Built for this crew.
Nobody here needs to become an AI expert. The point is that the next time you've got a contract to read, a spreadsheet to count, or a flyer to draft, you know which button to press and why. That's it.
Who you are and what you use
Start with these three.
Chat, file upload, image upload. Ten minutes and you've got the hang of it.
What are you trying to do?
Say it the way you'd say it to a coworker. We'll point you to the feature that fits, show you a prompt if we have one that's close, and give you a button that opens it in ChatGPT or Claude.
Two words you'll see a lot: a prompt is just what you type to the AI. A model is the version of the AI that answers. Before you start: sign in with your CDP work account. If a feature here doesn't show up for you, your account may not include it yet.
Browse instead.
Pick what you have, or what you want back, and we'll point you at the feature.
See every prompt on the site
The whole library.
Five prompts to steal.
Five jobs worth handing to AI again and again. Each one needs something to work on, so give it that first: attach a file, paste your material, or describe the situation. Then add the prompt.
The one idea to keep: don't just ask AI questions. Give it work. Goal, context, task, limits.
Make it last.
ChatGPT and Claude don't give you unlimited use. Here's what gets used up, how to stretch it, and what to do if you run out. What you see below matches the tool you picked at the top: ChatGPT.
Tokens and limits.
AI reads and writes in tiny pieces called tokens. Think of each one as a coin. A token is about one short word, or part of a longer word:
Everything costs coins: what you type, the files you attach, and everything it writes back. Spend more coins and you hit your limit faster. Neither tool tells you how many coins you have, though. You only see a percentage.
Claude gives you two limits. One resets every five hours. The other resets once a week. They're just yours, so if you hit your limit, your coworkers can still keep going.
To see where you are: go to Settings, then Usage. You won't see a number of tokens or messages. Claude only shows a percentage of each limit used and when it resets.
- Claude counts up. The bar fills as you use it. 100% used means you've hit that limit.
- Current session is the five hour limit. Weekly limits is the one that resets once a week.
- Your numbers will be different. These are just to show the layout.
Chat and Work are counted separately. Chat has its own limits. Work has its own allowance with two limits: one that resets every five hours and one that resets every week. You need some left in both to keep going.
To see where you are: go to Settings, then Usage. You won't see a number of tokens or messages. ChatGPT only shows a percentage left and when it resets.
- ChatGPT counts down. The bar shrinks as you use it. 0% left means you've hit the limit.
- Chat isn't included. This is for Work and the other tools named at the top. Chat has its own limits.
- Two limits: one resets every five hours, one resets weekly. You need some left in both to keep going. The wording on your screen may be a little different.
- Your numbers will be different. These are just to show the layout.
Squeeze it down, or start fresh.
When a chat gets really long, Claude squeezes it down on its own. It writes a short summary of the earlier messages so the chat can keep going. That's called compacting. Your full chat is still saved, and Claude can still look back at it. (This only works when code execution is turned on in your settings.)
The catch: a chat that's long enough to get compacted uses more of your limit. Claude's own advice is to start a new chat if you're getting close to your limit in a long one.
The longer a chat gets, the more coins each new message costs, and in a really long chat it can lose track of things you said early on. When that happens, don't keep pushing. Start fresh with New chat at the top of the sidebar.
- In the old chat, ask for a recap.SAY THIS
Give me a short recap of this chat: what we're working on, what's been decided, and what's still left. Keep it under 150 words so I can paste it into a new chat.
- Copy the recap it writes back.
- Open a new chat, paste the recap, then ask your next question.
Start a new chat when: the topic changes, it starts forgetting things you told it, answers are getting slow, or you're close to your limit. Is it ongoing work you keep coming back to? Put it in a Project instead.
Seven easy habits.
- Give it everything up front. Say what you need, why, and who it's for in your first message. Less back and forth means fewer coins.
- Ask for a few things at once. Got three questions about the same thing? Send them in one message.
- Keep files you use a lot in a Project. In Claude, files in a project count less each time they get reused, compared to uploading them again.
- Only attach what it needs. If you only need one tab or one section, don't send the whole thing.
- Use the lighter setting for easy stuff. Save the bigger model or higher thinkingeffort for hard problems. Higher doesn't always mean a better answer.
- Save Deep researchResearch for big questions. For one fact, plain web search is quicker and uses less.
- Read it over before you send. A clear message the first time saves you a do over.
- Check your usage before a big job. Settings, then Usage. If you're almost out, save the big job for after the reset.
Don't panic. It comes back.
- Check when it resets. Go to Settings, then Usage. The five hour limit comes back the soonest.
- Wait it out, or do the parts that don't need AI. Your limit is yours alone, so coworkers aren't affected.
- Check which limit you hit. Go to Settings, then Usage. It shows which one ran out and when it resets.
- Wait for the reset. If you have a saved reset, you can use that instead. Meanwhile, do the parts that don't need AI.
- Switching models won't help. The models in Work share one allowance, so changing the model doesn't give you more.
- Work ran out? Chat has its own separate limits, so you can usually keep going in Chat.
From OpenAI's and Claude's own help pages, checked October 2026. Limits change, so Settings, then Usage is always the real answer.
What to look for
Tap the one closest to what you're doing.
You'll get a prompt written the way you'd ask a coworker, and a button that opens it in your tool.
Claude shows a caution banner for any prompt that arrives by link. That's normal. Read the prompt in the box and send it.
Walk me through it, step by step
Grab something from your real work. Nothing fancy.
Pick something real.
Go do it in ChatGPT.
Quick tip: it does a better job when it knows the situation. Tell it the goal (what you're trying to do), the context (what happened, who it's for), the task (what to make), and the limits (what to avoid).
Where is this feature in ChatGPT?
One more nudge.
Stay in the same conversation and push it one step further. This is where it gets good.
No score or test. Your progress stays in this browser. Use real files you'd be comfortable working on in the tool, and skip resident or financial details unless you know that's allowed.