Technology User Experience

Stop Re-Explaining Yourself to AI — Train It Once, Use It Forever

Every time you open a new chat and spend the first five minutes telling the AI who you are, what you do, what tone you want, what to avoid — you’re doing something you should never have to do more than once.

That’s the thing most people miss about working with AI. They treat it like a vending machine. Put in a coin, get something out, walk away. Next time, put in another coin, start over. It’s inefficient. And the better tools — the ones actually worth your time — are built for something completely different.

Gemini Gems, Claude Skills, and ChatGPT’s custom GPTs exist for exactly this reason. Train your AI once for a specific task, and it shows up ready every single time. No re-prompting. No context-setting. No explaining yourself from scratch.

I cover the full setup process in this video — watch it here — but I want to use this post to add some context around why this matters and what each platform actually does differently.

The Re-Prompting Problem Is Costing You More Than You Think

Here’s a scenario most people recognise immediately: you need the AI to write in your brand’s voice. So you write out a detailed prompt — the tone, the audience, the dos and don’ts, maybe even an example. You get a great output. Then tomorrow you open a new chat, and you have to do the whole thing again.

Multiply that by every task you use AI for. Writing. Research. Summarising. Client emails. Social captions. Strategy documents. Every new chat wipes the slate clean.

McKinsey’s research has estimated that AI tools, used correctly, can save knowledge workers an average of 20 hours per week. The word doing heavy lifting there is correctly. Most people aren’t using them correctly — they’re using AI assistants the same way they’d use a calculator. One input, one output, done.

The trained AI model changes that equation completely. It’s the difference between a tool and a workflow.

What Each Platform Actually Gives You

All three platforms — Google, Anthropic, and OpenAI — have versions of the same idea. But they’re not identical, and knowing the differences helps you decide where to build.

Gemini Gems are Google’s version. You give your Gem a name, a role, a set of instructions, and any knowledge files it needs to draw from. Once it’s set up, every conversation with that Gem starts with that context already loaded. If you’re deep into Google Workspace — Gmail, Docs, Drive, Slides — Gems have a meaningful edge because of the native integration. The limitation worth knowing: sharing Gems with others is still restricted compared to the other platforms.

Claude Skills (accessed through Claude’s Projects) are where I’ve seen the most consistent quality, especially for writing and anything requiring nuanced instruction. Claude tends to follow detailed custom instructions more faithfully than most. You can upload knowledge files, set a persona, specify exactly how it should respond — and it holds that context reliably across long conversations. If your work involves a lot of writing, editing, or professional communication, this is worth setting up first.

ChatGPT custom GPTs have the most mature ecosystem of the three. OpenAI launched this in late 2023, and there are now over 3 million custom GPTs built by users. The GPT Store means you can share your GPT publicly or access ones others have built. It also has the broadest app integration — if your workflow needs third-party connections, ChatGPT is currently ahead. The tradeoff is that following highly detailed instructions can sometimes be less precise than Claude.

None of these is universally “best.” Each has a home. Build where your workflow already lives.

How to Think About What to Train

This is the part most tutorials skip. They tell you how to build. They don’t tell you what to build — and that’s actually the harder question.

The mistake people make is building a generic assistant. “A GPT that helps me with marketing.” That’s too broad. You already have that — it’s called ChatGPT. The value of training is specificity.

Ask yourself: what task do I repeat most often that requires the same context every time? That’s your starting point. Some real examples that work well:

  1. A writing assistant trained on your specific brand voice — with examples of your best content, a defined tone, words you never use, and audience context baked in.
  2. A client briefing assistant — trained on your service offering, pricing logic, and how you qualify work, so it helps you draft proposals and respond to briefs faster.
  3. A content repurposing assistant — takes a long piece of content and knows your format preferences for each channel without you explaining it every time.
  4. A research assistant for a specific industry — knows your sector, the key players, the terminology, and your typical research objectives so it skips the groundwork.
  5. A feedback and editing assistant — trained on your editorial preferences so it edits consistently, not randomly.

The narrower you define the task, the more useful the trained model becomes. I’ve seen this work for clients across retail, professional services, and media — the ones who built narrow, specific assistants got more value faster than the ones who tried to build one assistant to rule everything.

What the Setup Actually Looks Like

I walk through this step by step in the video, but here’s the short version so you know what you’re getting into before you watch.

The process for all three platforms follows roughly the same logic:

  1. Define the role — Who is this assistant? What’s its job? Be specific. “You are a brand copywriter for [Company Name]. Your job is to…” is a better start than “Help me with writing.”
  2. Write the instructions — Tone, constraints, audience, format preferences, what to always do, what to never do. The more specific, the better the output.
  3. Upload knowledge files — Any documents, guides, examples, or reference material the AI needs. Brand guidelines. Past work samples. Product information. FAQs.
  4. Test and refine — Run your most common tasks through it. See where it drifts from what you want. Adjust the instructions. Repeat.

The whole process takes under an hour for a well-defined use case. The payoff compounds every single time you use it.

Once your AI knows your business, your voice, and your standards — it stops being a tool you use and starts being a team member you brief.

Frequently Asked Questions

What’s the difference between a Gemini Gem and a Claude Skill? Both let you create a customised AI assistant with persistent instructions and uploaded knowledge. The main differences are in ecosystem fit and instruction-following quality. Gems are better if you live in Google Workspace. Claude Skills tend to follow detailed writing and communication instructions more reliably. ChatGPT custom GPTs are the best option if you need app integrations or want to share your assistant publicly.

Do I need to pay for a premium plan to use these features? No. Not anymore at least but these things change all the time.

How long does it take to train an AI assistant this way? For a well-defined, single-purpose assistant, you can have a working version in 30–60 minutes if you’re doing it alone but under 10 minutes with a skill trianed to help you build skills hehe. The setup itself is fast — most of the time goes into thinking through what you actually want it to do and writing clear instructions. The video linked above walks through the process live so you can see what it looks like in practice.

Can I build more than one? Yes, and you should. Most people who use this properly end up with several — one for writing, one for research, one for client work, and so on. Each trained on a specific context. That’s where the productivity gains really stack up.

The Bottom Line

If you’re using AI daily and still starting every conversation from scratch, you’re leaving the best part of these tools unused.

The trained assistant — whether that’s a Gem, a Skill, or a custom GPT — is what turns AI from something you interact with into something that actually knows how you work. That’s not a small upgrade. For most people I’ve worked with, it’s the moment AI stops feeling like a gimmick and starts earning its place in the day.

And if you want help figuring out which tools make the most sense for your specific business, or how to implement AI in a way that actually sticks —then reach out to me, I’d be happy to help.