If you happen to have emailed me recently, you’ll know that I’m currently on a sabbatical, taking time out to focus on various personal projects. One of these is to work on some writing of my own. Over the course of my three-and-a half decades in publishing, including 15 years as a ghostwriter, I’ve worked on countless non-fiction books by countless authors and clients – but this will be the first time I’ve been writing fiction in the hope of getting a book published under my own name.

Given my experience as an editor, writer and literary agent, you might think I’d be able to dash off a novel in no time, securing a publishing deal while I write. And I probably could crank out something quickly – but it’s unlikely that I’d enjoy the process or be happy with the end result. As for finding a publisher, that’s something I know is never guaranteed for a debut fiction writer anyway, so it’s the honestly the last thing on my mind just now. My whole focus is on the writing process itself and getting the most out of that. As well as being (mainly) a lot of fun, it’s an exciting challenge and an opportunity to learn as much as I can – about how to write fiction and whether it’s something I’m any good at; about my own take on life and the insights I’ve gained in the time I’ve been on this earth thus far.

The reason I’m telling you this now is that in my new guise as an aspiring fiction writer, I’ve recently had an interesting experience with AI which I think is worth adding to a debate which has been ongoing in publishing circles for some time – that is, how this rapidly evolving technology will affect the book business from here, and more specifically, whether it will eventually render human writers and editors of all kinds obsolete.

It’s important to be aware that when it comes to works of fiction, the use of AI in the writing and editing processes is still very much frowned upon by the key players in mainstream publishing. This was highlighted by Rebecca Ackermann in an excellent article in The Atlantic last month about human ghostwriters versus AI/ChatGPT: ‘Most traditional publishers won’t buy a book that’s been touched by the technology’. As a case in point, Ackermann cites the recent cancellation by Hachette US of the release of the novel Shy Girl by Mia Ballard due to ‘a barrage online accusations that it had been written with the unacknowledged help of AI’. (I hasten to add here that getting help from a ghostwriter in writing fiction is not regarded with similar distain by the industry – as long as you’re upfront at the start with your publisher about it.)

But of course none of us, not even AI insiders, knows exactly how things will evolve from here.

AI has proved to be a wonderful thing in many settings, including the book business. A good friend of mine, a sales director for one of the ‘big five’ publishing groups, has been telling me how her job has been transformed since she’s been using AI to compile the comprehensive, weekly territory-by-territory sales reports she needs to keep a finger on the pulse of buying trends in her many markets. A tedious task which used to take her days can now be completed by AI within a tiny fraction of the time; the technology also provides detailed analyses of the numbers at the touch of a button.

My own recent experience with AI has been far less positive however. Nothing dramatic or terrible – but just another example proving that at present there are key aspects of creative writing and editing for which the present technology just can’t do what an talented, experienced human being can.

I’ve been working on a short story collection for some months and for an initial time, I wanted to let the writing take its own shape, so hadn’t shown any of it to anyone. But not long ago, with 70k words under my belt, I began to feel stuck and quickly realised that I badly needed another perspective – ideally an expert one, if possible. After doing some research online, I was lucky enough to find an experienced and high-profile writer whose novels and short story collections have been published by Faber, to critical acclaim. She’s also been making a name for herself as a generous and skilled mentor for other fiction writers – and she had a space in her schedule for me! Let’s call her Ms X.

Our work together started with my submitting 40k words of the stories to Ms X by an agreed deadline, after which she needed a certain period to read and review everything. We then had two 2-hour Zoom sessions in which she presented her feedback, including a line-by-line analysis of my writing as well as an assessment of broader issues like structure, plotting and style. After this, she sent me a recording of the Zoom meetings, along with some notes she’d put together after her initial readings (‘brutally honest’, as per my request). Along with the audio files, she attached an AI summary of our sessions.

Let’s just say I completely lucked out, having Ms X as a mentor. The sessions with her were incredibly useful, productive, energising and – the icing on the cake – great fun too. Almost everything she said about my writing felt completely on point – and the few things which initially didn’t have proved, on reflection, to be worth taking into account. As well as all this, Ms X had valuable insights to offer based on her own experiences as a writer, and some brilliant suggestions when it came to possible reworking.

There was just one aspect of the process which fell down badly for me. As you’ve likely guessed, this was the AI-generated summary of the feedback sessions. I know that it was a very minor element in the mix – just an add-on, really perhaps offered by the meeting hosting site, but my first reading of this ‘report’ left me feeling really frustrated, and angry even. Dispirited too – in complete contrast with how all of the work with Ms X left me feeling.

Much of the report was simply not relevant (sometimes comically so – see below!), or woefully incomplete, evidencing a basic failure to identify what elements of the conversations would be of value to a writer in need of feedback. Some of Ms X’s most important points – about plot developments, how the dialogue was working, and possible alternative story endings had been omitted altogether.

For example, an exchange we had mid-way through the second session, when after an agreed short break, Ms X was late returning to the screen, had been minuted in excruciating detail, including a verbatim transcription of her apologetic explanation for the delay – i.e., that one of her cats had followed her downstairs, and projectile-vomited several times in quick succession on the newly mopped kitchen floor, which of course she’d had to mop again. There was a similarly unnecessary account of a digression in our chat when we talked about the routine horrors of a visit to the gynaecologist.

A human transcriber would of course have known that neither of these two exchanges (and some of the others AI had flagged) was germane to a writing feedback session, but simply the stuff of general chit-chat and female bonding.

Most painfully of all – from my perspective as a novice fiction writer needing encouragement as well as critique – the AI summary had simply excised any of Ms X’s positive statements about my work. Alongside the brutal honesty, our sessions had been peppered every so often with positive observations from her, but the most heartening feedback had come at the end of our second meeting, when she talked at some length about what I was doing right. But all of this was missing, not given a mention! Thank God I had the recordings, and could listen back to the comments which had given me such grounds for hope at the time.

In defence of the technology, I’m pretty sure this AI report was just a generic function offered by the meeting software, with a one-size-fits-all brief. And of course I in no way regard Ms X as having any part in this of – she was wonderful in every way! But imagine how long it would take a programmer to write prompts for an AI bot to enable it to identify just what a newbie writer wants to hear, balance this with what they need to hear and weigh up the kind of sophisticated and multi-layered feedback an experienced mentor may offer. There are just some things that elude programmers and tech.

No, when it comes to this kind of complex collaboration, I think it’s best to stick to real people for the foreseeable future. AI is good, excellent even, at so many things. But we can’t expect it to be good at everything – after all, it’s not even only human. . .