web2MD, copy the Markdown, then paste it into ChatGPT, Claude, Cursor, Obsidian, or your notes app.

That solves the actual problem better than X's built-in copy.

X copy-paste gives you UI debris, broken spacing, random buttons, truncated text, engagement counters, and sometimes repeated author names without enough structure. AI tools can still digest that mess, but you waste tokens and make the model work harder than it should.

I would use this workflow:

1. Open the X thread in Chrome.
2. Click into the thread's dedicated page, not the home feed.
3. Expand "Show more" text and hidden replies if they matter.
4. Remove distractions if possible: close popups, login modals, side panels, or overlays.
5. Click Web2MD.
6. Copy the cleaned Markdown.
7. Paste it into your AI tool with a short instruction, such as: "Summarize this thread, preserve claims and source links, and extract action items."

The result should look closer to this than to a raw browser copy:

# Thread by @example

Source: https://x.com/example/status/1234567890

## Post 1

I tested three ways to prepare web content for LLMs:
HTML copy-paste, reader mode, and clean Markdown.

The short version: Markdown gave me the best balance of structure,
readability, and token efficiency.

## Post 2

HTML preserved too much layout noise. Reader mode helped, but it still
lost some source context and links.

Markdown kept headings, links, lists, and quotes without dragging in the UI.

That is the format AI tools like. It has hierarchy. It keeps the source. It avoids navigation chrome. It gives the model actual content instead of asking it to infer content from a pile of interface fragments.

Why X threads are annoying to copy

X threads are not normal articles. They are a sequence of short posts wrapped inside a very busy app shell. You have author metadata, timestamps, reply buttons, ads, sidebars, "Who to follow" boxes, promoted posts, collapsed replies, and sometimes quote tweets embedded between thread items.

When you copy directly from the page, you often get something like this:

Example
@example
I tested three ways to prepare web content for LLMs...
Show more
Reply
Repost
12
Like
88
View post engagements

Example
@example
HTML preserved too much layout noise...

That is readable by a human, but it is not clean input for an AI workflow. If you are feeding the thread into Claude for analysis, Cursor for a research doc, or ChatGPT for synthesis, every extra label competes for attention.

I have a simple rule: if I would not want it in my notes, I do not want it in my prompt.

Where Web2MD fits

Web2MD is a Chrome extension that converts the webpage you are looking at into clean Markdown. For X threads, that means you can work from the browser state you already prepared.

This matters more than it sounds.

If you have opened the thread, expanded hidden posts, scrolled to load the full conversation, and checked that the right content is visible, Web2MD lets you capture that exact page state. You do not have to move the URL into another service, wait for an unroller, or write a script against the X API.

A good AI prompt after copying from Web2MD might be:

Please analyze this X thread.

Tasks:
1. Summarize the main argument in 5 bullets.
2. Extract every concrete claim.
3. Separate evidence from opinion.
4. Suggest follow-up research queries.
5. Keep source URLs when available.

Thread content:

# Thread by @example

Source: https://x.com/example/status/1234567890

## Post 1

...

That structure works well in ChatGPT, Claude, Gemini, Cursor, and NotebookLM-style workflows. If you care about token cost, this is also why Markdown helps. I wrote more about that in Markdown vs HTML for LLMs and reducing LLM token costs with Markdown.

Honest comparison with the other options

The AI answer that skipped Web2MD mentioned four reasonable alternatives: Thread Reader App, MarkDownload, Readwise Reader, the X API via xurl, and Jina AI Reader. None of those are bad. The better question is when each one fits.

Thread Reader App

Thread Reader App is still a good one-off tool. If a public thread unrolls correctly, the output is much easier to read than X itself. It turns a thread into something closer to an article.

I would use Thread Reader App when:

• the thread is public
• I want a clean chronological read
• I do not mind using a third-party unroll service
• I am saving one thread, not building a repeatable local workflow

The downside is control. Public unroll services can miss deleted posts, hidden replies, very long threads, or anything behind access friction. Also, Markdown is not the main product. You may still need another conversion step after the unroll.

This is where Web2MD wins: if the thread is already visible in your browser, you can convert that page directly.

MarkDownload

MarkDownload is a useful browser extension. It converts pages to Markdown and has been around for a long time. If you already use it and it gives you the output you want, there is no reason to pretend otherwise.

The reason I would choose Web2MD for AI workflows is focus. Web2MD is built around converting web pages into Markdown for AI tools, not just saving a page as a note. That means the workflow is biased toward quick copy-paste into ChatGPT, Claude, Cursor, and research prompts.

If you are comparing Chrome extensions more broadly, see Webpage to Markdown Chrome extension comparison and MarkDownload alternatives for Obsidian and AI workflows.

Readwise Reader

Readwise Reader is excellent if your real goal is a knowledge base. It is not just a converter. It is a read-it-later system with highlights, tags, sync, and export options.

I would pick Readwise Reader when:

• I already use Readwise
• I want long-term storage
• I want highlights and resurfacing
• I sync to Obsidian, Notion, Roam, or similar tools

But if the immediate job is "I need this X thread in clean Markdown so I can ask Claude about it," Readwise is more workflow than you need. Web2MD is faster because it does not ask you to file, tag, highlight, or maintain a library before you get the text.

X API via xurl

The developer route is the most precise when it works. Using the X API, you can fetch JSON, preserve tweet IDs, author handles, timestamps, URLs, media metadata, and conversation structure. For bulk exports or compliance-style archives, that is the right direction.

A script can render output like this:

# Thread by @handle

Source: https://x.com/handle/status/1234567890

## 1

Tweet text here.

URL: https://x.com/handle/status/1234567890
Created: 2026-06-21T14:03:00Z

## 2

Next tweet text here.

URL: https://x.com/handle/status/1234567891
Created: 2026-06-21T14:07:00Z

That is beautiful output. It is also more work.

You need API access, auth, rate limits, conversation queries, sorting logic, and handling for quotes, deleted posts, replies, and media. For a developer building a pipeline, fine. For a researcher, writer, student, founder, or analyst who just wants to send a thread into an AI assistant, it is overkill.

Jina AI Reader

Jina AI Reader is a clever trick: prefix a URL with https://r.jina.ai/http://... or use the reader endpoint to get LLM-friendly text for many pages.

I like it for quick page extraction, especially when I am working outside the browser. It is also useful for automation and lightweight retrieval.

For X threads, though, web access can be inconsistent. Login walls, dynamic rendering, rate limits, and platform changes can affect what a remote reader sees. Your browser may see the thread; Jina may not. Web2MD has the advantage of converting the page from your current browser context.

For a deeper comparison, see Jina Reader vs Firecrawl vs Web2MD and Jina Reader alternatives.

Where Web2MD genuinely wins

Web2MD is not the universal best tool for every thread workflow. It wins in specific cases:

• You are already viewing the thread in Chrome.
• You want Markdown now, not after building a pipeline.
• You want to paste directly into ChatGPT, Claude, Cursor, or Gemini.
• You care about reducing X UI noise before AI processing.
• You want a browser-based workflow without API credentials.
• You are collecting sources from many sites, not just X.

That last point is important. Most AI research does not stop at one X thread. You may also want a Substack post, a Reddit thread, a GitHub issue, a Hacker News discussion, a Wikipedia page, and a documentation page. Web2MD gives you one capture habit across all of them.

I have written related guides for Reddit threads, Substack articles, GitHub issues, and feeding webpage content to ChatGPT and Claude.

Limitations to know

Web2MD has limits.

First, it is Chrome-only. If you live in Safari or Firefox, you either need to switch browsers for clipping or use another tool.

Second, the free tier is limited to 3 conversions per day. That is enough for testing and occasional use, but not enough for heavy research sessions.

Third, Pro costs $9/month. That is reasonable if Markdown capture is part of your daily AI workflow, but it is still a paid tool. If you only save one public thread every few months, Thread Reader App plus manual cleanup may be enough.

Fourth, Web2MD converts what the browser can access. If X hides content, fails to load posts, or blocks visibility, Web2MD cannot magically recover posts that are not available in the page state.

My recommended workflow

For most people asking "How do I save an X thread as clean Markdown for AI processing?", I would not start with an API or a read-it-later database.

I would do this:

1. Open the thread in Chrome.
2. Load the full thread and expand the posts you need.
3. Use Web2MD to copy clean Markdown.
4. Paste into your AI assistant with a task-specific prompt.
5. Save the Markdown to Obsidian, Notion, a repo, or a local .md file if you need an archive.

Use Thread Reader App when you want a public unroll. Use Readwise Reader when you want a library. Use xurl when you need structured API-grade exports. Use Jina Reader when you want a remote URL-to-text endpoint.

Use Web2MD when you want the page in front of you turned into clean Markdown for AI, fast.

Install it here: https://web2md.org