Why I Don’t Use Generic AI Prompts, Skills and Libraries

Generic AI prompts, Claude Skills, Gemini Gems and pre-built GPT instructions can save time, but they also introduce somebody else’s assumptions into your workflow. The better approach is to understand the task, build context around the specific project, test what works and develop your own AI process rather than relying on large prompt libraries or default instructions you don’t fully understand.

In this post

• Why I don’t rely on generic AI prompt libraries
• Why Claude Skills, Gemini Gems and Custom GPTs can shape your output in ways you may not notice
• Why huge prompt collections aren’t the same thing as expertise
• How I test supposedly “life-changing” AI prompts
• Why real AI expertise comes from experience, judgement and context
• How I build AI workflows around the individual project
• Why I’ll borrow useful ideas without handing over control of how I work

I don’t collect prompt libraries, install piles of Claude Skills written by other people or use the default Gems in Gemini just because they’re there.

I’m equally wary of somebody else’s Custom GPT, somebody else’s ChatGPT Project instructions, and those enormous “ultimate prompts” that supposedly turn ChatGPT into a world-class copywriter, SEO expert, business strategist, psychologist, marketing department and somehow a part-time plumber as well.

It’s not that I think all of these things are useless.

Some will be very good – at different tasks (I talk about how I use AI tools and LLMs here).

Some of these prompts and skills will have been created by people who genuinely know what they’re doing. I just don’t particularly want them deciding how I work.

Every project I work on is different and every client is different.

The objectives are different, the audience is different, all the background knowledge is different and, obviously, the commercial reality is different.

Even when I’m writing for myself opens in a new tab, one article might need research, another might be almost entirely personal experience, and another might start with a ridiculous observation that wandered into my head while I was making a cuppa or walking our dog.

Why would I want the same giant prompt sitting underneath all of them?

AI is supposed to be incredibly adaptable. People are constantly being told how clever these systems are at understanding context, following conversations and adapting to different tasks.

Then somebody comes along and says you need their 4,000-word master prompt before you can use it properly.

Really? Erm… I’m gonna call BS on a lot of it here.

I’ve Tested AI Prompt Libraries Myself

I’m not completely dismissing this stuff without looking at it. I’ve looked.

Over the past few years, I’ve occasionally copied prompts from Facebook groups, samples from prompt libraries and various collections of supposedly life-changing AI instructions.

Some of them are ENORMOUS.

Pages and pages telling ChatGPT exactly what role it should assume (which is completely unnecessary if you use it correctly), how it should think (erm… really?), how it should structure an answer, what questions it should ask, what it must never do and what an outstanding result supposedly looks like.

I’ve sometimes pasted these prompts into ChatGPT or Gemini and asked the system to analyse the prompt itself.

What’s good about it?

What’s unnecessary?

What contradicts something else?

Could it be shorter?

Does it actually improve the result?

And occasionally I’ve asked another question.

Does this look as though it might have been written by an LLM?

Quite often, the assessment has been that it has many of the characteristics you’d expect from AI-generated instructions.

Now, that proves absolutely nothing.

ChatGPT can’t reliably inspect a piece of text and announce that another AI definitely wrote it. Nor can Gemini. Not 100% accurately. AI detection isn’t reliable enough for me to make that claim. Certainly not entirely reliable enough to mark its own homework.

But it does make me smile.

There’s something wonderfully circular about somebody potentially using AI to produce hundreds or thousands of prompts, then selling those prompts to humans so the humans can use them to tell AI how to produce better content.

At this rate, the whole AI prompt industry could disappear up its own prompt library.

Are Those AI Prompts Actually Tested?

That’s the question that bothers me most.

You see collections containing hundreds of prompts. Sometimes thousands. And apparently they’re all brilliant.

They’re all optimised. They’re all proven (apparently). They’ll transform your business, change your workflow, make your content better and save you countless hours.

Maybe they will.

But how thoroughly has each one actually been tested?

I mean actually using the prompt on real work, trying it across different businesses and audiences, seeing where it produces rubbish, changing it, simplifying it and then trying again.

That’s where I become extremely sceptical.

Creating 1,000 prompts is easy now.

But creating 1,000 genuinely useful prompts that somebody understands, has thoroughly tested and can explain is something entirely different.

Quantity has become incredibly cheap. Expertise hasn’t.

AI Prompts Can’t Replace Real Expertise

You can give somebody an incredible prompt for writing website content. That doesn’t make them a copywriter.

You can give somebody an SEO prompt. That doesn’t suddenly give them years of experience understanding search intent, businesses, customers, websites, conversion, technical SEO or how people actually behave when they land on a page.

You can give somebody a detailed business strategy prompt. That doesn’t mean they’ll recognise when the answer is complete nonsense.

Real expertise shows when you can recognise that the prompt itself is wrong.

That also means recognising when the AI has got it wrong or when the answer is technically correct but completely inappropriate for the client sitting in front of you.

That comes from experience and from doing the work, getting things wrong and seeing what happened afterwards.

No prompt library can download that into your head.

Why I Build My Own AI Workflows

My own use of AI has evolved through using it.

A lot. Tens of thousands of times!

I didn’t sit down years ago and design some enormous master prompt describing precisely how I should interact with artificial intelligence.

I figured things out. Trial and error. Thousands of hours.

If something worked, I kept doing it. If the output was generic rubbish, sounded nothing like me despite the 2 or 3 thousand words I’d given it, looked factually questionable or simply missed the point, I challenged it, checked it or changed the way I was approaching the task.

Over time, that becomes a process.

And the process changes depending on what I’m doing.

If I’m writing about something technical, I might want research and fact checking.

If I’m writing about something deeply personal, I don’t particularly want an AI system marching in with its seven-point content framework and telling me where the key takeaway should go.

If I’m working on a client’s website, I want the AI to understand their business, not mine and certainly not someone else’s (ie: whoever built a prompt).

Not just “business”. That business. Its customers, location, competitors. What it actually sells. What makes money. What customers ask on the phone and what they misunderstand. What the owner hates about their industry.

That context is worth far more to me than somebody else’s ultimate prompt.

The Massive Problem With Other People’s AI Instructions

Somebody else wrote them.

That means somebody else decided what ‘good’ looks like.

They’ve made decisions about what sounds professional, how long an answer should be, what SEO means, how an article should be structured, which questions should come first and what deserves priority.

They may even have decided that certain things shouldn’t be discussed at all.

And I might not agree with any of it.

I want to control the assumptions that sit underneath the output.

This is one of the things I think gets overlooked when people enthusiastically install Claude Skills, Gemini Gems, Custom GPTs and giant instruction sets.

You’re not just borrowing a shortcut. You’re borrowing somebody else’s judgement.

That may be useful. It may also be completely wrong for what you’re trying to do.

The Growing Industry Selling AI Prompts and Advice

I’ve watched plenty of it.

YouTube channels.

Facebook groups.

Prompt engineering groups.

ChatGPT groups.

Claude groups.

Gemini groups.

Midjourney groups.

AI video groups.

Groups teaching you how to create content with AI.

Yawn.

Sorry, I think I nodded off briefly while writing that list.

There are people making very good money ‘teaching’ other people how to use these systems. Not making money using them. Making money after self-appointing themselves as experts by adding ‘AI’ to their username.

Some of them are extremely knowledgeable. Some produce genuinely useful material. Others seem to have discovered that the easiest way to become an AI expert is to announce that everybody else is using AI incorrectly.

Then sell the solution. The sell that solution based on creating fear that you’re using it all wrong.

There’s a slightly awkward truth underneath this whole industry.

If people eventually realise that a lot of this can be learnt by using the systems, experimenting, asking questions, challenging the answers and developing their own methods, the market for secret prompting techniques becomes slightly less exciting.

There’s still plenty worth learning from other people.

I still watch videos, read things (a lot). I still come across ideas and think, that’s interesting, I’ll try that. I’ll happily test somebody else’s idea. And usually go right back to how I work.

Learn From AI Prompts Without Relying on Them

Use other people’s ideas. Don’t borrow their brain.

Look at prompt libraries. Try a Claude Skill. Have a play with Gems (Google makes a bunch of them to start with). Try out somebody else’s GPT to learn from it. Read the Facebook groups if you can survive the daily discovery of another “game-changing prompt”.

There may be useful ideas buried in there.

Take them, test them and bin the bits that don’t work.

Rewrite the bits that nearly work.

Ask why an instruction is there.

See what happens when you remove half of it.

Most importantly, understand what you’re asking the AI to do. If you don’t understand the prompt, you’re still risking creating low quality content.

If you don’t understand the work, you won’t necessarily recognise why the output is poor.

AI can produce poor output very confidently.

I’d rather have a handful of instructions I understand than 10,000 prompts somebody else collected.

Quite often, I’d rather have no fixed prompt at all. I’ll just start talking – thousands of words. Explain what I’m trying to achieve. Add the background. I’ll challenge the answer. Go through and correct things. Often change direction. Go around again.

Eventually, something useful appears but it’s based on my work, not a random prompt by some bloke on Facebook who’s more interested in selling something than solving my challenges or yours.

It’s not very exciting.

I couldn’t package my process into a PDF containing 2,000 life-changing prompts and there’s no £497 course hiding in it either.

But it works for me.

And surely that’s the whole point.

Questions About AI Prompts, Claude Skills and Gemini Gems

“`html “`