The Psychology of AI Content: Nuance, Bias, and Trust

I’ve been thinking a bit lately about AI/LLMs and business. There is a lot of chatter about how AI will take away jobs. There is also an emergence in business owners automating and using AI Agents to help them do repetitive tasks. But what do AI, LLMs, and AI Agents have to do with business psychology? Everything! This is the psychology of AI content: language shapes meaning, meaning shapes trust, and trust drives action.

Scientific American recently wrote on “AI and human intelligence are drastically different – here’s how” and led with this quote “engines of linguistic automation, not engines of understanding.” When we look at business, social media, and their psychologies understanding is key.

Before we get too far in, I want to clarify what I mean by LLMs, AI etc. LLMs, Large Language Models, are the programs and databases that sit behind the AI. AI, Artificial Intelligence, are the interfaces we use, Chat GPT/Nano Banana/Google Gemini. AI Agents are part programming and part user interface, and they can be both.

I’ve been seriously working with AI in my business since the end of 2024. Through 2025, I’ve moved more parts of my work into AI. Yes, to a degree even my blogging. I now use AI exclusively for my AI content. And I know I’m not alone in this.

There are a few things that spending this time working with AI has taught me.

1. Garbage in – Garbage out.

I am incredibly grateful that I learnt to program when I was 10, yes that was the 1980s. I then went on to learn how to build websites in 2000. While lawyers know that a comma can cost millions; programmers know that a comma can cost the whole job.

AI is currently reliant on LLMs, Large Language Models, and when we look at that term it’s clear. You have to be incredibly precise with your language. Why? You’re programming the AI/Agent to do a task.

You put a garbage prompt in, as in your not clear on your outcome or instructions, you can’t expect it to deliver anything but garbage. It just isn’t that smart. It’s not a mind reader, it’s a prompt reader.

2. Garbage can lead to disease

You give the AI garbage, then you can expect – somewhere down the track to get some confounding or hallucinating. Just as if it’s caught some sought of bug from the rubbish you’ve been feeding it; it’s going to run hot and start hallucinating.

Yes, I’ve had this happen to me. Yes, it’s incredibly frustrating. And yes, I had to remind myself that I fed it garbage and I was responsible for fixing it.

3. Buyer beware

It may seem that AI is the answer to our woes, and I’d be lying if I said that it hadn’t been helpful. It even warns us at the start that its answers may be inaccurate. So you actually need to know your stuff and be able to pick out the errors. You just can’t trust it. Even if you give it examples of what A+ work is, you can’t always trust it. (I’ll go into why shortly)

4. It’s a people pleaser

AI will absolutely stroke your ego and tell you that your work is good/great/awesome. It may even tell you that you are these things and more. And even if you were to give it strict instructions, its thinking and behaviour gets diseased, and it will start hallucinating.  And before you wonder what’s wrong with that, we need to be held accountable, we need to be told that we’re on the wrong path, we need to be told that we need to do better. AI is not going to do that.

So these are my experiences, but what have I noticed more and more in conversations I have with those who I respect in the area of AI & LLMs? This is where psychology comes in.

AI, Agents, LLMs and language

You might have realised, through your own experience or what I have said, that what we say or write has a lot to do with the efficacy of these tools. You’d be right.

These are Large LANGUAGE Models. At their core, they know and interpret language. They are only as good as their programming, including the prompt/question we ask. They can’t read our minds, and they certainly can’t read our body language.

LLMs and cultural bias

LLMs, in my experience, have a cultural bias toward the US. I have found that Google Gemini and the LLM it uses will recognise that I’m from Australia and reference Australian content or influence first. Chat GPT does not. I realised this when I was using Chat to help me write a job application, and I was referring to ‘Medicare card’. The Medicare system in Australia is universal health care and a different model to the US. Until I told Chat that I was in Australia and referring to the Australian system, it kept correcting ‘Medicare card’ to ‘Medicare’, which then changed the meaning of what I was writing.

Whilst it might seem that Gemini is devoid of bias, it’s not. It was biased until it realised my location.

LLMs and nuance

LLMs, while they have access to a lot of information, they do not always have access to the nuance of meaning in words that we do.

When I’ve been writing, Chat has removed words that I know are important and show subject-matter expertise. Whilst they may not seem to add to the text overall, there is a nuanced difference when including them.

Nuance is important; understanding the cultural background is important to understand nuance. AI, isn’t good at that.

Here’s a test I ran using a couple of AI models. Where I’m from, to “luck out” isn’t a good thing. It means that you ran out of luck, out of luck, lucked out. See the progression? I understand that elsewhere, to “luck out” is a great thing. By elsewhere, I mean even elsewhere in Australia. Just ask someone in Sydney for a fritz sandwich or someone in Adelaide for a devon sandwich. IYKYK. (Personally, devon is a poor substitute for a good butcher bung fritz)

So, as you can see in the tests, Gemini and Chat recognised the nuance, but both chose to ignore it and used the bias in their LLM.

So while we can upload all of the proforma, A+ graded work samples, avatars, and other documentation we would like the systems to reference, there are places where it is incapable of being our replacement.

LLMs and business psychology

Spend any regular amount of time on LinkedIn, and you will see someone complaining about AI-generated content. Fair enough too, when you realise the shortcomings. AI is not using the nuance it MIGHT detect or ignoring the inherent bias in its LLM. To be fair, the users aren’t checking either.

When we talk about social media, we are trying to do one of three things: educate, entertain, or inspire. But remember, the content needs to do this for our audience. While you could give the AI a list of things that educate, entertain or inspire your ideal client, or you could get your Agent to go off and research what content they are interacting with, in the end, you need to know the subtleties of the language and how it makes your ideal client respond.

If you’ve read content online, you’ll probably be able to see trends in writing. There’s a good chance that this is the bias of the LLMs shining through. The way they have read and “learnt” is the way things should be done. Again, they miss the nuance or subtleties of human language.

How your business can benefit from understanding psychology when using AI and LLMs

Please understand that language conveys meaning, and you need to understand the meaning that your ideal client gives to particular language, like my “lucked out” example. To help with this, I recommend my clients go through reviews and use the language clients have used when describing the service they have received. You can ask about what they had tried before, issues they have had beforehand, and things they were worried about. It’s a language, need, goal, mine of information that you can include in your psychographic ideal client avatar.

Why AI content sounds generic (and what that signals to your audience)

When we use AI generated content without considering, nuance, bias, or editing we run the risk of looking like all the other posts online (the main LinkedIn issue), we can look like we put in little to no effort which can make your ideal client fell like they are of little to no value, it can lack risk in its vanilla same-sameness which can signal that we were ticking a box in a business to-do list, and it can come across as impersonal and lacking vulnerability which just adds it to the pile of another generic piece of “writing”.
Hint: I’m seeing an increase in the everyday event being converted into a business lesson, and it’s screaming ticking a box.

How to make AI content sound human

Please give all of the content a close read and edit for bias. Do you want to refer to US Medicare or were you like me and talking about the Australian system? Remember that the LLM is only as good as the language in its database and its modelling, so there may be biases in there.
Hint: most AI systems have bias because of the information they’re trained on.
Reality Check: This works best when you are aware of your own bias and your customers’ biases.

Continuing on from bias, you need to have a really good idea of what you want to achieve before you start working with AI or Agents. Giving it the A+ examples helps, but it’s not foolproof. Be explicit in your prompting; you’re asking something to carry out a task, you need to tell it every step and the pitfalls it needs to avoid. This means you need to be skilled at procedure writing/recording.

Another way that you may need to use an understanding of psychology when using AI for your business is when it’s a people pleaser. This means that you will need to be able to be objective, step back, or even use “third person” skills to work out if the AI is telling you what you want to hear rather than what you need to hear.

You need to know when to stop working with AI. Many business owners, myself included, will run drafts through AI looking for gaps, errors, improvements and never really landing on a final version. Think of it as an AI-augmented analysis paralysis or perfectionism.

In the end, how much, and even if you choose to use AI, is a decision only you can make based on what is best for you and your business. Do not make using AI another “should” that you feel obliged to do. It’s a tool, and only you can decide if it’s worth the investment.

Final thoughts

When you outsource your writing to a tool that you know is unable to discern nuance and is biased, you risk your positioning. Outsourcing your positioning like this can weaken trust. This is the point where your audience is having that LinkedIn moment, doubting what you’ve written and your authenticity. And we know that people buy from people that they “know, like, and trust”.

I understand that having a tool to help with social media, in particular, reduces a lot of the mental load. Having something to write removes writer’s block, worrying that you’ll say the wrong thing, not being confident in your writing, and making sure that you hit all the parts of a post that convert. What relying on it does is add to the story that you’re not good at these things, and this interferes with your ability to critically assess what it has written and how it will connect and build trust with your audience.

And here’s one thing my AI suggested that I should add to this blog…

Practical “anti-LLM voice” rules

Here’s a simple checklist for you to use when editing AI-created content or even add them as rules in your prompts:

1. Add one strong opinion (even a small one).
2. Add one hyper-specific detail (place, timeframe, example, “IYKYK” moment).
3. Replace 3 generic phrases with client language (from reviews, DMs, consult notes).
4. Add one “cost of getting this wrong” line (what happens if they keep doing it).
5. Keep one slightly imperfect sentence (because humans talk like humans).

References

Quattrociocchi, W. (2026) ‘AI and human intelligence are drastically different: here’s how’, Scientific American, 18 February. Available at: https://www.scientificamerican.com/article/ai-and-human-intelligence-are-drastically-different-heres-how/ (Accessed: 26 February 2026).