Updated

Hacker News · Free account

Send a Hacker News thread to ChatGPT or Claude.

Pull every comment in an HN thread, nested replies included, and open ChatGPT or Claude with the comments and a prompt already in the chat box. Nothing is sent until you press send.

How do you send a Hacker News thread to ChatGPT or Claude?

Open the thread, pull it with Threadpull in Chrome or Edge, choose Send to AI and pick ChatGPT or Claude. The chat opens with every comment, replies nested as quotes, and your chosen prompt already in the box. Nothing is sent until you press send. It uses the ChatGPT or Claude account you're already signed in to, so you don't need an API key.

What one real thread showed

We pulled a live Show HN and asked for the themes a maker should act on. This is what came back, checked against the thread.

Show HN · pulled 1 Nov 2026

Show HN: Capsule – Single-file web apps that save their data into SQLite

news.ycombinator.com/item?id=49712278 ↗
Comments166
People122
Reply depth11

The change for the next launch post: place the product against the tools people kept naming, and say plainly how sharing and syncing work.

Comparisons to existing tools

about 30 comments
"This sounds like the modern take on MS Access."m-p-3
"This reminds me of Clipper/dBase software tools."molsongolden

Praise for local AI-built mini-apps

about 17 comments
"It's so easy to create small tools with AI now, but hard to install them as local apps or share them."andix
"AI has reignited my interest in building based 100% on browser native features."eleventen

How do sharing and syncing work?

about 16 comments
"Will that merge into my copy if I've added some more items or would it open as a "fresh" app?"domh
"I think data and apps should be separated, I might want to share an app with someone else without sharing the data"andix

A closed format needs a viewer app

about 13 comments
"they first need to download the capsule app to associate the file extension with the OS and have it launch, right?"al_borland
"Curious: this requires users to have a host app installed on their machine that can read .capsule file right?"lewisjoe

The thread in numbers

"Show HN: Capsule – Single-file web apps that save their data into SQLite" had 166 comments from 122 people when we pulled it on 1 Nov 2026. 95 of them (57%) were replies to other comments, nested up to 11 levels deep, and the median comment was 29 words long. The theme raised most, comparisons to existing tools, came up in about 30 comments (18% of the thread).

How we did this research

  • Pulled with Threadpull on 1 Nov 2026: every comment still live, at every reply level, from news.ycombinator.com/item?id=49712278.
  • Read and grouped into themes. A comment can raise more than one theme, so counts are approximate.
  • Every quote on this page is checked word for word against the export, under its author, each time the page is built.
The thread at a glance
Score on the post380
Comments166
Top-level comments71
Replies95 (57%)
People commenting122
Comments by the original poster13
Deepest reply level11
Median comment length29 words
Comments from the 5 most active people25 (15%)
Themes, by share of all comments
ThemeCommentsShare
Comparisons to existing toolsabout 3018%
Praise for local AI-built mini-appsabout 1710%
How do sharing and syncing work?about 1610%
A closed format needs a viewer appabout 138%

The words they used

How many comments used each word or named each product. Counted from the export each time the page is built; a comment counts once however often it repeats the word. Variants count together, so "self-host" includes "self-hosted" and "self hosting".

"Show HN: Capsule – Single-file web apps that save their data into SQLite"

Words and phrases
Word or phraseCommentsShare
share1710%
sync117%
File System Access85%
single file74%
open source64%
security53%
Products and tools named
NameCommentsShare
SQLite95%
Firefox74%
Google Drive53%
PWA53%
Electron42%
Microsoft Access42%
Excel32%
Tauri32%

Where the conversation happened

Comments per reply level. The deep chains are often where people disagree, and a skim of the top-level comments misses them.

Comments per reply level in the Hacker News thread. Top level: 71, Level 1: 45, Level 2: 25, Level 3: 12, Level 4: 5, Level 5+: 8.Top level71Level 145Level 225Level 312Level 45Level 5+8
Hacker News: 116 of 166 comments (70%) were top-level or direct replies, and the deepest chain went 11 levels. Half the comments arrived within 3.9 hours of the post, and 90% within 25 hours.

How to send an HN thread to your AI

  1. Open the thread

    Any news.ycombinator.com/item page: a story, Ask HN, Show HN or a single comment.

  2. Pull the comments

    Threadpull collects every reply in HN's order. A 400-comment thread takes about half a minute.

  3. Choose Send to AI

    Pick ChatGPT or Claude, and one of the ready-made prompts or just the comments.

  4. Read the answer

    The chat opens with everything in the box. Add your own question if you like, then press send.

Prompts you can choose

Pick one in the Send to AI sheet, or choose Comments only and type your own question.

PromptWhat it asks for
Comments onlyThe post and comments, no instructions
Top objectionsWhat stops people buying, with quotes
Praise and benefitsWhat people love, in their words
10 ad hooksScroll-stopping lines from real phrasing
Customer avatarWho they are, pains, desires, language

The prompt we used

Copy it into the chat box above the comments, or start from one of Threadpull's ready-made prompts.

These are every comment from a Hacker News thread about a product launch.
1. Find the 4 themes that come up most, with a count for each.
2. For each theme, quote one comment word for word.
3. List the tools the product is compared to.
4. List the questions the maker should answer in the post.
5. Suggest the single most useful change to the launch post.

Getting a good answer from a whole thread

Why send every comment instead of a summary?

A summary keeps what one reader noticed. Sending every comment lets the model count themes across the whole thread. In the Show HN above, 95 of 166 comments were replies, which a skim of the top-level comments would miss.

How does the AI know who replied to whom?

Threadpull writes each comment under its author's username, and each reply as a Markdown quote one level deeper than the comment it answers, so the model can follow a back-and-forth.

How long a thread can I send?

Threadpull includes up to 500,000 characters, roughly 80,000 words. Longer threads are trimmed, and a note in the chat says how many comments were included. The Show HN above was about 6,700 words, well inside the limit.

Which prompt should I start with?

Top objections before a launch, Praise and benefits for copy, Customer avatar to learn who's commenting, and 10 ad hooks for ads. For anything else, choose Comments only and write your own question, like the one above.

How to find threads like these

Searches that turn up the threads worth pulling. Open one, then pull it with Threadpull.

SearchWhereWhat it finds
"Show HN" plus your categoryhn.algolia.comLaunch threads to summarise for objections and requests
"Ask HN" how do youhn.algolia.comWorkflow threads where people explain what they actually do
site:news.ycombinator.com "I switched from"GoogleSwitching stories: what pushed people to leave a tool
Sort by comments on hn.algolia.comhn.algolia.comThe biggest threads in a time range, the ones most worth an AI pass

Threads worth pulling next

Comment counts as of 2 Nov 2026.

Hacker News to AI FAQ

Does Threadpull send anything to ChatGPT or Claude by itself?

No. It opens the chat and fills in the box. Nothing is sent until you press send there.

What if the thread is too long for the chat?

Threadpull tells you when it trims the comments to fit, and how many were included. Export the full thread as CSV or JSON if you need every comment.

Do I need an OpenAI or Anthropic API key?

No. It uses the ChatGPT or Claude account you're already signed in to.

Is it free?

Hacker News threads need a free account. Reddit threads work without one.

Ask your AI what a whole thread thinks.

Add to Chrome, it's free