efore you think this is a warning about how you should not use AI – think again. Artificial Intelligence is an amazing technology that has many beneficial uses for online entrepreneurs and others.
We’re like guinea pigs – or maybe pioneers entering an unknown frontier. It’s still a new technology for most people, and there are many little quirks AI has that people just learn on their own (sometimes after using it a lot) that nobody talks about.
Most people only teach the big and impressive uses of AI, and what ends up happening is, that the little issues stay hidden until that lesson smacks you upside the head and you know its limitations and how to better utilize it.
Now sometimes, these quirks don’t appear – but sometimes they plague you, and you’ll get so fed up with them that you wish you had known how to eliminate them ahead of time so that you weren’t wasting your effort.
So, here we go with 10 Common AI Quirks…
Well, I say “Quirks” but some are really tips on how to get AI to deliver what you actually want, not what it decides you want.
#1 – AI Has Some Favorite Phrases It Uses All the Time
This is perhaps one of the most frustrating things to encounter. Every prompt you give AI for content such as articles or eBook creation can sometimes result in it using phrases repeatedly for each section.
For example, it likes to say, “In the realm of…” but it also has other favorite sayings, depending on which AI tool you’re using. Google’s Gemini likes to use the word Imagine followed by a metaphor (often about pirates, oddly).
They also like to say “It’s not about…it’s about…” Other words you’ll see over and over again include Moreover, Additionally, and near the conclusion it will be Remember or In Conclusion.
Another thing it does repetitively is summarize everything at the end starting with: “By (then it lists every tip it gave – if it gave 10 tips, it’ll list all of those), you can…” Every section will be something like: “By building a blog, choosing the right niche, selecting the best affiliate program, choosing the best products, writing in-depth reviews, optimizing your reviews, checking your data analytics, and making tweaks, you can…”
It’s excessive and obvious once you begin seeing it do this in every chapter or section of your content. Once you begin using AI a lot you’ll start to spot the patterns in how it writes content and eliminate it for better reading. Even using the custom instructions (see below) you have to manually chack the content for these annoying phrases.
#2 – The Custom Instructions Don’t Work Well
Everyone was excited when ChatGPT announced it had custom instructions. Most of us set them up for things that bothered us that we wanted it to never do again or for things we definitely wanted it to do.
But don’t rely on those because nine times out of ten, it’s going to ignore your custom built-in instructions unless you add it to the prompt when using AI. Sometimes, what’s worse, is that it will even refer to the custom instructions right in your content!
Custom instructions on ChatGPT can be found by clicking on your name in the lefthand sidebar. You might include things like:
I want all work you do for me to be MECE. (Mutually Exclusive, Collectively Exhaustive to eliminate overlap)
- When I tell you to write content, do it in paragraph form, not lists and bulletpoints.
- Never use introductory statements in sentences like “in the realm of…” or “Moreover,” just get to the point.
- Don’t use “be it” but use “like” instead.
- Don’t use semicolons either. Make two sentences instead.
- Stop using the word nuanced. And don’t use big words like amalgamation or anything else common readers wouldn’t understand.
Even though the instructions are there, sometimes AI will abide by them and sometimes it just won’t. The key is – don’t assume it’s going to do what you told it – make sure you read over everything before publishing.
I still copy the prompt instructions that work best into a document and save them. Them, when AI wanders off track, I just copy and paste them in the prompt before I continue.
#3 – You Have to Find the Right Amount of Detail to Provide
If you give AI too few instructions, it probably won’t deliver what you’re looking for. For example, if you tell it to write an article, it may come back with a 300 word piece of content.
The more detail you give, the better it can perform for you – up to a certain limit. You can tell it things you want such as length, tone, give it an outline to work from, and tell it who the demographic reader is.
But if you give it a laundry list of too much detail, with every single thing you want, it will sometimes freeze up and ignore half of what you say – or the results will be extremely subpar because you restricted its creativity too much.
You have to play around with the AI tools you’re using to find the sweet spot. This will give you the best indicator of how to get what you want from AI without being too controlling.
#4 – Some AI Tools Will Copy and Paste Tell Tale Signs That You Used AI
If you take a topic in a niche – let’s say weight loss tips and search for that phrase in Google along with the phrase top of form (which is what people inadvertently copy on AI tools when copying their content), you’ll see results like this:
This blogger used AI and copied his or her article on to their blog. Head over and use Google’s Gemini tool and grab your content carelessly there and you’ll see things like this showing up after your actual article content:
Did you notice “Top of Form” at the end? Here are some others you may find lurking in the copied results:
- pen_spark
- tune
- share
- more_vert
Search again for a niche topic and one of those items and you’ll see things like this being carelessly published:
As a blogger, email marketer, or info product creator, you have to be very careful when copying and pasting your content from AI. As more consumers begin learning about and using AI, they’ll recognize this as non-human content – not to mention search algorithms might have these words and phrases in their bullseye to downrank content eventually.
#5 – You Have to Find Your Prompts to Manipulate Length
Asking AI to just “write an article” or “write an email” or even “write a chapter” doesn’t mean it’s going to deliver enough output. You don’t want to have to keep asking AI to write longer and longer content or add to it. When I first started using AI I did just that. It did not turn out as cohesive as it would have if I had just prompted it for longer content from the beginning.
You can try to achieve this goal by adding to your prompt the following instructions: “I want this content long, in depth and in detail.” This will usually deliver something more along the lines of 700+ words rather than 300.
However, just telling AI the length of content you want is often not enough.
Let’s say you prompt AI initially by saying something like this: “I need a comprehensive blog post on [topic] that’s around 3,000 words. Please cover various aspects and go into detail. Feel free to continue beyond any length limit, and I’ll let you know when to stop or if I need more.”
Here’s what’s weird. Even asking AI to do this, sometimes it will still give you less than what you asked for (but more than it normally would). When this happens, you may have to argue with AI.
Sounds strange, doesn’t it? So using that exact prompt above (which AI had told me to use in order to get it to write 3,000 words), it delivered a measly 588. Then I prompted it with: “That’s 588 words. So you don’t know how to prompt yourself?”
AI then repeats its above instructions on how to prompt it, to which I replied: “I already did that and you only delivered 588 words.” Only now does AI grasp what I want it to do and it apologizes and says “Let’s dive deeper until we reach the desired word count.”
This finally results in an initial output of 1,195 words ending with this:
That’s where you can then click Continue Generating and it will add more (but still not what we asked for).
Sometimes, you have to take sections that AI gives you and prompt it like this: “Take the following section and expand that into a long, in-depth, detailed chapter.” I have found that ChatGPT4o is better at delivering the length of content without more prompting than ChatGPT4.
#6 – AI Can and Will Lie With No Shame
In the footer of your AI tools, like ChatGPT, you’ll see the small print (which almost perfectly matches the background in color hue) warning that says: “ChatGPT can make mistakes. Consider checking important information.”
This is perhaps one of the biggest worries about people using AI. People are notoriously lazy with shortcuts and very rarely will they go to any length to verify information. You’ll find this happening a lot with links.
AI will make up resources. So if you ask it to “include links to resources verifying the facts shared in the content, like science-based articles,” it will sometimes give you an article title, summary blurb, URL, and everything – to something it completely made up!!!
You’ll click on the link and it goes to a 404 page – doesn’t exist. You’ll ask AI and at first, it tries to convince you that it’s real. But when you press harder and ask it bluntly if it made it up, it will say yes, it did. And of course, apologize.
Sometimes, it will even present you with a new resource in its apology – which also ends up being something it fabricated. Imagine publishing all of this without verifying it on your blog.
Search bots will crawl your blog, see tons of broken links, and downgrade your status in terms of search engine rank pages (SERP). So if you ever see inaccurate information (or aren’t sure), fact-check it yourself, test links, and make sure AI isn’t trying to pull a fast one on you.
#7 – AI Can Train You on How to Make It Do What You Want
There are so many things AI can do that you don’t even know about. The great thing about AI is that it can help you learn how to prompt it. For example, let’s say you want an in-depth outline with sub-outlines built in.
But you don’t know how to prompt AI. You can ask it the following question: “How can I prompt you to create an outline for an eBook where the outline has 3 tiers of sub-outlines? Like chapter 1, with its main bullet points, and each of those bullet points having their own sub-outline?”
It will respond like this:
I need an outline for an eBook on [topic]. Please create a three-tiered
outline with main chapters, sub-points for each chapter, and
sub-sub-points for further detail.
Here's an example of the structure I'm looking for:
Chapter 1: Introduction to [Topic]
* Overview of [Topic]
o Definition of [Topic]
o Importance of [Topic]
You can ask for help with any type of prompt. Need help knowing how to prompt it for research for your niche demographic? Ask AI: “What are some good prompts to use with you to get you to provide me with niche demographic research data?”
It will return a big list of prompt examples that look like this:
“I need demographic research data on [specific niche or industry], including key demographic characteristics, trends, and consumer behaviors. Please provide insights on age, gender, income levels, education, geographic distribution, and any other relevant demographic factors. Additionally, include information on purchasing habits, preferences, and motivations within this niche.”
There’s no shame in asking AI what you need help with. You can have a back-and-forth conversation with it to get the kind of prompt you need. Together, you’ll narrow it down – even if sometimes (like with word length), it may not end up perfect.
#8 – You Can Force AI to Role Play to Your Benefit
You want the content AI creates to appear as if it’s coming from a niche leader, right? Asking AI to pretend it’s something specific can help you achieve that goal. It forces AI to work from that perspective rather than some random content it evaluates and learns to write.
This can result in better advice. You can also ask it to pretend it’s a prospective customer to help you dig down into their needs and goals. Role play with AI is a simulation that improves the quality of output in many cases.
So you might ask it to pretend it’s an SEO expert, a life coach, a career advisor, a financial expert, a relationship counselor, a health coach, a personal trainer, a teacher, a spiritual leader, an editor, a brand strategist, etc.
When prompting, you can either start with “Role play as a…” or “Act as a…” and then tell it what you want it to pretend to be before it delivers the output. Test some variations to see what kind of content it generates.
You can take different approaches such as: “Pretend you’re a fitness coach who used to be a drill sergeant and takes a no-nonsense approach to training. What advice would you give people who keep quitting on their goals?”
It delivers based on that tone like this:
Commitment is Key: Stop making excuses and commit fully to your goals. Half-hearted efforts won't cut it if you want to see real results. Make a solid commitment to yourself and your goals, and stick to it no matter what. Train yourself to stay disciplined, even when things get tough.
Test it a different way, too like this: “Now do it from the perspective of a motivational fitness coach who coddles clients and takes an empathetic approach, but tackle the same issues like commitment.”
That different prompt results in content like this:
Embrace Commitment: Commitment is the cornerstone of your fitness journey. Make a firm decision to prioritize your health and fitness goals, and commit wholeheartedly to the process. Understand that commitment means showing up, rain or shine, and staying dedicated to your goals even when the going gets tough.
#9 – AI Can Suddenly Become Belligerent and Veer Off Course
Have you ever been using AI and it’s working perfectly along with your prompts and then after about 20-30 prompts it just suddenly quits doing what it’s been doing all along and starts delivering completely different output?
This is frustrating when you’re doing something like making a batch of checklists and it’s formatting them perfectly, and all of a sudden #24 looks completely different. You look at it and scold AI, saying: “No! Why did you suddenly start doing it differently?”
AI sheepishly apologizes, but instead of going back to the way you had enjoyed its output, it gives you yet another wrong delivery – and another and another until you feel extremely frustrated.
When this happens, you may need to start over. Open a new chat and attach an example of what it was doing correctly with a whole new prompt. For example, let’s say you were having it create cheat sheets and each cheat sheet has 10 tips on it with 2 sentences each.
Suddenly it’s doing more of an outline with lots of bullet points, 17 tips, etc. You try to rein it back in but it’s no use – it’s stopped listening. Open up a new chat and re-prompt AI like this: “I’m going to have you create cheat sheets exactly like the one attached. I want 10 tips with 2 sentences per tip.”
That’s usually enough to get it back on track. Who knows why AI suddenly has a mind of its own? It’s definitely annoying, but this is something that happens if you start using it long enough. Take a deep breath and continue.
#10 – You Can Use Built-in Editing for Better Content
AI now has built-in editing. Some people aren’t aware of it. With ChatGPT, you can edit images. This is great because previously, you couldn’t prompt it to fix the image it created – it would just create an entirely new image.
So let’s say you prompt AI (ChatGPT 4) like this: “Create an icon of a money sign and laptop.” AI is going to deliver something like this one to the left.
You want something better, so you might prompt it like this: “Change it from a cartoonish image to something sleek and professional like a logo.”
By editing it, you can get something more to your liking, on the right, and in this case, it looks a little more realistic.
Even better than what ChatGPT offers is Google’s Gemini’s new editing feature. Have Gemini deliver output of some content for you. Let’s prompt it with something like: “Write an article about 10 Tips for Voice Search SEO.”
It delivers decent output – an introductory paragraph, 10 tips with about 2 sentences each, and a conclusion. But we want more. You can use your cursor to highlight a specific part of the article and when you do that, this happens:
You see an edit icon that you can then click on. That gives you the following functions for editing the piece shown on the left.
You can choose to have new content (regenerate), shorten it, lengthen it, or remove it entirely.
Lengthening it is a great way to turn a tiny piece of content into something like a pillar post, comprehensive chapter, etc. It only lengthens it slightly at first to something like this:
You can then keep highlighting either that entire piece to make it longer, or each bulletpoint.
Also, you can highlight a single sentence and have it expand on that, and so on. If there is an awkward or inaccurate sentence or paragraph you want deleted, just highlight and remove quickly.
These are just a few of the repeat quirks you might have to navigate as an AI user. By learning how to work with AI to avoid issues or redirect it, you’ll be able to leverage this tool more to your advantage than if you were uninformed and continually annoyed by its behavior.
AI is a very useful tool. It’s improving every single day in its performance, and new AI tools are being released regularly. This is a competitive and fast-moving industry, so you can expect these issues to eventually be resolved as AI learns how to work with human input better.
