Are you writing for humans? Then work with human editors

If you’re writing for humans, and seeking to engage with those humans through your words, then it’s a human you need in the editorial process. Why? Because machines can’t feel. 

In this article …

I’ll argue that the ultimate value a human editor brings to the table lies in their ability to experience genuine emotion, rather than the illusion of it.

Emotion is what we use to inform the editorial judgements we make, and to help our clients create texts that are truly engaging.

Without it, editing – really good, effective, mindful editing – will be absent.

The issues writers deal with

Writers inevitably experience head-scratching moments during the creative process. For example:

• the thriller: a fight scene needs to ooze with tension and urgency
• the academic article: the punctuation style in direct quotes doesn’t always match the publisher’s style
• the historical novel: a character’s dialogue risks alienating readers from a marginalised group
• the economics textbook: a complex equation must be accessible to undergraduate students
• the English Language Teaching course book: the language must be culturally sensitive to a particular regional audience
• the business website: the copy needs to be written in plain English but also be retain the brand voice
• the health communications pamphlet: the advice on managing grief needs to be practical but empathetic.

If you throw your text into an AI programme, the machine will indeed offer you suggestions for tackling all the above conundrums.

Amazing. Cheap too.

But …

How AI edits text for you.

Let’s look at how AI carries out editing work. I asked ChatGPT to explain, and it told me:

The AI breaks the writing into tokens (small text units) and analyzes grammar, sentence structure, meaning, tone, and context. Large language models are trained on huge amounts of writing. They learn patterns of clear, correct, and natural language, so they can recognize awkward phrasing, spelling mistakes, repetition, or unclear structure. The system evaluates multiple possible edits and selects suggestions that best fit the surrounding context and likely user intent. The AI generates this by recognizing grammatical patterns and predicting a more fluent version based on its training data.

So, writers and editors, here’s a question. What is ‘correct’ and ‘natural’ when it comes to language? Or more specifically, what is ‘correct’ and ‘natural’ when it comes to the specific piece of writing you’re working on right now?

Chances are, it will depend at the very least on the following:

• the audience you want to engage with
• the actions you want them to take
• what their sensitivities are
• the message you’re trying to convey
• any conventions you’re seeking to challenge
• the emotions you want to arouse
• any guidelines you’ve been asked to follow. 

Easier said than done. The choices you’ll make if you’re writing a thriller set in London might be different from an historical romance set in Mumbai, and will certainly be different from the ELT course book being published for the Middle East market, or the bereavement pamphlet that will go into every hospital trust in the UK, or the business website promoting insurance services in the US.

Here’s another question: Even if the ‘system’ can adapt its suggestions based on your prompts, do you trust it to feel its way through those critical conundrums and make editorial judgements that will protect you, your reputation and your creative effort?

The limitations of AI editing

The reason I asked that second question is because even though AI has rapidly become extremely good at mimicking human interactions, it’s still not there.

There may come a time in the future when a writer will give their text to AI and a human, compare the results and declare that they can’t tell the difference. But that time is not now.

Here’s why.

The prompt problem

My experiments with AI have taught me something interesting and critical: the quality of the output is determined by the quality of my prompts. So a simple prompt like ‘Tell me who publishes the Journal of Peace Research in 2026’ will give you a simple answer and a URL for the source of that information, which enables you to verify its answer. All well and good.

But a prompt like ‘Edit this article for me’, followed by an upload of the piece of text, might sound simple but in fact is more complex. The machine also needs to know the comprehensive set of ‘rules’ you want it to apply like (according to, say, a style guide, a set of preferences, or the particular goals you want to achieve).

And that means you, the creator, need to know information in the first place so that you can tell the AI – using appropriate prompts (not as easy as it might seem) – what to do.

Given that lots of creatives want editorial help precisely because they

• don’t know that information
• do know that information but would rather spend their time on being creative
• aren’t sure how to communicate with the AI efficiently to get the result they want

a human who has the required knowledge and can intuit the writer’s needs over a Zoom call, the phone or an email is still your best bet.

The IP problem

There’s another very serious issue to consider. If that prompting information includes the information published in a recognised style guide – commonly used in the editorial process for the purposes of consistency and clarity – the writer needs to check that the AI has:

• access a verified version of that guide
• the permission to use that intellectual property (IP).

Some software and closed AI systems have these permissions built in, but we should never assume that this is the case if we’re using open-source AI.

For example, the human who wrote this article pays for an annual subscription to The Chicago Manual of Style. In exchange for that fee, I’m provided with a licence to access, use and apply the guidance.

The leaving-alone problem

Editing isn’t just about knowing the comprehensive set of ‘rules’, conventions and preferences. Critically, it’s just as much about deciding when to ignore them, if doing so will better serve the client and their readers.

Let me give you some examples from my fiction editing work, where I regularly make mindful decisions to ignore formal stylistic conventions:

• I might choose to introduce or remove a comma because of how its presence or omission would affect the lyricism of a sentence or a character’s voice.
• I might choose to delay amending a slur in dialogue until I’ve identified whether it serves a particular viewpoint character’s arc.
• I might allow a viewpoint character in a UK-based novel to use a subtle but uncharacteristic idiomatic term because it will be relevant to the later reveal.
• I might allow thoughts to be formatted in three different styles because each one is best in the moment owing to factors like the number of words in the thought, the ease on the reader’s eye, or the formatting that’s been used in the surrounding narrative.

The point is that AI can do lots of amazing things with text, but its output must be checked thoroughly to ensure it hasn’t broken the first law of good editing: Do no harm.

Put another way, a human needs to edit the AI’s edits.

Helping or harming – who cares?

Maybe you’re thinking that all that stuff – the prompt, IP and leaving-alone issues – are things you can live with. So here’s another question. Do you want the entity that edited your text to care about whether their work helped or harmed?

Human editors: The stakes and the rewards

When your thriller ends up on my desk and I see the way you’ve ended a chapter with a line that just rips with suspense, I might call out a quiet ‘Yessss!’ and make sure I let you know that you absolutely nailed it. On the other hand, if I notice an opportunity to inject missing suspense, I’ll feel excited that I’ve earned my fee by helping you make your prose more engaging for your readers.

In other words, I give a damn. I want you to succeed.

It’s not just about you; it’s also about me and the personal stakes involved, which include:

• my pride: I get satisfaction from doing a good job.
• my reputation: I want you to say positive things to other people about my work.
• my business: I want you to pay my fee, not send an email asking for a refund.
• my sense of self-worth: Not delivering what you asked me to would make me feel ashamed and embarrassed, feelings I find unpleasant.
• my conscience: The idea that I might harm your text is abhorrent and unacceptable.

I’m not unusual. My colleagues feel exactly the same way. It’s natural for humans to feel these things; it’s in our nature. Which is a good thing – it means it’s in our best interests to do a great job for you.

AI doesn’t give a hoot

AI, on the other hand, doesn’t give a damn. It doesn’t have an interest in you or your success. If the ‘system’ gets it wrong, there are no stakes – no vanishing recommendations, no negative feedback, no shame or embarrassment, no conscience, no pride in a job well done.

It might seem like it cares, but it’s an illusion. AI’s great at learning – from experience – how to engage in ways that its users find palatable. Which is why when I asked ChatGPT if its discussion with me about AI editing could feel a bit more friendly it said:

Yes – it can, and it should be. Sorry, I got a bit stiff and textbook-y there! You were asking for something simple and conversational, and I answered like I was writing a school handout. I'll keep it more natural from here on.

It’s good. It’s so very good. It sounds emotional, but it’s not – not at all. It’s just echoing what I’ve asked for, which is great if I’m having a bad day, but useless if I genuinely want to know how it feels about the words I’ve given it.

Why editing is more than analysing ‘small text units’

If you’re seeking editorial assistance with your writing, do you want that analysis to emerge from data that’s been broken down into ‘small text units’ by a ‘system’? Ask yourself:

• Does the AI feel the sting in a phrase that wasn’t intended to discriminate but does?
• Does the AI appreciate the elegance in the explanation of a complex equation in a physics textbook?
• Does the AI sense the tension in a novel’s scene or the wow of the twist in the penultimate chapter?
• Does the AI understand the pain being addressed in a booklet about grief?
• Does the AI grasp the power of a brand voice on a business website and the compulsion it can create to buy a product?
• Does the AI think, in the moment, that it would be the perfect time to challenge a stylistic or grammatical convention?

It does not, even though it might seem that it does when it suggests solutions. It might even start apologising to you – remember that ‘Sorry, I got a bit stiff and textbook-y there!’ I mentioned above? – or reassuring you.

And you might just be fooled into thinking it cares about your feelings and your work, and might even make a good alternative to your mother, your partner or your best mate because, just sometimes, they have the audacity to challenge you and disagree with you!

And isn’t that the best and worst thing about AI – that, ultimately, we – the users – are always right?

It’s the best thing if all we want is validation. It’s the worst thing if we want mindful, considered and informed output.

Thankfully, there’s another option: an entity with a frontal cortex, whose editorial judgements will be based on empathy and emotional intelligence developed from a life’s worth of living and learning.

Why caring and feeling matters

It is only humans who can genuinely, truly feel the potential for capturing clarity, readability, brilliance, pain, joy, voice, tension, amazement, manipulation, compulsion, and every other experience that comes with absorbing a piece of text.

It is only humans who can feel the sting of a slur, the wonder of a well-explained theory, the hurt that comes with loss.

It’s only humans who can choose to ignore a respected style guide’s suggestions or a set of historical conventions and take another path … because that path is best for your text.

And when you know how your text made your human editor feel through the amendments they suggested, the comments they made or the advice they discussed with you, you’re well on the way to knowing how it will resonate with your human readers.

If you want to use AI as part of your editorial process, fine. But please, please, please use it in addition to – not as a replacement for – a person.

It’s people you wrote for. It’s people you’re seeking to engage with. And if you need help that embodies editorial excellence, it’s people you should be hiring.