12:47Futurepedia · Tutorial
Understand the FULL Claude Ecosystem in One Video
A 13-minute map of every Claude product, feature, and layer — from Chat to autonomous Scheduled Tasks — with one simple framework to keep it all straight.
June 24thThe argument in one line.
ChatGPT Work's Sites feature turns any chat, skill, or automated workflow into a live, persistently stored website, closing the gap between a one-off answer and a tool worth reusing.
Read if. Skip if.
READ IF YOU ARE…
- You have ChatGPT Work or Codex access and want to turn research, decisions, or gathered data into something more useful than a chat thread.
- You build or use custom ChatGPT skills and want a simple front end that displays what the skill actually finds.
- You want a fast way to spin up an internal tool, tracker, or database without writing or hosting code yourself.
SKIP IF…
- You're on standard ChatGPT Plus Chat without Work or Codex access. Sites isn't available there.
- You need a production application with real engineering controls. This is closer to a fast internal prototype than a scalable product.
The full version, fast.
ChatGPT Work's Sites feature deploys a live, hosted website straight from a chat prompt, complete with its own URL, persistent storage, file storage, authentication, and analytics. Instead of ending a research thread with a text summary, you can turn the full context into an interactive site, one that updates in place rather than duplicating when you ask for changes. The same mechanism can wrap a custom skill in a persistent interface, preserving individual sources a summary would otherwise discard, and Scheduled Tasks can keep that site's underlying database updating on its own, turning a one-time build into a standing tool.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:31
01 · Intro
Sites is introduced as a ChatGPT Work feature that builds and deploys real websites and apps, complete with persistent data, file storage, authentication, analytics, and a live URL.
00:31 – 02:26
02 · Getting started
Work mode is explained as ChatGPT's more agentic mode. Tagging @Sites in a prompt is enough to build and deploy a working site. A quick 8-bit game example goes live in about nine minutes, then gets shared publicly and inspected in the Settings and Analytics panels.
02:26 – 04:43
03 · Use your existing context
Instead of asking for a text summary of a long research project (planning a Disneyland trip across many chat threads), the same context is compiled into an interactive site. Re-requesting the site updates the existing deployment instead of duplicating it. A second example, an RV buyer's guide, shows the same pattern turning research into a decision-making tool.
04:43 – 09:51
04 · Give a skill an interface
A third-party GitHub skill ('Last 30 Days') that researches a topic across Reddit, YouTube, and other sources is installed via the skill creator, then wrapped in a Sites interface so its full findings (not just the final summary) are preserved, searchable, and linkable. The site is then updated so the skill can push new research runs into it directly, and its persistent storage keeps a growing research library across sessions.
09:51 – 13:24
05 · Automate a living database
A searchable database of AI newsletter content is built by connecting a secondary Gmail account and backfilling over 2,000 past newsletters. A Scheduled Task is then set up to run automatically every day at noon, keeping the site's database current without any manual re-running.
Lines worth screenshotting.
- ChatGPT Work's Sites feature builds and deploys a live, hosted website directly from a chat prompt, complete with its own URL, persistent data, file storage, authentication, and analytics.
- Tagging @Sites in ChatGPT Work turned one short prompt into a playable game, deployed to a live URL, in about nine minutes.
- A new Sites deployment defaults to private, requiring the creator's own ChatGPT login, until it's explicitly made public from the site's Settings panel.
- Site Settings let you rename the URL, attach a custom domain, add environment variables, and view visitor analytics without touching any separate hosting configuration.
- Instead of asking ChatGPT to summarize a long research thread into a wall of text, you can ask it to compile that same context into an interactive site.
- Re-requesting a site from a project ChatGPT already deployed updates the existing site in place instead of creating a duplicate.
- A ChatGPT skill, a packaged tool built by someone else, can be installed into ChatGPT Work just by pointing the built-in skill creator at its GitHub repository.
- A skill's raw output is often richer than the final summary it prints. Building a site around the full output preserves individual posts, sources, and takeaways a summary would otherwise discard.
- Sites can carry persistent storage, so a research library built this way kept growing across sessions instead of resetting after each conversation ended.
- ChatGPT Work can connect multiple Gmail accounts to one login, which let a single backfill pull in over 2,000 historical newsletters to seed a searchable database.
- Scheduled Tasks is ChatGPT Work's built-in automation. Describing a recurring job in plain language, like processing new email at noon every day, is enough to set it running with no external script.
- No specific storage ceiling is published for a Sites database, but a text-heavy use case holding thousands of records worked without issue in testing.
- The general Sites workflow runs research, analyze files, connect data, build a workflow, then site, positioning the site as the last step of a task rather than the first.
Sites turns finished ChatGPT work into a tool you keep.
WHAT TO LEARN
Sites closes the gap between an answer ChatGPT gives once and a persistent, shareable tool, by adding a live URL, storage, and automation on top of any chat, skill, or workflow.
- A Sites deployment gets a real hosted URL, persistent data storage, file storage, authentication, and analytics automatically, with no separate hosting step to manage.
- Turning a finished research thread into a site beats asking for a text summary, because the reader can filter, search, and revisit the underlying context instead of scanning a wall of text.
- Re-running a site request against a project that's already deployed updates the existing site instead of creating a duplicate, so a URL you've already shared stays current.
- A skill's full output is usually more useful than its final summary. Building a site around the raw results preserves individual sources and takeaways a summary would otherwise discard.
- Persistent storage means a site-backed research library keeps growing across sessions instead of resetting, which is what makes building the interface worth it in the first place.
- Multiple Gmail accounts can be connected to one ChatGPT Work account, which is useful for pulling in a large historical backlog in a single pass to seed a database.
- Scheduled Tasks turns a manually run process into a standing automation just by describing the cadence in plain language, so a site can stay current without anyone re-running it by hand.
Terms worth knowing.
- Sites
- A ChatGPT Work feature that builds and deploys an actual live website from a chat prompt, with its own URL, persistent data storage, file storage, authentication, and analytics.
- Work
- ChatGPT's more agentic mode, used instead of standard Chat for longer tasks. It can use tools and take a series of actions to deliver a finished result.
- Skill
- A packaged, reusable tool or workflow that can be installed into ChatGPT and invoked with an @ mention, similar to installing a plugin from a code repository.
- Scheduled Tasks
- ChatGPT Work's automation feature. It runs a task described in plain language on a recurring schedule, such as processing new emails every day at a set time.
Things they pointed at.
04:44toolLast 30 Days research skill
05:15toolScrape Creator
Lines you could clip.
00:00
“Sites is a feature in ChatGPT Work that lets you build and deploy actual websites and apps directly from ChatGPT.”
13:01
“One way I've started thinking about sites is that they're kind of the last step of a bigger task in ChatGPT.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
Sites is a feature in ChatGPT Work that lets you build and deploy actual websites and apps directly from ChatGPT. They can have persistent data, file storage, authentication, analytics, and their own live URL. I'm going to start with a really simple example, then build up to some more interesting ones, turning existing work into an interactive site, adding an interface to a skill, storing data long -term, and automatically updating a living site every day with an automation.
Now, this Sites feature is only available in Work or Codex. I'll be focusing on Work here. So if you're on the web app, go up top, switch from Chat to Work, and you're good to go.
And if you don't know, work is basically ChatGPT's more agentic mode for larger tasks. It can work for longer, use tools, and take a series of actions to deliver a finished result. Taking the capabilities of codecs and making them easy for anyone to use.
And once you're in work, anytime you want to create a site, all you have to do is type out what you want, then tag it. Hit at, sites. Then it will call that tool and build from there.
And just from that silly prompt I had, in nine minutes, it built Capybara protocol. It's still private, so I log in with my ChatGPT account. And this is now a live URL I can go to and play.
And it seems to be working well. Then if I want to make any changes to the site, I can just ask for them. Then I reload the site and it's updated.
And just like that, I have an actual live website. It's hosted by ChatGPT. That's all built in and taken care of already.
From here, I can click Share, then I'm able to make it public if I want, or keep it private so I have to keep logging in with ChatGPT to access it. Then I can click Settings for some more options. The beginning of the URL is customizable, so I can rename it.
I could add a custom domain if I own one already. I can also add environment variables if the site needs it. Then there's also an Analytics tab to track how many visitors you get.
That's the basics, just tag Sites and it will build and deploy a website for you just like that. It's fun that you can make a game this way, and you could make it much more complex than this one, but really if you want to make a good game you'd want to do that in Codex or Cloud Code. So let's do some practical examples and show what I think Sites is actually the best solution for.
One of my favorite uses for sites is taking work you've already done in ChatGPT and turning the result into something you can actually interact with. It might already have your conversations, uploaded files, research, links, data, and decisions. Instead of ending with a giant summary or a static document, you can turn all of that context into a site.
I'm going to reuse an example from a recent video I made all about ChatGPT work, because this one demonstrates that really well. We were planning a trip to Disneyland. In this project, I've had conversations going back and forth to pick a hotel, figure out which rides to prioritize, find unique things we don't know about, and just a bunch of other things.
Instead of asking for a summary of everything and just getting back a wall of text, I'm going to start a new chat in this project, and I can say, compile everything we've talked about in this project into an interactive and useful, then I'll tag, site that we can use for our trip. Now this got started and it was smart enough to realize that I had already made a site this same way.
So it decided it's going to just update the site it already made rather than make a duplicate. But it's updating it with some newer conversations we've had. So here is the site.
It's really helpful. I'd had a lot of back and forths and gathered really good information across multiple conversations. And it's hard to utilize that when it's just a long chat thread or multiple threads.
This is much more organized and useful. From there, I can share that with my wife and someone else we're meeting there. So that is super simple to make a site like this, but it's very useful.
And that could be for something you're working on with your team, some data analysis you've done, helping you to make a decision. You can turn research you've been doing into a decision -making tool. For an example, I made this quick RV buyer's guide where you adjust things like budget, family size, towing, comfort, and storage.
Then it dynamically changes which type of RV makes the most sense for you. So that's another just quick example. Let's move on to something more substantial.
I'm just diving into the sites feature in this video, but there's a lot more to ChatGPT work. I have a full guide we put together in the description that covers everything else. All the other features and techniques.
Work is really powerful. We also included nine example use cases with full prompts you can copy paste so you can get started right away. Grab that for free using the link in the description.
For this next demo, I'm going to take a skill that was built by someone else, but I'm going to add steps to the skill and turn the results into a more useful site that includes persistent storage of data. The skill I'm going to use is called Last 30 Days. It's a free skill you can find on GitHub.
I'll link it below as well. It's an agent skill that goes out and finds what people are actually saying about a topic right now. Instead of relying mostly on normal web search, it searches across YouTube transcripts, Reddit.
Polymarket. It will use a tool called Scrape Creator to pull from Instagram and TikTok and a bunch of other places, and then it scores and filters what it finds based partly on how much people are engaging with it. Then it synthesizes everything into a short report.
The skill is awesome, and I've been using it for a while, but it collects way more useful information than what fits into that final summary. There are individual posts, use cases, weird opinions, and sources that I might want to dig into more. So instead of just Throwing that all away after the summary, I'm going to build a site around it.
The first step is to download the skill to ChatGPT. That's really easy. To install it, I'll just tell ChatGPT to use the skill creator to add the skill, and I'll give it the GitHub URL.
The built -in skill creator it has already knows how to package the repository into a ChatGPT skill. So that's all I need to do. And I already had this as a skill, but it did actually create a brand new one for me.
So I'm going to open up a new chat and use that. To call one of your skills, hit at, then select it. Then all I need to do is type in a topic I want it to research.
How about Grokbot? It did its thing for about five minutes and came back with a very helpful report summarizing what people are saying about it and how they're using it right now. That is awesome, but I want a way to make this information even more useful for my purposes and easier to read.
I'm going to ask for what I want and get a rough version of the site using this information that came back. I want the summary, but also use cases listed out and unusual takes, since I often use this to find aspects of a topic I might have missed. I usually test something and use it for myself for a while, then when it comes time to make a video, I also want to make sure I cover uses that would apply more broadly, rather than only if you happen to do similar things to me.
You know, I just try to include all of that. The other part is I want it to be able to link out to all the original posts easily and also be able to see them embedded without having to leave the site whenever that's possible. So I listed out all of that.
You could, of course, customize that to whatever you want to be able to see. I've built out the whole site using the information it got from that first run. And the site is looking great.
The main summary page is nice. Then the use cases tab is very helpful with these filters by role. Then unconventional takes is great to have.
I can also see the original posts. And for the sources it's possible to embed, I can click Show Post and see it right here. This was to make sure I'd like it before adding the full functionality.
I probably will come back and change some things later, but the core is good for now to finish the demo. Next, I need it to be able to add new topics and also save the history. So I sent a very simple prompt explaining that, and it came back with the history functionality there.
And it has the option to add a research run, but a site can't actually run a skill. It wants me to run the skill separately, then copy the JSON from that and input it into the add section. I don't want to have to do that, so I'm going to tell it that.
I'll ask it to figure out the best way to update the skill. I want to just run the skill, then have the skill automatically add the information to the site itself at the end of the process. So I'll send that and let it figure it out.
And it actually left the name of the site as Grokbot Research still. So I'm going to change that and also change the URL to Research Lab. It made the update, so I'm going to see if this works now.
I'll do chat GPT work, sites, use cases, then tag the skill. It actually says it was blocked from publishing it unless I explicitly authorize it. That's strange, but I'll do that really quick.
Then I can actually see the issue right here while it's thinking. It's because I changed the URL to the site, but it had already hard -coded the previous URL. That makes sense.
It was able to update the research library now. Since it already discovered the issue, I'll just say update the skill so it will work on the first try next time, and it should be good. Now that came back, so I will open up a new chat and try one more time with Fable 5 .1.
Send that. The skill is finished running, so I'm going to open up the site and click refresh. And there it is.
The new research run plus the previous two are all in here now. So now anytime I call that skill to research a topic, when it's finished running, I open up the site and it's all there. The site has persistent storage so these research runs don't disappear when the conversation ends.
I can keep adding topics and build up a research library over time. I wanted to use that example because it was a little more involved and also needed some troubleshooting. I usually think it's helpful to leave that in, especially in this case to show I didn't need to go digging through the skill myself to figure out what went wrong.
Work. figured out what broke, fixed it, and I could tell it to just update the skill so the same problem wouldn't happen again. And hopefully that example shows you ways to get a little creative and expand what sites can do.
For the last example, I want to really use that persistent storage and combine it with an automation. I've subscribed to a bunch of AI newsletters, but I don't want to read 10 versions of the same news story. What I actually want is one searchable database that removes duplicates, tracks the important stories, and pulls out useful workflows or use cases buried inside each newsletter.
And because new newsletters get sent every day, I want the site to keep updating itself automatically. For this one, I'm just going to show some of the first steps I did rather than build out the whole thing again. Then I'll set up the automation for it.
I explained what I wanted, all the things I just described, but I also wanted it to backfill it with some of the history. I have a separate email that's subscribed to all these newsletters, so I asked it to pull in some of that. This actually shows another nice feature they added very recently, that you can connect multiple gmails to your ChatGPT account.
These don't go to my main email I use in here, so it's nice to be able to add multiple now. I didn't specify how far back I wanted it to go, so it just went for all of them. It pulled in the whole catalog of over 2 ,000 newsletters and extracted each of the use cases it found and populated the database of the site with all that information.
So at least for... text -heavy use cases like this, you can store a pretty substantial amount. I looked and couldn't find a specific storage limit that OpenAI has published for these.
I assume it's not unlimited, especially if you're storing like large files like videos, but a database with thousands of text records worked no problem here. Also, ChaffGPT really seems to be liking this green color. I'm just showing setting up the functionality of these.
You can definitely fine -tune the aesthetics really easily once the functionality is there. You can see when I search for something like ChatGPT sites.
It shows me each of the workflows and use cases that have been covered about them. Then if I switch over to the first tab, it's the news stories, usually around when something was first released. And it also does track mentions of tools over time and has a saved section.
But these first two tabs are the main ones I'll be using. So that is really useful for me for the same reason I mentioned in the last demo, just to be able to see other perspectives on the tools rather than my own little bubble of how I've been using them. But from here, I need to be able to populate this with the new newsletters that come in.
So I'll set up an automation. Before I do that, I do always want to make sure it can do it right first. So I will say intake and process the new newsletters I've received into the site.
So that did it. So I'll go to the site. There's 2081.
I'll refresh. And now there's 2094. So that worked.
Now I want to set up an automation that adds the new newsletters automatically every day. And the automation feature in work is called scheduled tasks. And to set one up, just like most of these steps, all you have to do is ask.
Now set up a task to process the new newsletters at noon every day. Just about every newsletter will have come in for the day by noon. And just like that.
It's all set. You can open up all your scheduled tasks on the sidebar and this will run every day. So this site will be consistently updated with all the newsletters and they'll be separated out into the different categories I have that make it searchable and more useful for me.
And that history will be stored and stay there over time. One way I've started thinking about sites is that they're kind of the last step of a bigger task in ChatGPT. You do the research, analyze files, connect your data, and build a workflow.
Then instead of leaving the results just buried in a chat, you can turn it into an interface you can actually keep using. And once you combine that with persistent data and scheduled tasks, these start becoming actual tools.
The Hook
The bait, then the rug-pull.
ChatGPT Work has a feature most people never open: Sites, which turns a chat prompt into an actual live website, complete with its own URL, storage, and login. This walkthrough builds one from a single prompt, then pushes it further into a research tool, a skill's front end, and a self-updating database.
Named ideas worth stealing.
13:01list
The Sites workflow
- Research
- Analyze files
- Connect data
- Build a workflow
- Sites
Sites is framed as the last step of a bigger ChatGPT Work task chain: gather information, work with files, connect data sources, and build an automated workflow, then turn the output into a persistent, usable interface.
Steal forany internal-tool or dashboard project where the 'last mile' of turning finished research into something usable keeps getting skipped
How they asked for the click.
VERBAL ASK
04:17link
“Grab the free guide to ChatGPT Work, with nine example use cases and full prompts, using the link in the description.”
Soft mid-video plug for a companion lead-magnet guide, delivered as a break between the simple game demo and the more advanced examples, not a hard sales pitch.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
