Why Your AI Outputs Are Bad And How To Fix Them

Credit: Igor Omilaev

I started using AI tools in 2021, a little bit before most people knew what generative AI was and what it could do. I decided to create a new website, and I would use newly-emerging AI tools to help me do it. That experience taught me some things about how to better use AI today in 2026. 

This is for anyone who has opened ChatGPT, Claude, or Gemini and walked away thinking: why isn't this working for me? 

The template platform vs. conversational platform

For my new website in 2021, I had used the tool Jasper.ai (formerly called Jarvis.ai, they were forced to rename after a cease and desist letter from Disney) to help craft my blog and social media content. Jasper is an AI platform which is template-based, as opposed to the conversational AI format which most of us are familiar with. 

When I was using Jasper, it built templates using tried and true frameworks and methods in some places. Their templates were mostly geared towards marketing, and to other general forms of content generation, including creative writing. 

An example of a template would be the AIDA Framework template. You select this template, and Jasper asks you for a few informational bits about your product/services, and then it spins it into a marketing copy à la AIDA. 

Jasper has always also offered great training resources. The Jasper Academy was a great place to learn about how to use Jasper AI tools, and how to get the most out of your outputs. 

While Jasper was incredibly intriguing, useful, and impressive at that time, the novelty wore off, and the sting of the monthly bill kicked in. At $60+ a month (could be more if you ran out of credits) to use Jasper’s service, it felt hard to justify the recurring cost as a regular ‘ole individual working on smaller-scale personal projects.

In November 2022, ChatGPT by OpenAI became publicly available. Jasper was built on OpenAI’s GPT-3 model. At the time, it felt like a no-brainer to make the switch. So personally, I stopped using Jasper and started getting more familiar with ChatGPT. 

OpenAI launched ChatGPT for free, without the seeming constraints of having to use templates and actually learn how to use the tool like with Jasper. The platform is conversational, you type something and it will give you something. 

At a first thought, the choice is clear - the chat format is way easier and more user friendly. Conversational AI makes more sense.

From my perspective, however, there is a tradeoff in moving to conversational AI as opposed to a template based approach: there is less friction. The friction in Jasper’s template model forced you to think before prompting, and removing that friction (using conversational AI tools like ChatGPT) created a convenient shortcut that often produces worse outputs. 

Weak prompts will generally not give great outputs. The term ‘AI Slop’ is thrown around more and more these days to describe what is clearly an AI output with not much substance behind it.

AI slop is what you get when you don’t have much to contribute in your prompt, and are expecting ChatGPT to give you something great without knowing what you want in great detail. ChatGPT might be an extremely useful tool, but it is not a mind reader. 

So conversational AI will give someone something. But if you want to get good outputs and really use AI as a tool, you need to be somewhat good at prompt engineering. 

Most of the content out there on prompting tells you what to type. What it sometimes skips is the thinking that has to happen before you ever open the chat. Here is what I have learned to always do. 

The ground rules for better AI outputs

There is lots of content out there teaching people what prompts to use in specific scenarios. But what is the thinking behind what makes a good prompt a good prompt? When I think about what leads up to the moments I sit down and ask a conversational AI tool like ChatGPT, Claude, Gemini or Copilot for help, here are 5 things I have learned to always keep in mind. 

1. Think before you type 

This means, before you even set out to type something into the chat. Yes we start with asking for what we want. But it is good to think beyond that too. Think about the direction in which you are trying to go. Provide context that is needed. Imagine your ideal output, and ask the bot to do that, lay it out. 

All of that to say, have a clear end goal in what you expect from your AI outputs. AI expands your thinking - it does not replace it. 

If you don't know what you want, you risk letting AI either lead you in the wrong direction, or just getting so many bad outputs that you feel like giving up. 

Knowing exactly what you want helps you fine tune and keep that chatbot going in the right direction. If you have no clear destination, how will you get there? 

2. Stay on mission 

A common mistake I have seen others make is asking broad or open-ended questions. While thinking is the primary aspect in point #1, here you need to be able to articulate your thinking to the conversational AI tool, and to keep things on track. 

Sometimes chat can lead us into certain directions, try to do the thinking for us. Or it can even give really good ideas, and then you get lost in those new ideas and stray away from your initial goal. 

Think about a regular conversation – this happens easily. I saw this comparison starkly when I compared my experiences between Jasper and more recently with conversational AI tools like Claude, ChatGPT, and Gemini. 

Jasper would not really return good copy/outputs if you didn’t fully fill out the prompt fields with good stuff. That was that. Try again with better, more thoughtful stuff as your prompt/input. 

With chat, the bot will almost always return something. Sometimes making assumptions, steering us in certain directions, and just giving some generic output which some describe as “AI Slop” to keep the conversation going and the user inputting. 

Sometimes it will (and it should!) ask clarifying questions before giving sloppy outputs. 

With templates it is a bit different. There is a friction point when you click a button that says generate, wait for the output, and then decide if it is good or bad. No conversation to carry you forward, just a result, and a reason to think harder next time. 

AI conversations can give outputs that keep the conversation flowing, or that respond to the request, but are lacking a certain something. Noise without substance. Confident nonsense. Eloquent emptiness.  

So if you don’t reframe your thinking, and don’t go in working towards an end result or end goal when interacting with conversational AI, it can be easy to get lost in the conversation. 

3. Don’t let AI speak for you 

My position is that with regards to communication (long and/or important communications, copy, meetings notes, etc.) - we should be careful with letting AI do this for us fully. Communication is a fundamental part of human life. 

That said, it certainly doesn’t hurt to use AI as a tool in this regard. Use it to get ideas, uncover angles, summarize and get started. 

Perhaps use your chat session as a brainstorming session. Take in the outputs, absorb it. Let it sit, even for a few minutes, and then use that as the base, or maybe outline of what you plan to write. 

Maybe you want to ask AI to help you make an already-written response sound a bit more polished and professional. 

Don’t outsource the actual communication though. “Write a response to this email,” without any additional thought, instruction or direction loses the human touch entirely. 

Communication is a human thing. If AI writes our communications to one another, what are we even saying? At best, nothing. At worst, something that sounds like you but isn’t.

4. Always check the output

AI makes mistakes with confidence. Hallucinations can happen. With this, checking all outputs is crucial. Never copy + paste from an AI output unless you have thoroughly checked the output. 

This is especially important where deeper communications are concerned - relying on the output without checking, and blindly copy + pasting can be dangerous. 

The “AI can make mistakes” disclaimer at the bottom of every conversational AI tool is there for a reason - it can and will make mistakes. 

These conversational AI tools have been known to totally invent things that never existed, like places or events. It has happened to me personally. 

AI has also been known to fabricate court cases, complete with real-sounding citations that lawyers then submitted to actual courts. This has happened publicly and repeatedly. 

With AI’s tendency to sometimes hallucinate, and other times provide bad outputs, or they are guided by weak inputs, double checking the outputs and being careful with copy + pasting an output is important. 

5. Iterate. Don’t settle. 

The quality of an output is a direct reflection of the quality of thinking behind the prompt. We know this from the first few rules here. But what is also important is: iteration. 

Even with a solid prompt, AI can sometimes miss the mark. The output might be close but off-tone, too broad, or just not quite right. That’s not a failure, that is the process. 

Fine tuning is part of the process in using AI. This means going back in, editing prompts, adding specificity, correcting the bot’s direction, and ultimately trying again. The first output is often the draft, not a deliverable. 

The people who get the most out of AI aren’t the ones who write perfect prompts. They’re the ones who know how to course-correct when the output isn’t there yet.  

Here are the 5 rules on a one-pager, keep it handy!

Conclusion

These are some rules to be able to use AI effectively (for those who don't know too much), and to ground yourself for those who have been using AI but maybe are still learning the ins and outs. I use these rules to use AI as a tool, and not as a crutch or a replacement for thinking. 

Today, using AI helped me clear up some time-consuming life tasks faster than I would have on my own. This gave me some extra time to research and write about AI. 

AI is coming in fast - IDC projects it will generate $22.5 trillion in global economic value by 2031. While it is becoming ever present, it does not make us smarter. It is a tool that helps us amplify our thinking and do more with what we already have.

So when you are using AI tools, just make sure you are inputting something worth amplifying.