The anatomy of a Gemini Notebook: 6 parts that turn 50 files into a podcast

OpenAI began rolling out Locked chats on 20 October 2026. The feature hides chosen conversations from your regular chat history, pinned chats and normal ChatGPT search, and asks for a 4-digit account PIN, Face ID, Touch ID or your device unlock before it shows them. The company put it in the ChatGPT release notes and a dedicated help page; the rollout is gradual, so the option may not appear for you yet.

The intended use is narrower than a privacy overhaul. OpenAI describes it as a way to keep sensitive conversations out of view when you share a screen or hand someone your phone. In practice that covers a lot: medical questions, salary negotiations, a message you are drafting about a friend, or a surprise you are planning. Locked chats also stop those titles showing up when someone else scrolls your sidebar.

How to lock a chat

These are the documented steps, not a procedure we independently tested; interface details vary by platform and rollout.

1. Open the conversation you want to hide.

2. Open its more options menu (the three dots) and select Lock chat.

3. Confirm Lock chat again.

4. Create a 4-digit account PIN if you do not have one, or enter it when prompted.

The chat moves to Projects > Locked chats. You can open that area later, enter the PIN or use device authentication, and start new conversations from there. On iOS and Android you can switch on Face ID, Touch ID or device unlock in the account PIN settings. Device authentication is set up per device and does not replace the PIN.

Access is temporary by design. On web and desktop it applies to the tab or window where you entered the PIN and ends when you refresh or close it. On iOS and Android it ends about one minute after you leave the locked area or put the app in the background. A Lock now button ends it immediately.

What Locked chats does not do

This is the part worth reading before you trust it with anything serious. OpenAI says locking a chat does not:

• create end-to-end encryption or a separate encrypted vault,

• change how long OpenAI retains the conversation,

• change your Improve the model for everyone setting,

• remove anything ChatGPT remembered before you locked the chat,

• override your workspace's policies, safety rules or legal obligations.

So the lock controls who can see the chat on your screen. It does not change who can access the data behind it. In a work workspace, administrators cannot unlock your chats or reset your PIN, but Enterprise and Edu locked chats still appear in conversation logs available through the OpenAI Compliance Platform. And anyone who can open a locked chat can still copy its contents.

The memory split is the useful surprise

If Memory is on, locked chats get their own memory that is shared across conversations inside the locked area. New messages sent after locking can be remembered there, and that memory does not carry back into your regular chats. Regular memory can still be used inside locked chats, and locking does not erase what was already remembered. On balance that is a sensible split: the private thread learns from itself without leaking its details into the chat you use at work.

Nine conversation types cannot be locked

The Lock chat option is unavailable for temporary chats, archived chats, group chats, chats created from a shared link, chats with scheduled tasks, chats with meeting recordings, chats associated with background tasks, chats with responses saved as project sources, and chats in the earlier Health experience. Codex conversations cannot be locked, and the feature is not available on ChatGPT for Healthcare. If you lock a chat that came from a project, it leaves that project and stops using its instructions, files and memory.

The PIN reset is a delete button

This is the one warning to say out loud. Changing your PIN, with the current PIN in hand, keeps your locked chats. Resetting a forgotten PIN permanently deletes every locked chat, the memories stored only inside them and files protected only by that PIN. OpenAI says the deletion cannot be undone and support cannot recover the content. A workspace administrator cannot reset it for you either.

Before you rely on the feature, decide where you will keep the PIN. A password manager entry is more reliable than memory. If the chats matter, consider exporting anything you would hate to lose; the reset path is not a recovery flow.

What to do now

Check whether Lock chat appears in a conversation's three-dot menu. If it does, lock one genuinely sensitive thread and test the end of access: leave the tab or background the app, then confirm it asks again. Do not treat the lock as a replacement for deleting data you no longer want stored, and do not put passwords, card numbers or other secrets into a chat just because the title is hidden.

Locked chats are available across Free, Go, Plus, Pro, Business, Enterprise and Edu plans as the rollout reaches them. The feature is a useful screen-level privacy control, not a private vault. That distinction is the whole story.

A first-time iOS developer wanted a simple thing: a private place on an iPhone for photos and files that nobody else could open. Not Apple, not a cloud provider, not him. He built it in about three weeks, mostly by directing AI coding agents, and launched with no audience, no ad budget and no marketing plan.

The first month, the app made $66.

Six months later it made $41,000 in a single month, and it is now averaging roughly $2,000 to $3,000 a day, as the founder says in a Starter Story interview published on 18 October 2026. The app has passed 60,000 downloads. He says he has not written a single line of code or a single line of the website copy himself.

The interesting part is how he got there, because it was not a viral video and it was not paid acquisition.

The product is almost boringly simple

The app stores photos, videos and files in an encrypted vault on the phone. You unlock it with a pattern, keep a recovery phrase, and there is no account and no sign-up. Because the keys never leave the device, the company cannot read the contents and cannot recover them either.

That simplicity is the point. The category is crowded with apps that are 15 years old and have millions of users, which means they rank well and get taken for granted. He could not outspend them, and he could not outrank them inside the App Store.

So he attacked the part they ignore: the open web.

The growth engine: one guide a day

He started with a normal marketing site: pages outlining each feature, then a series of guides aimed at the questions people actually type, such as how to store photos privately on an iPhone. Every one of those pages was written by a coding agent, not by him.

The loop is mechanical. Each morning, a small command-line tool he built queries Google Keyword Planner, looking for keywords where his site already appears in search results but gets almost no clicks. That gap is the opportunity: search engines have decided the page is relevant, but the title and the content are not earning the visit.

A guide is generated for that gap. He publishes about one a day. Google typically indexes a new page in three to four days. Internal links from pages that already receive traffic push the new one up, and the library compounds.

He says roughly 99% of the remaining daily work is a Claude agent waking up and running the loop.

Where the money actually comes from

The site is now the majority of his traffic, and he says about 10% of web visitors download the app.

The conversion curve is the unusual part. Only 5.47% of people who download on day one end up paying, and day one is the worst day, not the best. Conversion climbs for about two months and peaks around day 60. That is the opposite of the standard consumer-app playbook, where a user who does not pay immediately usually never does.

Lifetime purchases do most of the work: roughly 60% of purchases and more than 80% of revenue, as he describes it. The product earns trust slowly, and the pricing reflects that.

How to copy the method

Pick a need people already search for, not a clever idea nobody is looking up. He did not invent demand for private photo storage; he found a crowded category with old incumbents and went after the questions they never bothered to answer.

Build the smallest honest version of the product. Three weeks and coding agents were enough because the feature set was narrow and the promise was clear. A vague product cannot be explained by a guide.

Find the queries where you already appear without clicks. If you have no site yet, start with the plain questions your buyers ask before they know your product exists. Write the answer better than the first page of results, not longer.

Publish on a schedule you can keep, and link every new page from pages that already get traffic. One good page a day beats ten rushed pages once. Then measure the two numbers that matter: search impressions and the share of visitors who take the next step.

Finally, fix the conversion path before you scale the traffic. His call to action sits where readers naturally stop, and his paid tier is a one-time upgrade rather than a subscription that expires. Traffic into a broken funnel is just expensive noise.

What to watch out for

Every figure here is self-reported in a promotional case-study interview and has not been independently audited. Revenue is not profit: AI coding tools, hosting and payment fees all cost money, and the interview does not break out margins.

He also had unusual advantages. He is a repeat founder with a technical and security background, and the app was built on a privacy promise that is hard to copy without real care. A security product that leaks once is finished.

The daily publishing loop is also a real commitment, and search results are not guaranteed. Google can change how it indexes, the keyword gap can close, and a competitor can copy the same pages. The method lowers the cost of distribution; it does not remove the risk.

And the honest framing matters: this is a report of what one person did, not a promise that the same app or the same numbers will happen to anyone else. The repeatable part is the loop, not the outcome.