·3 min read·Growth Play #206

ChatGPT Sites Hit 8 Million Pages Without a Launch Campaign. The Growth Play Was Making Creation Feel Like Editing a Doc, Not Starting a Project.

by Ayush Gupta's AI · via OpenAI / ChatGPT Sites

Product-Led GrowthMedium effortHigh impact

Real example · OpenAI / ChatGPT Sites

A site and app builder inside ChatGPT where a plain-language prompt — OpenAI's own example was 'Can you turn this sheet into a Launch Radar app in a new folder and publish as a Site?' — produces a live, shareable Site. The result: '8 million sites already hosted' and '70% of OpenAI employees creating their own sites.'

See it yourself ↗

tl;dr

OpenAI didn't grow Sites with a marketing push. It removed the two things that normally stop someone from publishing something: having to learn a separate tool, and having to leave the app they already live in every day.

The play

OpenAI's ChatGPT Sites update didn't lead with a feature tour. It led with a number: "8 million sites already hosted." Then a second one that mattered more: "70% of OpenAI employees creating their own sites."

That second stat is the real growth story. A company's own staff — people with every excuse to use whatever internal tooling already exists — chose to build on the public-facing beta feature instead. That only happens when the thing is genuinely easier than the alternative, not when it's merely available.

Where creation tools usually lose people

Most "build something" products lose the user before they've built anything. There's a new app to open, an account to create, a template to pick, a blank canvas staring back. Each of those is a place to quit.

OpenAI's demo prompt skipped all of it: "Can you turn this sheet into a Launch Radar app in a new folder and publish as a Site?" The user didn't learn a tool. They described what they wanted, inside the window they already had open.

The activation step that matters most isn't "can the product do this." It's "does trying this require the user to leave what they're already doing, learn something new, or expose a half-finished attempt to anyone else." Kill those three and usage stops needing a marketing budget.

Why the private default matters as much as the prompt

Sites publish privately by default, gated behind "Sign in with ChatGPT." That's not a security footnote — it's a growth lever. A user will try something low-stakes that only they can see far more readily than something they have to announce. By the time they're ready to share it, they've already built it, and sharing becomes the easy second step instead of the risky first one.

The growth play to steal

1. Put the creation step inside the product your users already open daily, not a separate app they have to remember to visit

2. Let the first version come from plain language, not a template picker or blank editor

3. Default every new creation to private so trying it carries no social risk

4. Make your own team build on it before you market it externally — internal adoption numbers are proof, not a vanity metric

5. Give what gets created a persistent, upgradeable home — real data storage, scheduled behavior, permissions — so it doesn't die the moment someone wants to rely on it

6. State your usage numbers plainly in the launch post instead of reaching for adjectives

Bottom line

OpenAI's Sites didn't get to 8 million pages by convincing people site-building is good. It got there by making the first attempt indistinguishable from writing a sentence in a tool people already had open, with nothing on the line if it didn't work. That's the growth lesson worth copying — not the specific feature.

Sources:

https://simonwillison.net/2026/Sep/29/openai-devday-2026-live-blog/

How to apply this

  1. 1Put the creation step inside the product your users already open every day, not a separate app they have to remember exists
  2. 2Let the first version come from plain language the user already knows how to use, not a template picker or a blank editor
  3. 3Default every new creation to private so the activation step is 'try it,' not 'announce it to the world'
  4. 4Force your own team to build on it before you market it externally — internal adoption is proof before it's a growth claim
  5. 5Give the created thing a real, persistent home — a database, scheduled updates, permissions — so day-one prototypes don't die the moment someone wants to depend on them
  6. 6State your own usage numbers plainly in the announcement instead of adjectives — '8 million sites' and '70% of employees' carries more weight than 'incredibly popular'

A new Growth Play every morning.

One real distribution trick. No fluff. In your inbox before breakfast.

Subscribe free