Making AI Agents More Humanlike

With the method that I showed you, you can put together pretty easily huge lists um without much effort. Like this, this is going to take another like 5 10 minutes, right? In order to find these things cuz I asked for huge examples, but the huge amount of examples and the report that you get out of it is big, right? The analysis of what is and isn't human writing and then a list of just straight up what are words that we shouldn't use prevents most the problem. Hi, my name is Demetri Bonichi and I'm a content creator, agency owner, and AI enthusiast. You're listening to the AI Agents podcast brought to you by Jot Form and featuring our very own CEO and founder, Idkin Tank. This is the show where artificial intelligence meets innovation, productivity, and the tools shaping the future of work. Enjoy the show.

AI doesn't sound like people. They don't sound like people because they're not people. They're AI. They're like robots. Heard of them? and their writing style just simply isn't that natural. We've all seen the downfall of the world we live in when somebody's on LinkedIn and there's a clear and obvious double dash that you can't even write on a keyboard. So, you know that was AI written. Well, what if I told you you can actually make your AI never use the double dash and never be nonhuman sounding again. It's actually not the hardest process ever and I've gone through it pretty extensively utilizing a tool that I like to call person Pete. So person Pete is a agent that I made inside of text cortex and it's used via two different parts which are both equally as important. If you go to person Pete in here

which is this agent I've created. I've given it the background to essentially be the most human sounding, the best, the greatest, the least weird AI word usage, AI sounding robot ever. I gave him a huge list of always and never to do rules. And I gave him more importantly a knowledge base that founded everything on this. So this knowledge base is called anti-AI human writing. Okay? And it's actually pretty basic. I looked up using Gemini deep research which you can do right now. What are the grammatical patterns in AI versus human sounding marketing writing? What are the analysis of AI writings for Cheshi PT Claude and Gemini? And then what are human sounding terminologies? And then on top of that, I made a blacklist of each and every term and phrase that is used like gamecher, clockwasher, close the loop, compelling, competitive advantage, all these

silly I wasn't going to say stupid. All these silly terms that don't sound human and don't sound personal. Who says insights and they're not trying to be a corporate shill? Who says actionable and doesn't mean in the context of an AI written piece of content or some weird meeting that you're having with somebody that's for business, consumer driven, competitive advantage, core competencies, corporate culture, cutting edge. Yeah. Yeah. Guess what? AI uses these words because whoever trained all these models thought we were all in a corporate meeting. But guess what? We're not. When we're trying to use this stuff, it's usually for content or chatting with someone via email. And there is no reason to use the term laser focused unless you're being overly corporate. So what I did was I literally went to Gemini, so simple, and I did deep research and I asked a

myriad of different questions. Please find for me a deep dive of the AI stylistic choices in writing versus human writing. And then I cleaned up the results of that and I asked a similar question. And by the way, my best hack for doing deep research for agents is asking follow-up questions up to three times with Gemini because on my pro plan, it's unlimited, but it's three at a time. So, I can look up three deep research questions. So, I could say like, "Give me 100 examples of overly corporate and non-natural human sounding jargon words, phrases. give me the word slphrase followed by an example of it being used in a sentence. So I do this I do this I do like three at a time, right? So I can ask another question that's um more focused on how AI generates it. So I can do

another one. Give me 100 examples of overly corporate non dragon. Yep. Overly corny. Yeah, I I like corny because what ends up happening is uh with corny that happens a lot with AI. It's just like weirdly corny for no reason. Okay, so deep research. deep research and then I can do it a third time. And when I have all of these then put together, I can make an agent that actually sounds like a person. You'll see this example pretty easily. And by the way, all I needed to do to add these were this essentially spits them out into Google Docs. I put it into Google Docs. I download them. I upload them as PDFs here. Now when I write something whether it's with a different thing like a scripting agent or a uh AI hermosi which is essentially a business type guy or bisdev Brian

which is to essentially attempts to figure out how to move a conversation forward. I then always do a follow-up prompt with person Pete which I have called basically improve writing. Yeah AI to human writing word choice and then you can do length and formatting too but the prompt goes like this. The goal is to create a text analyzer that removes or replaces AI identifiable blacklisted words. So since I have those blacklist databases, a return format, which should just be the improved text with no changes uh or with no commentary before after and then warnings maintain the original meaning while replacing the language and then it essentially has a variable here to add in the context. So let's test this out by using like no agent and telling me to write an email. So even with you know even with Claude, which is a good writer,

it'd be bad. So, let's try to be honest here and give Claude a fair shake. So, I'm going to say, "Art me an email to ask somebody what is the status of the video that is due to edit in 2 days. Be pushy to try and see what the status is without being mean." Right? So, this is essentially just going to write it out. Okay. Hi, name. Hope you're doing well. I want to touch base. I know you've been working hard on it. I'm eager eager. Who the heck says eager? Nobody. Could you give me a quick update on where things stand? I'd love to know what percentage you'd say is complete. Are there any roadblocks I can help improve? Do you feel confident? I'm asking now because I want to make sure we have buffer. Yeah, this is pretty well written. It would be

much worse written with something like chat GPT. I'm going to open up another tab and just show you how it would be slightly worse with like GPT4, even 4.1. Um, but we can do 40 just to prove the point. And then what we're going to do is in this other tab, I'm going to go here to person Pete. I'm going to go to AI to human writing word choice. I'm going to copy this, put it in the prompt, press enter, and it's even the best with that claw for in my opinion, which sounds really humanlike nowadays. And you'll see it's searching AI identifiable blacklisted terms. It's going through the knowledge base and fixing it. So, hi name, hope you're doing well. I wanted to touch base about video edits that's due in a couple days. And now we can see in the 40 example, I

hope you're doing well. Wanted to touch base regarding you due for the next couple days as the deadline approaches. I'm keen. Oh gosh. Who says I'm keen to understand where things currently stand and ensure everything is moving along smoothly? Yeah, how about no? I I do not like that writing whatsoever. It's potential issues promptly. Talk about a mouthful. So, what we're going to do is we're going to copy this and I'm just going to start off with a new page cuz honestly that'll make it easier. So, we'll use person Pete. We'll use this prompt. We'll paste this in here. Keen to understand. Could you let me know? Oh my goodness. It would be really helpful to have a brief update. Brief update. Who says brief? Hi, hope you're doing well instead of I hope you're doing well. That's much more human actually. Wanted to check

in about the video that's due for editing in the next couple of days. With the deadline coming up, I'm curious where things stand and want to make sure everything's on track. Could you fill me in on how the project's going? A quick update on the progress or any roadblocks you're hitting would be great. Would be great is very human. This will help us stay on the same page and tackle any issues that pop up. Very human. And thanks for all the hard work on this. I know you'll nail the final product. So, uh, just you can see the before and after is night and day. Thank you. you in advance. Looking forward to your update. One of those sounds like you're talking to a coworker. The other one sounds like you're attempting to be corporate on purpose or you just sound AI. So, this can

be used not only in like a chat GPT sort of opening up the product in text core text format, but it also can be used for like automations. You know, if you're sending out mass emails, I use this for uh a CRM workflow or essentially I'm following up with a client uh or a potential client about uh working together. I'm following up with uh cold email prospects. all of these different situations. I know that whatever the AI will output, if it is accurate in regards to what is inside it, this will at a minimum make it sound like I wrote it and not an AI wrote it. And that's the key thing here. We're moving more and more and more towards the world of people know you use AI. You need to automate that out of it. You need to AI the AI out of

the AI. And lucky for us, text core text is something I use often and other people can use. And it's not the only product, but with the method that I showed you, you can put together pretty easily huge minutes, right, in order to find these things cuz I asked for huge examples. But the huge amount of examples and the The blacklist in my opinion are so great because it just prevents the writing from ever using these bad word I don't want to call them bad words but and these silly words that make no aha moment nobody says that out loud. At the end of the day is the the most annoying colloquialism in history. Baked in could be used I guess but isn't that great? Big bang for the buck. Oh god that's corny. Best of breed. Nobody would say that out loud, but AI

would say it. AI would call it a game changanger. And here's another example. Circle back. Ooh, heck no. Take this offline. Heck no. Put a pin in it. Heck no. Let's table this parket. Heck no. Let's punt this. Back burner. Out of pocket. Run it up the flag pole. Rain check. All these corporate jargon phrases are awful. Unpack. Scalable. Core competency. Action items. Idulate. Operationalize. Value ad. Alignment. Ooh, that's bad. I can't stand the word alignment. So, as you can see, you can keep iterating on this. And what I like to do is anytime I see the AI write a phrase or a word that I don't like, I add it to that blacklist. And then you can iterate on it, continue to improve, and the next thing you know, like me, you have an agent that prevents the problems. So I would recommend that

you try out this process with whatever AI tool you're using, add these PDFs, add a knowledge base, make an AI prompt that turns an agent into a human sounding agent. Its entire job should just be to sound like a person. And that'll get you more advanced in your writing, whether it be for social media, newsletters, general client follow-ups than anything else. And when you're using a tool like text core text, it feels as easy as open up chat GPT. And it also has an API capability, too. So, if you're interested in this, try out text and let me know what your thoughts are in the comment section down below. With that being said, thank you so much for watching this episode. I know it's going to be helpful for those of you that take advantage of the advice. It helps me every single day. We'll

see you in the next one. Bye.