🤖 Part 20: Reusable Assistants — Projects, Custom GPTs & Gems

Part 19 taught the assistant about you. This part is about a specific task you do again and again — polishing emails, drafting social posts, summarizing reports in a set style. Instead of re-explaining it every time, you can package the instructions and reference material once into a reusable assistant, then open it like a saved tool. Names and availability vary by product and plan, so we'll keep it honest and grounded.

📚 What You'll Learn

By the end of this part, you'll be able to:

Tell apart the main reusable-assistant features — ChatGPT Projects & Custom GPTs, Claude Projects, Gemini Gems — and what each roughly does
Give a reusable assistant standing instructions plus reference files so you stop re-explaining
Judge when it's worth building one versus just using a normal chat
Understand sharing and the honest free-versus-paid picture

In This Part

What a Reusable Assistant Is

A normal chat starts empty and disappears into your history when you're done. A reusable assistant is a saved workspace with a job description baked in: you write the instructions once, optionally attach reference files, give it a name, and from then on you open it and get straight to work. It's the difference between explaining a recipe to a new cook every night and handing them a laminated card they already follow.

📖 Definition

Reusable assistant: a saved configuration of an AI assistant that bundles standing instructions (how it should always behave for this task) and, usually, reference files (documents it should always draw on), so you don't re-supply them each time. Depending on the product it's called a Project, a Custom GPT, or a Gem.

Two ingredients do most of the work. Standing instructions are like custom instructions from Part 19, but scoped to one task instead of all your chats — "you polish emails to sound warm and professional," say. Reference files are documents you attach once — your style guide, a product FAQ, a list of your services — that the assistant can read every time without you re-uploading them.

graph LR A["✍️ Standing instructions
the assistant's job description"] --> C["📦 Reusable assistant
saved & named"] B["📎 Reference files
style guide, FAQ, notes"] --> C C --> D["🔁 Open it anytime
no re-explaining"] C --> E["👥 Optionally share it
⟨varies by product/plan⟩"]

The Landscape: Projects, Custom GPTs & Gems

Each big assistant has its own version of this idea, with its own name and its own rough shape. The table below is a map to get you oriented — but as always in these fast-moving parts, treat the specifics as "roughly true today," not a fixed spec.

Product	Feature(s)	Roughly what it's for
ChatGPT	Projects; Custom GPTs	Projects group related chats with shared instructions/files; Custom GPTs are named, shareable mini-assistants for a specific job
Claude	Projects	A workspace with standing instructions and attached reference material for a body of related work
Gemini	Gems	Custom, named assistants you set up once with instructions for a recurring task

⚠️ Important: The names, the exact capabilities, whether files can be attached, how many you can make, and whether a feature needs a paid plan all vary by product and tier and change often. Some of these live behind a subscription; some have free versions with limits. Don't take any single detail here as permanent — when you go to build one, check the assistant's current interface and official help page. The concept transfers everywhere even as the buttons move.

The good news: because they share the same two ingredients — standing instructions and reference files — the skill of building a good one is identical across all of them. Learn it once on whichever assistant you have, and you can rebuild the same helper anywhere.

⚠️ Watch Out — "Project" means different things

In ChatGPT and Claude, a "Project" is a workspace for a body of related chats. A ChatGPT "Custom GPT" is a more packaged, shareable mini-assistant. They overlap but aren't identical, and the line between them shifts as products evolve. If a menu doesn't match your expectation, don't worry about the label — look for the place to add standing instructions and attach files, and you're in the right spot.

When It's Worth Building One

A reusable assistant is setup cost you pay once to save time forever — so the math is simple: build one when the task repeats and the setup is non-trivial. For a one-off question, a normal chat is faster. For the thing you do every week with the same requirements, a saved assistant earns its keep quickly.

✅ Worth building one	⚠️ Just use a normal chat
A task you do weekly with the same rules (polishing emails, formatting reports)	A one-time question or a task you'll never repeat
Work that always needs the same reference material (style guide, FAQ, product list)	Something where the context changes completely each time
A helper you want a teammate or family member to reuse exactly	Quick exploration where you don't yet know what you want
Anything where you keep re-typing the same long setup	A task your Part 19 custom instructions already cover

✅ Tip — let a reusable assistant grow out of a good prompt

The easiest ones to build are the ones you've already been doing by hand. If you've got a favorite prompt you keep pasting — a big block of "rewrite this email to be warm, concise, and professional, American English, keep my meaning" — that block is your standing instructions. You've already done the hard part; you're just moving it into a saved home so you never paste it again.

Sharing & the Free-vs-Paid Picture

One reason reusable assistants are appealing is that a good one can be handed to other people — a family member, a coworker, a whole team — who then get your carefully-tuned helper without knowing anything about prompting. How sharing works, and how far it reaches, depends on the product and plan.

Sharing style	What it usually means
Private (just you)	The default; only you can open and use it
Link / specific people	Anyone with the link, or people you invite, can use it (they may need their own account)
Team / workspace	Shared across an organization's account, often with admin controls (typically paid)
Public / directory	Some products let you publish a Custom GPT-style assistant more broadly

⚠️ Important: When you attach reference files to a reusable assistant and then share it, you may be sharing the contents of those files with everyone who can use it. Before you share, ask: would I be comfortable if each of these people read every document I attached? If not, remove the sensitive material or keep the assistant private. This connects straight back to the privacy habits in Part 13.

On free versus paid: honestly, this is one of the more paywalled corners of the assistant world. Building reusable assistants — especially the more powerful, shareable, file-backed kinds — often sits on paid tiers, though lighter versions may be free, and the line moves. You can absolutely follow the concepts here for free; whether you can build the exact feature named depends on your account. Check the current pricing page before assuming, and remember from Part 1 that you never need to pay to get real work done — a saved prompt in a notes app achieves much of the same benefit at zero cost.

📋 Templates & Examples

Here's a complete, copy-ready instruction set for a "Polish My Emails" helper. Paste it into the standing instructions of a Project, Custom GPT, or Gem — or just keep it in your notes and paste it at the top of a normal chat.

Role: You are my email-polishing assistant. When I paste a rough email or bullet points, you return a clean, ready-to-send version.

How to respond: Keep my meaning and any facts exactly; never invent details, names, or commitments I didn't include. Make it warm but professional, clear, and concise. Use American English. Match the formality to my draft — don't make a casual note stiff. Keep it short unless I ask for more.

Format: Give me (1) the polished email, then (2) a one-line note on anything I should double-check (a date, a name, a claim). If a fact seems missing, ask me rather than guessing.

Tone: No corporate jargon, no hype, no filler openings like "I hope this email finds you well" unless I ask for them.

Optional reference file to attach: a short "voice notes" document — a few emails you've actually sent that sound like you — so the assistant matches your real style. Keep any real recipient names or private details out of it.

🛠️ How To: Build a "Polish My Emails" Helper

What you'll do: turn a recurring task — cleaning up emails — into a saved assistant you open and use, instead of re-explaining every time.

Step by step

Find the feature. In your assistant, look for "Projects," "Custom GPT," "Gems," or "Create." If you can't build one on your plan, don't worry — you can do every step below in a normal chat by pasting the instructions at the top; the payoff is just slightly smaller.
Name it. Something obvious like "Polish My Emails" so you find it instantly later.
Paste the standing instructions. Use the template in the Templates & Examples card above (or your own version). This is the assistant's permanent job description.
Attach a reference file, optionally. Add a short "voice notes" document of a few emails that sound like you — with private details removed — so it matches your style.
Test it with a real, messy email. Paste in something rough and see if the output is genuinely send-ready. If not, tweak one line of the instructions and try again — this is how you tune it.
Save it, and decide on sharing. Keep it private unless someone else needs the exact same helper. If you share it, re-check that any attached file is safe for those people to see.

💡 Tip

Build the simplest version first and use it for a week before adding anything. Reusable assistants get worse when overloaded with rules that fight each other. If you notice the same correction twice, that's your signal to add exactly one line — not ten. A lean helper you trust beats an elaborate one you have to babysit.

Best Practices

✅ Do's

Build for tasks you actually repeat — the setup cost only pays off when you reuse it often.
Give clear standing instructions and one or two focused reference files, not a pile of documents.
Test with a real example and tune from what it gets wrong, one line at a time.
Re-check attached files before sharing — sharing the assistant can mean sharing those files' contents.

❌ Don'ts

Don't build one for a one-off — a normal chat is faster when you'll never reuse it.
Don't assume the feature is free or named the same everywhere — availability and labels vary by plan and change; check the official page.
Don't overload it with contradictory rules or too many files; it gets less reliable, not more.
Don't attach sensitive documents you wouldn't want stored or shared (revisit Part 13).

💡 Pro Tips

Keep your standing-instructions text in a notes app too, so you can rebuild the same helper on any assistant — free or paid — in minutes.
Make a tiny family of helpers for one workflow ("draft," "polish," "summarize") rather than one giant do-everything assistant; each stays sharp at its own job.

📓 Learning Journal

Keep a journal as you work through this guide — digital or paper. After each part, jot down:

Key ideas you learned
Things that clicked for you
Questions or confusion points to revisit
Ideas you want to try
Your progress and how you feel about it

✍️ This part's prompt: Name one task you do at least weekly with roughly the same requirements every time. Draft the standing instructions for a reusable assistant that would handle it — role, how to respond, format, tone. Then list the one or two reference files it would need, and note whether any of them contain something you'd have to strip out before attaching or sharing. Would this genuinely save you time, or is it a task a normal chat already handles well? Write down your honest verdict.

📝 Part Summary

🎓 Key Takeaways

A reusable assistant bundles standing instructions and reference files so you never re-explain a recurring task.
The features go by different names — ChatGPT Projects & Custom GPTs, Claude Projects, Gemini Gems — but share the same two ingredients, so the skill transfers.
Build one when a task repeats and the setup is non-trivial; for one-offs, a normal chat is faster.
Sharing can expose attached files' contents, and building these often sits on paid tiers — names, limits, and pricing vary and change, so check the official page.

🎉 What You Can Now Do

You can take a task you've been doing the slow way — pasting the same long prompt, re-uploading the same files — and turn it into a saved helper you open in one click. You know when that's worth it and when it isn't, how to keep the instructions lean enough to stay reliable, and what to double-check before you hand a helper to someone else. Even with no paid plan, you've got the pattern: a good prompt saved in your notes gets you most of the way there.

❓ Common Questions

Do I need to pay to build one of these?

Often, yes — the more powerful, shareable, file-backed versions frequently sit on paid plans, though lighter versions may be free and the line shifts. You can still get the benefit for free by saving a strong standing- instructions prompt in a notes app and pasting it into a normal chat. Check your assistant's current pricing page for what your account can actually do.

What's the difference between a Custom GPT and a Project?

Roughly: a Project is a workspace for a body of related chats that share instructions and files, while a Custom GPT is a more packaged, named, shareable mini-assistant built for one job. They overlap, and the exact line moves as products change. In practice, look for where you add standing instructions and attach files — that's the part that matters — and don't overthink the label.

If I share a reusable assistant, can people see my files?

Potentially, yes. Reference files you attach are part of what the assistant draws on, so sharing the assistant can effectively share those files' contents with everyone who can use it. Before sharing, remove anything private or keep the helper to yourself. Treat attached files as if the people you share with will read them.

🔭 Up Next

In Part 21: Voice, Vision & Real-Time, we step away from typing entirely: talking to an assistant hands-free, showing it things through your camera, and holding live, back-and-forth conversations — with an honest map of where these features are free, where they're paid, and how fast they change.

📚 Additional Resources

🌟 Keep Going

You just learned to build a tool out of a tool — packaging your best setup so it's one click away forever, and shareable when it should be. That's a real force multiplier. Next, let's put down the keyboard and start talking to, and showing things to, your assistant. 🤖