Announcing the Wikiprompt Open Prompt Dataset: 55,000+ AI Prompts, Free

Announcing the Wikiprompt Open Prompt Dataset: 55,000+ AI Prompts, Free

Today we're opening up the entire Wikiprompt catalog. Every prompt we've curated across ChatGPT, Claude, Gemini, GPT Image, Midjourney, Seedance, Veo, Kling, Nano Banana, Grok and more is now available as a public, structured, free dataset. No signup, no API key, no scraping required.

If you've ever wanted a real corpus of working AI prompts, the kind other people actually used and got results with, rather than a list scraped off a random blog, this is that corpus. Over 55,000 prompts, each with its full record: title, description, the actual prompt text, category, tags, media, the model it was written for, structured quality metadata, author, and a link back to the original post it came from.

Why we're doing this

Wikiprompt has always worked like a wiki: people submit prompts, we curate and organize them, and the result becomes a public reference anyone can browse for free. That worked well as a website. It didn't work as well for the people who wanted to *build* something with the data: train on it, analyze it, mine it for patterns, feed it into their own tools. Their only option was to scrape wikiprompt.org page by page, which is slow, fragile, and unnecessarily hard on our servers.

So we're doing what Wikipedia did decades ago: shipping the dump. Instead of scraping, you fetch. Instead of parsing HTML, you get JSON. The whole catalog, in one predictable format, updated as the catalog grows.

What's actually in it

Start at the dataset. That endpoint is the manifest: it tells you total_prompts, the record_fields you can expect on every record, and how pagination works. The real content lives at /dataset/prompts, which returns the catalog itself as JSON.

Each record includes:

Only active, cleaned-up prompts make the cut, the same editorial bar we apply to the live site. Categories span creative, marketing, personal, productivity, coding, education, business, research, and other, so there's real range here, not just one niche.

To make it concrete: the dataset is how a prompt like this editorial smoking poster prompt or a hand-cut linocut travel poster prompt ends up as a row of structured JSON instead of just a page you'd have to visit and copy by hand. So does an ethereal Tang dynasty portrait photography prompt, and every other prompt in the catalog.

How to grab it in one line

This is the part we're most excited about: there's no setup. One curl command gets you a page of real records right now.

curl "https://www.wikiprompt.org/dataset/prompts?limit=500"

That's it. No key in the header, no auth dance. CORS is wide open (Access-Control-Allow-Origin: *), so you can call it straight from a browser or a client-side app too. It's also heavily edge-cached, which means it's fast for you and cheap for us, a rare combination.

The dataset is paginated with a keyset cursor rather than page numbers, which keeps it stable even as new prompts get added mid-crawl. Each response includes a next URL; keep following it until next comes back null. You can request up to 500 records per page with limit, or take the default of 200. A minimal loop looks like this:

import requests

url = "https://www.wikiprompt.org/dataset/prompts?limit=500"

prompts = []

while url:

data = requests.get(url).json()

prompts.extend(data["results"])

url = data.get("next")

print(len(prompts), "prompts fetched")

Run that and you'll have the full catalog sitting in memory in a few minutes.

A note on attribution

Every prompt in this dataset was aggregated from a public post by its original author. Wikiprompt curates and structures the catalog, but we don't hold a formal license over the content itself. If you build something with this dataset, publish research on it, or reuse individual prompts, please credit wikiprompt.org as the source and, where it matters, the original_source link on the specific record you're using. That's the whole ask.

Beyond the dump

The dataset is the bulk option, but it's not the only way in. If you just need to answer a single question, the search API returns JSON for any query without downloading the whole catalog. If you're building an AI agent, the MCP server exposes search, retrieval, and submission as tools Claude and other agents can call directly. And llms.txt gives any LLM a quick map of what's here and how to use it.

We built Wikiprompt to be a public reference for how people are actually prompting these models today. Opening the dataset is us taking that idea seriously: the whole catalog should be as easy to get as it is to browse. Go grab it and see what you build.