How to Use AI Without Sounding Like AI

AI generated posts are everywhere lately. Whenever I look at my Facebook, LinkedIn or Instagram feeds I see the same generic style where every post sounds exactly the same as the one before and the one after. It’s not so much the words and phrases that are used (though ‘It’s not X, it’s Y’ is a classic tell), it’s that the voice, tone and rhythm are identical each time. I’m guilty of posting similar bland content myself because it’s so easy. Giving an AI model a very rough idea and asking it to create something post-ready for LinkedIn or Facebook saves time and stress, and avoids the dreaded blank screen syndrome (when you really need to get anything at all up so you can stick to the three posts per week minimum). An AI model can do in seconds what it takes most of us many minutes, and sometimes much longer, to do, and it does it with perfect spelling and grammar to boot. 

So, having seen more and more of these generic posts, I decided that I need to up my own posting game. My output will from now on be in my own words, with my own typos and grammatical errors, but also my own voice, style, verbal tics and personality. And because it’s what I do, I decided to write a blog to help other people find their own voices again too, and continue to use AI to help speed up the process. Because this blog isn’t about not using AI just because the output all sounds the same, it’s about using AI while ensuring that your own voice is still coming through, and your posts sound like they’ve been written by a human and not come off an AI conveyor belt. There are two ways to go about this - templates for when you need something up fast, editor-mode for when the voice really matters, and I’ve covered both below.

Tell AI How You Communicate
The first step to making sure you’re getting genuinely personalised content from AI is by personalising the AI model that you use. Have a look at the sidebar or settings of whichever one you prefer - there should be a ‘customise’ or ‘personalise’ button to click on. From here you can change various settings, including the ‘voice’ the model uses to chat directly with you. You can also include stuff here to help it remember who you are, what you do and what your preferences are. 

Next up create a simple description of your voice, style and tone. You can ask AI to do this for you by giving it some examples of your actual writing. That could be letters, emails, blog posts, social posts or a mix of everything. Five examples is more than enough as long as your own voice and personality shine through. Here is what Claude says about my style when I gave it a couple of my previous blogs to analyse:

“You write in plain, direct British English with a wry, slightly exasperated edge. You're warm but you don't fuss, and there's an unsentimental honesty running through it, you'll happily admit your own mistakes and you're just as happy to call out other people's nonsense. Your sentences vary in length and you use rhythm deliberately. You'll build a longer, flowing sentence and then land a short one after it for emphasis.”

ChatGPT described my style in six words: ‘Plain-speaking, practical, opinionated, dry, direct, human’. It’s a fun exercise, of course (I particularly like that ChatGPT described me earlier in that same conversation as ‘slightly grumpy’ - who says machines don’t have feelings…), but it has practical applications too. You can take this 50-100 word voice profile of you and how you sound and ask any AI model you use to create content in that voice. It can also be tweaked to whichever task you have - more professional for a letter to a client, more lighthearted or casual for an Instagram post.

Don’t forget to tell AI what to avoid too - I hate em-dashes (though I love a hyphen) and corporate jargon, so this also forms part of my ‘personality’. If you have a particular dislike for certain punctuation, or there are words and phrases that you never use, include this. 

As with most tech, the old ‘garbage in = garbage out’ idiom is true of AI models. You will of course get usable output from a carelessly written prompt with no paragraph breaks and no hierarchy, but the output will be so much better if your prompt is clear and explicitly states what you want, who the audience is, the style you want it to follow and how the output should be formatted. I’m not talking about perfect spelling and grammar here (AI is very good at deciphering typos), just that all of the relevant information is included and laid out clearly. Writing a good prompt once means you’re more likely to get a useful and usable response, rather than going back and forth with the AI for 30 minutes before you get to where you want to be. Plus, if you’re on a free or limited tier, conserving tokens with focused and tight prompting means you’re less likely to hit your usage limit at the wrong moment. 

The basic prompt layout for pretty much every prompt follows more or less the same rules each time, and should go something like this:

Context > Task > Objective > Audience > Requirements > Output

Here’s what that actually means in language that us non-techy people can understand:

• Context: what does the AI need to know?
  Who you are, what you're doing, relevant background. 

• Task: what do you need the AI to do?
  Create, write, rewrite, summarise, analyse, brainstorm. 

• Objective: what are you trying to achieve?
  Cold-calling script, LinkedIn post, quote chasing email, supplier complaint. 

• Audience: who is the output for?
  Customer, colleague, supplier, HMRC, general public. 

• Requirements: what must it include or avoid?
  Length, tone, key points, an example of the style or format if you have one. 

• Output: what should the finished result look like?
  Email, list, blog post, table, bullet points, analysis. 

A great trick for this is to keep a prompt templates library (even just a single-page document). For example, you post on Instagram three times a week with varying content - a marketing post, a behind-the-scenes post and a new product post. You can create three prompts, one for each post style. Include all of the above and just change the Objective and Output parts each time. So you now have targeted prompts that include instructions on how to sound, and you’ve just streamlined your content creation. You can do the same with client emails - one for enquiries, one for confirmation, one for money chasing. Get the prompts written once and all they need is a quick edit and you’re good to go. 

Use AI as an Editor not a Copywriter
The most foolproof way to retain your voice and personality is to use AI as an editor. Scanning for typos, flow, logical progression, grammar and accuracy is a breeze and can be done in seconds. I can (and have) read through a blog or social post I’ve just written ten times over and still miss a glaring typo, only to find it immediately I click publish. But Claude never does, with the added bonus that it’s on call 24/7, so if I’m writing early morning or late at night, I can still get the post finished without waiting for office hours. I write the post, run it through Claude for a quick SPaG check and I’m good to go. If you’ve asked Claude to write something for you, in your own voice, you can get the AI model to check its own work by including your voice profile and asking for a comparison between the AI output and your voice profile, and an assessment of whether it stays true to your personality.

Of course if you want an objective assessment of something you’ve written, then you need to tell the AI that specifically. AI models are not deliberately designed to be lovely and flattering, but if they were horrible to talk to or made us feel stupid and dull they would be penalised during training. The sycophancy is actually a by-product of the humans training them, rather than a cynical ploy to exploit us. What this does mean is that you need to explicitly ask the AI model to adjust away from their normal ‘you’ve just written the next bestseller’ attitude and into a more balanced approach. 

The words you need to use for this are critique, assess, analyse, evaluate and so on. You can ask the AI to ‘critique harshly’ or to ‘be realistic, do not be kind in your assessment’ or you can ask it to ‘give me some practical feedback on how I can improve this piece’. My most used phrase when it comes to blog editing is ‘Hi Claude, please read the attached and critique. Please also check for SPaG, flow and accuracy.’ Two seconds later you have an assessment of your post, letter or blog, with any errors clearly flagged. 

One last word here - don’t be afraid to disagree. Just because the AI has access to more knowledge and information than we could ever learn in a lifetime, does not mean that the AI is better, cleverer or more intuitive than us. You know your customers, your business, your life, how you react, what makes you tick. AI knows none of these things, so don’t assume it knows better than you. 

Staying Safe
Before you skip down to the prompt templates I’ve included here, please do read this part first. I’m always banging on about GDPR and personal data and it’s boring, I know. But it’s also really important, and in this age of data breaches, phishing and cyber threats, it’s more important than ever to be up to speed with protecting your own and your clients' data. Before you attach that document or you upload those letters, check there is no identifying information included. This is especially important if you’re using letters and emails you’ve sent to clients or suppliers that contain their details. Don’t give any AI model any personal details about yourself beyond your name and what you do. Don’t type the names or contact details of clients into the chatbox if you want the model to create an email or letter for you: ask it to include placeholders for the name, email and any other identifying information that needs to be included. You can find out more about GDPR laws and how they apply to UK sole traders and freelancers in the Resources section of this website, where there is a link to the Information Commissioner’s Office. I’ve also written a blog about GDPR which you can read here.

Prompt Templates
Here are the two templates. The first builds your voice profile, the short description you'll paste into everything else you write. The second uses it to turn a rough idea into a finished post, email or blog in seconds. 

PROMPT 1: Create My Voice Profile

Below are examples of my writing. Analyse them and write a 50-100 word

description of my voice, style and tone, one I can paste into another AI

as instructions for writing as me.

Cover:

- Vocabulary and sentence rhythm

- Tone, personality and formality

- Humour, opinion and quirks

- Anything I consistently avoid (words, phrases, punctuation)

Describe how I write. Don't rewrite me, improve me, or tell me how to write.

[PASTE OR ATTACH YOUR EXAMPLES]

PROMPT 2: Create My Content

Write the content below in the voice described. Follow the requirements exactly.

VOICE PROFILE

- [Paste your 50-100 word voice profile]

CONTEXT

- [What you need to know about the situation]

TASK

- [What to create]

OBJECTIVE

- [What it needs to achieve]

AUDIENCE

- [Who will read it]

REQUIREMENTS

- [Length, key points, anything to include or avoid]

OUTPUT

- [Blog / email / letter / social post]

Don't invent facts, claims or opinions I haven't given you unless I ask you

to research. No generic AI language, corporate jargon, unnecessary headings,

filler or clichés. Return only the finished content unless I ask for an explanation.