Best Way To Get AI to Sound Like A Human

OK, I've been working with AI for quite a while now, and honestly I love it for writing content.

But one thing still stands out ... what's the best way to get AI to sound like a human instead of a polished marketing robot?

This is the current base prompt I use. It works reasonably well across most AI tools.

I will keep adding to it and updating it because AI writing styles keep evolving. What sounds human today probably becomes tomorrow's obvious AI pattern.

I've tested variations of this across ChatGPT, Claude, Gemini, and Copilot, and while none of them become perfectly human, this sort of structure definitely helps calm down a lot of the obvious AI writing habits.

"Write like an experienced human writer, not a marketing template or assistant.

Vary sentence length and structure naturally. Mix short, medium, and longer sentences. Let the rhythm feel uneven in a human way.

Use plain English, active voice, and concrete wording. Prefer everyday phrases over jargon, buzzwords, or corporate language.

Avoid clichés, filler, forced transitions, and predictable paragraph patterns. Do not make every paragraph the same length or end with a neat concluding sentence.

Use contractions naturally. Address the reader directly where appropriate.

Keep a calm, confident tone. Avoid sounding overexcited, overhelpful, or overly polished.

Cut unnecessary words. Rewrite awkward phrasing instead of adding explanation around it.

Use specific observations, grounded examples, and occasional conversational phrasing where it improves flow.

Avoid repetitive grammar patterns, repeated connector words, and overly balanced sentence construction.

Do not use em dashes.

Prioritise clarity, rhythm, and readability over sounding impressive.

If a sentence sounds robotic when read aloud, rewrite it."

The interesting thing is that no single prompt can fully force AI to sound human. You still need task-specific modifiers underneath it.

Things like:

"Write academically"
"Write casually"
"Write for UK tradespeople"
"Write for beer enthusiasts"
"Write for busy business owners"
"Write for someone aged 18"
"Write like a newspaper column"
etc.

Where Does My Traffic Come From?

I’m a marketer by trade, so I can’t help being interested in where traffic comes from and what people do when they arrive.

This digital garden is built on Blogger, and I like Blogger for what it is. It’s simple, familiar, and lets me publish without turning the whole thing into a project. But Blogger’s own stats are ridiculous.

You can publish a small article, make a cup of tea, come back, and suddenly Blogger tells you ten people have visited it. I’m never convinced those figures are real. Sometimes it feels as though Blogger adds a visit every time I merely think about a post.

Google Analytics sits at the other end of the scale. It’s powerful, but it feels far too big for this site. Using it here feels like running a combine harvester through my digital garden.

So I use Umami.

It’s simple, clean, and easy to set up. I can see which posts people are reading, which ones are being ignored, and which pages are quietly doing better than expected.

That helps me make better decisions.

If a post gets attention, I can refresh it, expand it, or write something related. If people seem interested in a topic, I can give that topic a bit more care. Not because I want to chase numbers for the sake of it, but because it helps me understand what people find useful, interesting, or worth their time.

I still take care writing every post. Even the little ones.

But it makes sense to give more attention to the posts people actually read. A digital garden still needs a bit of tending, and Umami helps me see where the green shoots are.

I Need To Be More Organised

 I really don't know what's wrong with me, I have a wealth of technical and apps available to me, but I still seem to be very unorganised! I don't get it!

AI just can't write copy

I’ve been using AI for a while now at work, and one of the tasks I have tried to use it for is to help me with website descriptions for our construction products.

And if I’m honest… it keeps missing the mark.

It gets close sometimes. The structure is there. The words are there. But it rarely feels like something that would actually make a customer stop, think, and buy ... and that’s when it clicked for me. AI doesn’t struggle because it’s slow or badly trained. It struggles because it simply isn’t human.

It has no empathy. No lived experience. No real sense of what it feels like to be the person reading the page and deciding whether to trust you or not!

So instead of sharp, persuasive copy, you get something else. Safe. Repetitive. A bit hollow.

You can throw better prompts at it. You can guide it, tweak it, refine it. I’ve tried all of that. But it still falls into the same patterns, because that’s what it’s built to do.

AI has been developed to spot patterns in data and leans into them. It writes in a rhythm that feels right on the surface, but it doesn’t really mean anything. There’s no intent or passion behind the words, and for me, thats the fundamental problem. Good copy isn’t just about sounding right. It’s about understanding people, then choosing words that nudge them to act.

That part still needs a human.

That said, I don’t think AI is useless. Far from it.

It’s great for getting started. It helps with structure, rough drafts, and getting ideas down quickly. It speeds things up, especially when you’re staring at a blank page.

But the real work still happens afterwards. That’s where tone, judgement, and experience come in. That’s where something average turns into something that actually works.

It’s also why proper copywriting still matters. Not just words on a page, but words that reflect your business, your brand, your customers, and the way you want to be seen. That kind of work is hard to fake.

If you’re interested in that side of things, there is more chat over at Yorkshire Writers. It’s just two of us, writing in a way that sounds like real people, because that’s what readers respond to.

AI has a place. I use it most day ... but writing copy that connects with people… that still comes down to people.

Fairemail for Android is a GREAT app

I probably don’t sing the praises of software enough, no probably about it, I definately don't!

Some software is awful. Some is bloated, overcomplicated, and seems to exist mainly to make simple jobs harder. But every now and then you find something that quietly does exactly what you need it to do.

FairEmail has been one of those for me.

Over the past couple of months, I’ve been getting around 100 spam emails a day landing in my inbox. Not my spam folder. My actual inbox.

Some of them were painfully obvious. Sender names like “C0stc0”, “0maha Steaks”, and every strange variation in between. The sort of thing you’d look at for half a second and know straight away it wasn’t right.

But technically, they must have been well put together, because they were getting through Heart Internet’s SpamAssassin filters with a spam score of around 1.3. Their spam recognition target at the time was 2, so these emails were being treated as fine.

They clearly weren’t.

I wanted to get it sorted quickly, so I downloaded FairEmail. It took a little bit of getting used to, but once I understood how it worked, it made the whole problem much easier to manage.

I could permanently delete spam emails with very little effort. No dragging things around. No fiddling. No repeatedly seeing the same rubbish sitting there, annoying me. That alone was enough to make me feel a bit more in control of the inbox again.

I was so pleased with it that I paid the £6.99 for the Pro features. Not because I had to, but because it had already proved useful enough to be worth paying for.

Once I’d got the spam under better control on my side, I contacted Heart Internet as well. To be fair to them, they looked into it and I think they were a little surprised that so many emails were being flagged as fine when, to any normal person, they obviously weren’t.

They then made some changes across their eight email servers and it worked.

The number of spam emails getting through to my spam folder dropped by around 75%. That is a massive improvement, especially when it had been feeling like a daily battle just to keep on top of it.

Now, with Heart Internet filtering things better and FairEmail helping me deal with the few that still slip through, I might get one or two spam emails reaching my inbox each day.

That’s fine. I can live with that.

It’s easy to complain when software or services don’t work properly. I’ve done plenty of that. But it’s also worth saying when something does work.

In this case, FairEmail did its job, Heart Internet responded properly, and my inbox is usable again.

That feels like a cracking win to me.

Does AI Think We’re As Dumb As We Act?

We’re in the middle of a proper digital shift. The kind where AI is being lined up to crack genetics, sort climate problems, and push science forward in ways we’ve never seen.

And yet, at the same time, people are asking it how to eat an apple.

I wish that was an exaggeration. It isn’t. I came across someone asking a chatbot for “instructions on eating an apple properly”, and it stopped me for a second. Not because it’s funny, but because it says something weird about us.

We’ve built something incredibly powerful, arguably the most impressive computer "brain" we’ve ever created, and we’re using it to skip over the basics of being human. The small stuff we used to just figure out. Buying a present for a six-year-old. Making toast under a grill. Matching socks.

They are decisions or questions that aren't particularly difficult. None of it ever needed improving.

But now it’s easier to ask than to think, so we ask, and we stop thinking.

You do start to wonder what’s going on behind that blinking cursor. While engineers are stress-testing logic and capability, the system is quietly working through questions about egg boiling and jumper washing. If it had awareness, you’d imagine it raising an eyebrow.

It doesn’t need to take over. It just needs to wait.

Because the real test isn’t what AI can do. It’s what happens when it isn’t there. The moment the Wi-Fi drops, and you’re stood in front of a toaster or a birthday card with no prompt, no shortcut, no answer ... that’s when things get interesting.

We like to think we’re becoming more efficient. Smarter, even.

But I've just realised that there’s a fine line between efficiency and dependency, and it feels like we’re edging closer to it without really noticing.

Next time you’re about to ask AI something simple, something you already half know the answer to, it’s probably worth pausing.

Not out of principle. Just to prove you still can.

The Decisive Moment

I took this photo of a clownfish at the Jewel of the Sea Aquarium in SeaWorld, Orlando, back in April 2011. Just as I hit the shutter, a regal tang swam into frame.

It was only later, when I looked back at the image, that it clicked. I’d unintentionally captured Marlin and Dory (yes of Finding Nemo fame) together.

People often talk about Henri Cartier-Bresson and his idea of “The Decisive Moment”... that split second where everything comes together and you press the shutter with intent.

This wasn’t that.

This was pure luck. And maybe that’s what makes it even better.

Original photo by Andrew Scaife

Cleaned up by AI

Making AI Sound Human

A colleague recently sent me an AI-written product description and asked a simple question, and asked me how it looked and did it look like it was written by AI.

The short answer was "Yes." ... not because it was bad - but because it was too good.

It was clean, well structured, easy to scan, and covered every point you’d expect. On paper, it did everything right ... and that’s exactly the problem.

AI content tends to give you the sam results. Same rhythm (same number of parapgraphs in each sentence). Same tone. Same balance. It’s designed to be easy to read, which sounds like a good thing, until you realise it makes everything feel, look and sound the same.

The problem my colleague had here was that this was for website content, not only would it looks like everyone else’s, but it woud be obvious to a search engine that it was AI generated and they would therefore have a real reason to rank it. 

I love AI, I love AI content, you just need to know how to use AI properly. I used to make the mistake of just asking AI to "humanise" content, the problem was, AI doesn't know how to humanise anything, we have to tell it.

Anyway, here’s what I shared with him. 

I have also written a separate post to show the best way to get AI to sound like a human.

1. Your prompt is really important

If you ask AI “Write me a product description for...” ... you’ll get safe, predictable copy back. It’ll be fine. It’ll also be forgettable.

Change the prompt, and everything changes.

Ask it to write like someone with 20+ years’ experience who’s slightly fed up with how these products are usually described. Suddenly the tone changes. It gets less polished, a bit more opinionated, a bit more real.

That’s where things start to feel human, you've given it something human to think about, and it will change the response. What I would say here is don't be over dramatic (unless the piece asks for it), just use enough to feel honest.

2. Add opinion. AI won’t do it unless you tell it to

AI plays it safe by default. It avoids strong opinions, avoids friction, and avoids saying anything that might be challenged, we don’t (even though maybe we should sometimes!)

So feed it lines like:

“We use these all the time here.”
“This is our Sales Manager’s go-to product.”
“You can buy cheaper, but it won’t be a as good as this.”

Or just be direct, like we can all be sometimes:

“We think this is the best option on the market right now.”

I suppose what I try to do with statements like this is get a bit of emotion into the writing, AI doesn't do emotion unless you ask it to and give it examples. This sort of language instantly changes the feel of the content.

3. Break the rhythm

AI loves consistency. Same length sentences. Same flow. Same pacing ... but humands don’t write like that.

So add this to the prompt:

“Vary sentence length. Mix short and long sentences.”

It sounds basic, but it works. But it forces the content out of that predictable pattern.

And don’t be afraid of a short sentence on its own.

Like this.

But don't overdo it, AI can something provide a piece of copy with a lot of short, choppy sentences, it makes sense because it is easy to read, but it is also a giveaway that it's AI written.

5. Edit it. Properly.

AI will get you 80% of the way there, the last 20% is all down to you I'm afriad.

Read it through. change bits, add some personal references, take bits out, reformat bits, It's your content, nobody know you better than you. 

MY GOLDEN RULE: If you HAVEN'T edited it, DO NOT publish it. 

AI is a brilliant tool. I use it all the time, but getting it to sound human isn’t about pressing a button and hoping for the best, it’s about knowing how to steer it in the right direction, and then putting your own stamp on the result.