You can always tell when something's been written by AI with.
Little human involvement. You know what I mean?
A LinkedIn post that starts with in today's fast paced world, or an article that ends with it's not about X, it's about why. A paragraph but technically fine, but somehow feels like it was written by nobody. And that is the uncanny valley of AI writing. And the problem isn't just that it sounds robotic, it's that people stop reading.
The frustrating part is that this happens even when you have done the thinking, You write the ideas, hand it to AI for a cleanup, and it AI effies the whole thing, flattening your voice and coding it with M dashes until nothing sounds like you anymore. So today Neo and I get into exactly how to stop that from happening. We cover specific instructions you can give to AI to keep your writings sounding human, things like mixing sentence lengths, avoiding stiff transitions, and banning the.
Rule of three. To see what I did there, I just choose it. We also go through.
Some of the most overused AI words and phrases to strike from your prompts entirely and now as a little bonus, you will find a link in the show notes to our anti AI slot prompt, which is a ready to use resource you can customize with your own personal pet hates and run any content through before it goes out. Okay, let's get on with the show. Welcome to how IAI with me, Doctor Amantha Imber and Neo Applin, head of Inventium AI. Each episode we share one practical way to use AI better at.
Work and in life.
No fluff, no tech jargon, just things you can use straight away. So much content in the world is clearly written by AI now, and I know that it.
Really frustrates us, But like, why is it a problem?
Why should we actually be aiming to not sound like AI when we are all using AI to create the content and words we are putting out into the world.
Now, Yeah, I think it's really two parts. One is don't be AI, in other words, don't just get AI to do all your work for you. And then the second part is don't sound like AI. And I think because a lot of people are now doing things where they're going like five dot points and say write me a big article, and it might be there are five pages that really Ai wrote all of it apart from five dot points worth, and it just feels and smells like AI, and it just has all the tales like
everyone knows about the M dashes. But you can just read it and go, ah, this doesn't sound It sounds like CHGBT. And so you can get AI to do all the work and it sounds like AI. The other thing is you can do most of the work and then get AI to do a cleanup at the end, and it will sound like AI. And so everyone's going to look at that and go, the whole thing's written by AI, and then they'll disregard it like they probably should about the first one, which is just all written
by AI. So if you build some stuff that sounds like AI, people are going to roll their eyes and they may not read it with the same kind of intent or kind of care that they would if it was sounding like you. And they go, oh my god, I Mattha right the whole thing, And I really want to listen to everything she was said, and I really want to read it intensely rather than I don't know if AIS written this. Therefore I'll just skim it, or I'll get AI to summarize it.
So the point of this episode is not to say don't use AI to help you write things, but perhaps don't use AI to do all the thinking and all the content creation. And so then the point of us helping you not sound like AI means that the work that you do produce, that you do genuinely think about and work with AI to refine and edit and so forth, gets taken a whole lot more seriously rather than people
just dismissing it. So Neo, I know that in the Inventium AI training, when we're teaching people how to use AI for writing, you actually do go into this. So what are some of the things that you're teaching people to avoid sounding like AI?
The first one I will say is this, Like often we'll get it like a paragraph and you might say, look, I need you to clean up this paragraph AI, so you're not getting it to write the whole damn thing. And the first tip is telling AI don't make this sound more complicated than the original. So IT often likes to have these big words and be very stiff and quite formal unless you've previously told it you know what you like it to sound like, and things like that.
So the quickest easiest one is just saying something along the lines of, don't make the output more complicated than in the regional when you give it a paragraph and ask it to critique it and clean it up and things like that. But there are other ones that you can add in. So the first one is the most obvious one, but there's a bit of a catch on how to do it. You know those M dashes that seem to pop up everywhere, So I'll just explain for those who aren't sure what an M dash is, and
M dash is a longer dash. This is a I probably should get you to explain this, you're the actual author here, but yeah, it's a proper way to write things, right.
Yes, yeah, So they're like visually they look like a double hyphen So if you can imagine two hyphens connected,
that's what an m dash looks like. And because AI was trained on books, if you read any book and they're using you know, like language like and I'm just trying to think of a sentence that has like an M dash in it, so you know it might be run every piece of content through this list before publishing, and if you spot three or more, rewrite, so like you could theoretically have an M dash before the end, just to kind of break up the sentence.
You'll see it in any book that you're reading.
And because AI was trained on all those books, a lot of M dashes went into the system in terms of its knowledge, and it's.
Really hard baked in there because AI is a product of its training, because it was trained on so many books, then M dashes are everywhere throughout its training, and it just pulls out M dashes an awful lot, and it's really hard to get AI to not use M dashes. But so rather than just say don't use M dashes, I found that that has a very low chance of working. Sometimes it does gish kind of Sometimes. The phrase that I use, which has given me the best success is this.
Now it's don't use M dashes, So that is, don't use em space dashes. And then I actually give it an example of an M dash, So I actually paste in an M dash within brackets, so it then knows that's an M dash. I've been really clear about this, and you could if you don't like short dashes either, you could say or short dashes, and then you can
put the short dash in an inner brackets. Rely on commas full stops or short shorter sentences instead, So you could do things like that, or you could write it if you're affine with the dash, you just want them to be short. It'd be don't use M dashes, paste in an M dash instead. I want you to use
short dashes and then paste in the short dash. So it's up to you whether you want to ban the dash or you just want to keep the short dash, but yet give it first off, the words and also the example of what the dash is you're talking about, and I found that is better, but not one hundred percent why it loves the M dash and it's really hard to get it to stop doing.
What are some of the other things that you teach in inventium dot AI programs.
These are all to undo some of the very common things that AI does, and with AI doing in the
commonly things, it's doing it the right way. And I'm doing your quotes here because the right way to write is the very formal way to write, and it's been trained on the right way to write by a whole bunch of books and sources, and also you know the guides on how to write well, and so therefore you know if you're a native English speaker and you're you know, you've done a lot of writing, chances are you more get caught of these AI detectors and things like that.
So it's almost like putting the human back into it, making it a little bit messier than the perfect way to write. Okay, So with that first one's the second one is mix sentence lengths. And the reason for this is AI generally has the same sentence length again and again and again, and so when you and I read this, we may not actually pick up that these sentence lengths are the same, but we just go this doesn't feel like a human so we get that kind of like
the uncanny valley of writing. So say mix sentence lengths, these short ones for punch. He's longer ones when you need detail, And I found that one does pretty well. So if you remember hending Away famous for short, punchy sentences, so get some punchy things in there, get some detailed ones that mixes it up and you and I reading it go okay. Another one is the language that AI uses. It can try to be the smartest AI in the room and can like to use those long words as
part of that don't make the thing more complicated. So you could do something like this, use playing language at a ninth at about a ninth grade and reading level, avoid jargon and buzzwords. Now that the ninth grade thing or the fourteen year old type brief is common along in industry practice, I understand when the newspaper or pro writing and things like that, that's the kind of goal to get to. Why because everyone can just read that pretty simply,
pretty easily. It's not really difficult to read, and you're not going to pull out university style words that are only a handful of people going to understand. So that's why you do that. The other ones are sometimes it deals with the way that sentences is constructed. Particularly start
of sentences can be a bit stiff. So one of the things is it's fine to start a sentence with and but also although you should do this sparingly, so occasionally we will actually start a sentence with a handbuilt but or so that's not the right way to do it. I'm doing air quotes again, but humans do that, and so therefore let's let AI do it. Not do it all the time, but we'll let it do it. The other one about starting sentences is avoid stiff transitions such
as additionally and moreover. Who's start human says moreover blah blah blah blah blah. You know, you feel like you're like a stuffy person in the eighteen hundreds with a pipe or something. So I avoid stiff transitions such as additionally and moreover. Now this is a weird one, but I find this one. I used to like, but now I hate that AI does. Often it'll frame things in a particular way near the end of a sentence, end of a paragraph. It'll say half bunch of explanation stuff,
and I want to say, it's not this. It's so it maybe like, it's not just a good way of writing, it's the way to communicate something.
Like that, right, AI.
It does this a lot, and actually was at the start when it was coming, I was, Oh, that's really clever. It does that really well. I like, but it over uses that mechanism, and so therefore I actually had to google to find out what that mechanism was called. It's called corrective antithesis. So I say, never use corrective antithesis, And then in brackets, given an example, I say, for example, not X, but why and then that certainly cuts that down. And there's one more if I can throw it in there,
which is don't lean on the rule of three. Now, the rule of three is generally you give examples. You can give three examples, and it kind of makes you know, it reads nicely. But AI over uses the rule of three. It over uses that in so many things, and so I say, don't lean on the rule of three. Use one or two items, or use a longer list if it strengthens the point. So you can do one or two,
or you can do five. You know, up to you, but don't always use rule of three, because it's that consistency, like the sentence lengths being the same, the rule of three being the same. It's not this, it's that being the same that feels very robotic. Because robots are really good at doing things the same ais really good things the same. Humans we're messier, and so we're used to seeing things that a little bit messier. And it's also easier and nicer for us to read because it kind
of feels more like us. And so this is just a whole bunch of points you can put in if like you've got this as a standard prompt, all those what have I got one to three, four, over, six, seven, eight, seven or eight dot points? You can just drop them in and then say, when you rewrite my paragraph because I needed to be cleaned up, don't do these things. That's what you're trying to tell them.
Okay, I love all those.
Now what I did ne in preparation for this episode is I went into all the llms that I use and like I love comparing the output, so I went into chatpt Claude, Gemini and also Perplexity, and I gave it all the same prompt, which was something along the lines of.
Tell me what are the fifty.
Most overused or over indexed words and phrases that AI likes to use, and I think I suggested to look at LinkedIn for some you know, some examples. And then what I did with the output that I got from each of those four models, I then put them all into the one model Claudes generally the one that I default too, and I said, give me a summary of the ones that are appearing most frequently across these four outputs.
And so with that, I've then created a bit of a mega prompt that people can use that I've called the anti ai slop prompt and if you want that, I will link to it in the show notes. But I'm just going to go through a few things just for some additional advice for people that are trying to un ai aify their writing so often. And these are grouped into some categories. So there's the first category is
empty opener phrases. So you will have seen things like this in today's fast paced world, in an ever changing landscape when it comes to.
Blah, picture this black like.
These are all sounding really familiar because these are how a lot of certainly a lot of LinkedIn post starts start. Now, then we've got a category of fake profound hooks, which I think, you know, it's not X, it's why is certainly one of those. Here's the truth, which is one of my pet hates, or sometimes he's the uncomfortable truth, he's the inconvenient truth. I hate that one because I use that language here the truth or here's the thing.
I actually use that a lot in my writing. And what is really frustrating, as someone who was writing books before the age of AI is that I feel like I now have to change my writing style, which was you know, certainly in some ways similar to what AI produces to you know, just added like I used m dashes very often, but now I will lean more towards brackets and semi collins for like little asides or quips as my editor calls them, which is very frustrating when
you have to change your own writing style so it doesn't look like AI. And we've got lots of corporate buzzwords. Probably the most common is delve. I was hearing some research actually that was done in twenty twenty three where academic authors that were most prolific in twenty twenty three in terms of getting things published over indexed heavily on the word delve. I don't know if you've heard that research. I heard Ethan like sharing that recently, which made me laugh.
I didn't hear that researcher. I'm a bit sad about delve because I use delve as well, but AI uses it more. And so you may add a bunch of words. Can we also talk about other words we may want to add to that banned list?
For sure? For sure? Give me like three.
More quietly is another one that I was going.
To say quietly.
So the next category actually is what AI has termed hollow adjectives and nouns and adjectives, and if you don't remember from your English class, and adjective is a descriptive word for a noun, So it describes a noun which is like an object or a person, a person, place or thing. I think is how my English teacher probably
would have described it. Quietly is so overused. And again that really bothers me because that is a word that I would have used, but now I feel like I have to strip it out mostly from my writing.
Okay, what are the words near this?
Ultimately, I would recommend you build your own list of words that you don't want AI to use, because there's some there that AI will have used, like Delvin quietly, but there may be others that IT would use that I find words but you would never use. So I
use the example of like buzzword bingo, so like synergistic. Right, if AI uses the words synergistic and you would never use it, then put a list together and add to that list, but sudogistic in there, maybe put delve in there, maybe put quietly in there, and go, don't ever use these words, And so build up that list and enhance that over time as AI uses words that just great on you, because if you've been grading on you, they don't sound like you.
So hopefully.
I mean, this is so much to take in, as I said, for those that are listening and like, yeah, I really, I really need to improve my writing output because I could relate to a lot of those things, and I think about my writing. Yes, I'm using a lot of the things that Neo and a manter I have talked about. Go to the show notes. There is a link to get our anti AI slopped prompt, which I highly recommend you customize.
So it's a lengthy prompt.
It includes a lot of what we've spoken about today, but do read it and put in your own personal pet peeves so that you can customize it and run any content that you create or create with AI through
this prompt. It will improve your work a lot. And also I would say, please share this episode with maybe some people that you know that are perhaps producing a bit of AI slot, but they are actually doing the thinking themselves, which is the worst combination where the human is doing the majority of the work and AI is then going over it and AI are fying it and completely destroying that person's credibility, so do share this episode with those that need it. How i AI was hosted
by me Amantha Imber and Neo Applan. A big thank you to Martin Imba who does our sound editing and Jem Rubio for production support, and thank you to John Kilby who composed the theme music