Copyediting and AI: a manifesto

---

If words have power, then editors will be one of the most important bulwarks against the potential negative effects of AI on our culture and indeed our entire species. These are the eight principles on copyediting and AI that I live by in my editorial work.

In a widely read post I wrote on copyediting and AI in 2023, I argued that AI won’t replace human editors – and AI agreed. Nothing I’ve seen since then has changed my mind to any significant degree.

Editing remains a human profession requiring judgement, contextual knowledge, connection, nuance, tact, patience, problem-solving, prioritization, strategic thinking and more. AI tools can be interesting adjuncts to editors’ existing arsenal (spellcheckers, dictionaries and thesauruses, non-AI Word add-ins, style guides, calculators and the like). However, I have seen no sign of a paradigm shift in how editors work.

In 2024, I hesitated to call this a manifesto because I was aware it could quickly become out of date. The word suited my strength of feeling, so I went with it. And I was pleased, when I made my latest updates, that none of the fundamentals of what I wrote had been rendered obsolete or shown to be too idealistic. Nevertheless, I will continue to update this manifesto as the AI landscape settles.

1. AI is a tool, not a solution

There are countless examples of how AI cannot be trusted to get its facts right (hallucinations) or make reliable editorial decisions (just Google 'examples of AI getting copyediting wrong'). And it’s all very well AI presenting potential edits to a text, but how is a person without editorial training to know if what is being suggested is valid, sensible, and appropriate for their specific publication, readership, voice and context?

My clients choose to come to me rather than use AI. As such, they have made the decision to pay for my unique perspective and experience as a human, and my ability to select the most appropriate solution for their project. So why would I then give them a lesser service level they could get for free by using AI themselves?

I will keep an open mind about how I can use AI to better support my clients (as I explored here). However, I currently almost exclusively use AI for editing-adjacent tasks, such as basic reference editing and checking facts and calculations, and I cannot see that changing. Certainly, nothing suggested by AI will ever make it into my clients’ files without me (or a qualified human editor I am working with) personally reviewing and approving it in detail.

Key principle: I will not allow AI to work unsupervised.

2. My words, my insights

However much time it takes and however much I struggle to encapsulate what I want to say, I will continue to write this blog and my socials posts the good old-fashioned human way. I do not care if AI gets to the point where it can do as good (or better) a job. Text produced by AI is inherently meaningless because it is a product of probability, not experience (just as it is inaccurate to say that AI ‘writes’ anything; my use of ‘generates’ in this manifesto is deliberate).

When I write, I learn things, I process things, and I get an opportunity to try to use the squishy stuff in my head to spark the squishy stuff in your head. I understand that these aspects of writing are not important to everyone, but they are important to me and I will not give them up. While I will sometimes use AI to help me brainstorm if my thinking is stuck, AI-produced content will never form the core or substance of anything I write.

Likewise, when I’m explaining editorial concepts and principles to my clients, I believe the value I provide as a human editor comes from tailoring my explanations to those clients’ specific needs, backgrounds, industries, goals and levels of experience, using my unique professional insights and analysis.

So I will not use AI to answer my clients’ questions, provide them with information or generate queries on their texts. Instead I will continue to respond organically, tailoring my communications to the client and context.

Finally, I will not adapt my writing to avoid the ever-evolving list of AI tells. In AI-produced text, these tells stand out because they are used monotonously and indiscriminately. If AI-generated text reads badly, that’s the AI’s problem, not mine.

Key principle: I will not pass off AI-generated words as my own.

3. By humans, for humans

I can't claim credit for inventing the phrase ‘by humans, for humans’, but I've seen it around and I like it. For me it means embracing the unpredictable, serendipitous, inspiring synergies that emerge when humans work together (I'll expand on this point towards the end of this manifesto).

It also means being mindful of the ethical and legal problems inherent in generative AI. For example, much AI-generated art is only possible because models have been trained on artists’ work without their consent (as explained by The Art Newspaper and The Guardian). I do not use generators that have been trained in this way. When I need images, I usually use licensed stock images (sometimes free, sometimes bought). I occasionally use ethical AI image generators (such as Tess, described as ‘the first properly-licensed image generator that enables artists to own their style’).

Similarly, my enthusiasm for experimenting with generative AI for editing has been heavily blunted by learning more about its sordid beginnings. Regardless of whether AI companies somehow adopt more ethical approaches in the future, generative AI only exists today because of the theft of copyrighted works. Numerous lawsuits and petitions are ongoing, and organizations such as the UK Society of Authors are taking action to protect authors and launching schemes to identify human-authored works.

Finally, of course, there is the severe environmental toll of the infrastructure required to support AI. Until we can establish a better cost–benefit ratio, I am uncomfortable using AI without a compelling reason.

In principle I am not against the use of AI as a support for human endeavours. However, because of these ethical concerns, as of August 2026 I have mostly adopted a watching brief. AI may yet offer incredible advances to humanity. However, for me the concerns are currently too great to justify casual or habitual use of AI.

(I have no affiliation with Tess.)

Key principle: I will not use AI to take from others.

4. Give us more

In my previous post on copyediting and AI, I wrote about the concept of ‘AI plus’, or the idea that clients may want the efficiencies of AI as well as the unique contributions of humans. Since then, I’ve kept coming across examples from history of when new technology, rather than shutting down creativity, opened the door to new possibilities.

Robin Ince had a wonderful take on this in the lecture he gave to the Chartered Institute of Editing and Proofreading conference in September 2023. He pointed out that when photography was introduced, painters didn't just go away. They expanded into different movements (Cubism, Dadaism, etc) and became more creative.

Another example comes from fashion history. When the sewing machine was invented in the nineteenth century, it revolutionized garment production. But people weren’t satisfied with having functional garments quickly produced: they wanted the time saved to go into producing more – more frills, more ruffles, more trimmings, more embellishments.

Aristotle said that ‘nature abhors a vacuum’, and I would argue that so does human creativity. AI is opening up a blank space previously filled with humdrum necessities, and there our creativity can flourish.

We’re told that AI will make everything more efficient, and maybe my editing will end up taking less time, if I start to feel I can trust AI as a tool. But this doesn't mean I will take unjustified shortcuts or give my clients less substance.

Key principle: I will seek to give my clients more, not less.

5. Hear each other

When I'm talking to a client or emailing them about their project, there's a moment I often notice. Sometimes it happens in our initial conversation and sometimes it happens when I start sending them queries on the details of their manuscript.

It's a moment of relaxation. I can almost hear the relieved mental 'aaaaaaaaah' as they realize that someone understands all those little things they’ve been stressing about (and more) and has the expertise and sensitivity to deal with them appropriately. They know they can now return that stressed-out part of their brain to where it belongs – their core business – and leave the editorial stuff to me. I've experienced this too as a client (as I've recorded in my post Confession: I am an editor and I love being edited).

I don't believe the same feeling of being heard can come from a computer program that at some point in the past ingested data points somewhat related to a client's concern. AI does things like acknowledging, processing, analysing, parsing and executing. But it doesn't 'hear' in the sense that humans can, and any support it offers is just whatever it thinks the user is most likely to want it to say (probability again, potentially with a sprinkle of sycophancy).

Of course, this puts a responsibility on us as humans to hear each other better and truly tailor our support around what our clients need. Just like AI, we can acknowledge, process, analyse, parse and execute. In some cases, AI can do these things better than we can (or at least more quickly), but we can also hear. If we don't hear each other, we lose our edge.

Key principle: I will continue to endeavour to make my clients feel heard.

6. Drivers, not passengers

For me, refusing to let AI into the driving seat is about more than good professional practice. It is an active choice to stand firm against the potentially homogenizing effects of AI on human culture, creativity and language.

Ultimately, while generative AI might seem to be able to create things, it cannot. Instead, it works by synthesizing responses from its vast banks of human-created data. It doesn't have genuinely new ideas.

In the short term, that's probably not going to be a problem: AI has access to enough data to smash existing ideas together for a while to create interesting facsimiles of novelty. But long term, I worry about the potentially degenerative cultural, scientific and psychological effects of humanity ceding creative primacy to AI and only ever recycling our own ideas (the technical term for this is model collapse and it has potentially horrifying consequences for the preservation and availability of pre-digital content).

This means I will encourage my clients to find and retain their unique voice, and to use language in new ways when the context and intended audience allow. I will champion human writers, with AI as a potential support but never the substance. I will continue to seek out and build on the views of other humans (editors and others). I will resist the idea that knowledge is monolithic and can always be delivered through a simple search query (as editors are fond of saying about editorial matters, 'it depends'). And I will continue, in my own little way, to add my ideas to the pool of human thought.

Key principle: I will consider the consequences.

7. Trust clients and readers

When I originally wrote this manifesto in June 2024, I had decided not to enter into conversations with my clients about their use of AI until and unless it became necessary. This decision was largely based on a gut feeling that, as I wrote, ‘in the fields in which I edit, [no] credible writer is going to think they can cobble together some AI text and expect to be taken seriously’. This instinct has been borne out in conversations I’ve since had with my clients.

Nevertheless, I do now have a clause in my terms and conditions requiring clients to disclose their AI use prior to editing beginning. This is partly to give myself choice about what I do and don’t take on (while I sometimes work on texts with heavy AI involvement, I prefer to avoid it). But it’s also part of my duty to my clients: AI-generated texts read differently (there’s a reason that ‘humanization’ has become a new editing skill, as editors Josh Bernoff and Mary Kole explore) and as an editor I need to adapt my approach accordingly when I’m working on these projects.

So far so good: trust cannot be blind, but it is still my default position. But what about my instinct to trust readers – in other words, to have faith that (enough) readers will reject AI slop?

The picture here is more complex as clearly a lot of people are using AI to create content (if you’re wondering whether this reality contradicts my conclusion about trusting clients, I’ll draw your attention to the words ‘in the fields in which I edit’ above). However, I believe the extent of the pushback from consumers shows that many simply don’t want the slop.

In the original version of this manifesto, I applauded authors, such as Kester Brewin, who insert AI transparency statements into their books. It was also great to see the new (18th) edition of the Chicago Manual of Style making various strong recommendations about authors disclosing their AI use (even if only to their publisher). Since then Hachette’s dropping of Shy Girl has underscored that readers still expect mainstream fiction to be original. The AI Sidequest newsletter gives several examples of the extent to which publishers and awards panels are willing to go to avoid pushing AI-generated content. Initiatives like the C2PA and Books by People (alongside ones like the Society of Authors scheme that I mentioned in section 3) are all focused on transparency and thus giving readers the right to choose what they consume. And the EU’s new AI Act means that ‘companies must ensure people know when they are interacting with an AI-powered chatbot or viewing a manipulated image or AI-generated text’ (as reported by The Guardian).

There is a way to go as we learn what living with AI means. However, my hope is that AI-generated text will simply become the new average against which human creativity is measured. Because, as we saw in section 4, we humans don’t sit on our laurels. We may now have an arms race of creative expression on our hands, but I’m confident we will come out on top.

Key principle: I trust that human content will continue to be valued.

8. No harvesting, thank you

As I’ve mentioned, my AI use so far is fairly minimal. Nevertheless, just as I take care to protect my clients’ data by using secure storage and backup software, I seek to protect their text from AI training.

On the most basic level, this means that I always turn on the ‘don’t allow training’ button in any AI tool I use. I also make my general AI policy clear in my terms and conditions. And, given that we’re long past the point where consumer-focused AI lived neatly in tools like ChatGPT, my privacy policy lets my clients know which software I habitually use, so they can decide whether they are happy for their data to be exposed to those tools.

More broadly, though, wherever possible I customize my AI use to each client. I seek to learn their levels of comfort and agree boundaries, making any AI use a mutual choice rather than an imposed policy. With my longer-term clients, I share my learning about AI openly so we can update (or not) our consensus accordingly.

Key principle: I will treat my clients as partners in my choices about AI use.

Do we value our humanity?

I believe we are at a rare (perhaps unique) moment in history when we have to decide as a species whether we value our humanity.

The doomsayers believe we will choose in the negative: that AI will sweep human agency away and establish itself as a new hegemony that we must depend on for our economic and creative sustenance.

But I believe we can have the best of both worlds, using AI's potential where it serves us and continuing, as a species, to explore the vacuum Aristotle spoke of so long ago.

This is a matter of individual choice. We don't have to let AI run roughshod over everything we have built over the past 10,000 years, just because some commercial interests want us to. And I believe we won't, because doing so would be against human nature.

Imagine two identical books, one written by a human and one generated by an AI. In a blind test, we would expect two groups of humans to react to the two texts in similar ways. But now imagine that the groups come to the respective texts knowing who or what authored them. I suspect their reactions would be markedly different, with the human text valued more highly (and not just because of biases against AI). I believe we will always value texts more once we know they come from a human (see this article by Maggie Appleton on the ethics of this issue in art).

This is, of course, nonsense. A text doesn't change because of the nature of its author. Except that, in a way, it does. Once we know something was created by a human (or an organization representing humans), we want to know more about that specific human (or group of humans). Our experience of what they created is enriched by our knowledge of them, their history and their unique personhood or purpose (or, in some cases, by a dearth of knowledge and a consequent urge to speculate).

Perhaps this is narcissism on a species-wide scale, but if it is I would argue that it is good, healthy narcissism. We should be interested in each other, as well as feeling variously inspired, shocked, validated, informed, horrified, hopeful, energized, frustrated, enthused, bored, saddened, loving, enraged, joyful, confused, blessed, regretful, angry and every other human emotion in response to each other's words.

In short, we will never be curious about AI as an ‘author’ in the same way that we are curious about other humans as authors. And, when it comes to copyediting and AI, editors have a crucial role to play in championing human thought and creativity, so that this curiosity continues to have opportunities to manifest.

To return to Aristotle for the last time: we are nature and we will fill any vacuum caused by AI.

---

What does AI say about whether it will replace human editors? (June 2024)

Just as I did in my post on AI in January 2023, I'm going to give the last word to AI. Here are ChatGPT-4o‘s responses to the same questions I asked it last year, plus a new question on the dangers of using AI in copyediting (except where noted, the answers are presented verbatim and are the first ones ChatGPT-4o produced). I leave you to make of them what you will.

What is copyediting?

ChatGPT-4o: Copyediting, also known as line editing, is the process of reviewing and correcting written material to improve accuracy, readability, and fitness for its purpose. This includes checking for grammar, spelling, punctuation, and syntax errors, as well as ensuring consistency in style and format. Copyediting also involves verifying facts, checking for proper word usage, and sometimes suggesting improvements to the text's clarity and flow.

Here are some key aspects of copyediting:

1. Grammar and Spelling: Correcting grammatical errors, typos, and misspellings.
2. Punctuation: Ensuring correct and consistent use of punctuation marks.
3. Syntax: Improving sentence structure for clarity and readability.
4. Consistency: Maintaining consistency in style, tone, and formatting throughout the document.
5. Fact-Checking: Verifying facts and figures to ensure accuracy.
6. Clarity and Flow: Suggesting changes to improve the overall readability and flow of the text.
7. Adherence to Style Guides: Ensuring the text conforms to specific style guides (e.g., APA, MLA, Chicago).

Copyeditors work closely with authors to polish the text and make it ready for publication, ensuring that the final document is clear, accurate, and professional.

Are you a copyeditor?

ChatGPT-4o: As an AI language model, I can perform many tasks related to copyediting, such as correcting grammar, spelling, and punctuation, ensuring consistency in style and format, and suggesting improvements for clarity and readability. However, there are a few differences between what I can do and what a human copyeditor does:

1. Subjectivity and Nuance: While I can provide objective corrections and suggestions, human copyeditors are often better at understanding and preserving the author's unique voice and style.
2. Contextual Understanding: Human editors might have a deeper understanding of the specific context, industry standards, or niche subjects that an AI might miss.
3. Creativity and Judgment: Copyediting often requires nuanced judgment calls, which humans can make based on their experience and understanding of the audience and purpose.
4. Interaction and Feedback: Human copyeditors can interact with authors to clarify intentions, provide feedback, and discuss potential changes.

How can you help professional copyeditors?

(Note: ChatGPT-4o's original answer to this question contained a very long, complex list, so I asked it for a second answer, saying 'Please rewrite your previous answer without using a list'.)

ChatGPT-4o: As an AI language model, I can significantly enhance the workflow and efficiency of professional copyeditors. My capabilities include identifying and correcting grammar and spelling errors, and suggesting improved sentence structures for clarity. I ensure adherence to specific style guides and check for consistency in terminology and formatting throughout the document.

By handling large volumes of text quickly, I can process bulk tasks efficiently and automate repetitive duties such as formatting citations and abbreviations. I offer preliminary fact-checking for dates, names, and other details, and assist in correctly formatting references and bibliographies.

In terms of improving text, I provide suggestions to enhance overall clarity, readability, and narrative flow, ensuring smooth transitions. I help maintain a consistent tone suited to the target audience and adjust language to align with audience expectations.

My role extends to providing feedback on drafts, highlighting areas that need attention, and acting as a second set of 'eyes' to catch errors that might be overlooked. Additionally, I offer explanations on grammar rules, style guides, and best practices, supporting the ongoing learning and development of copyeditors.

Do you think that one day you might replace professional copyeditors?

ChatGPT-4o: While AI has made significant advancements in copyediting tasks, it is unlikely to fully replace professional copyeditors. AI can handle many aspects of copyediting efficiently, such as correcting grammar and spelling, ensuring consistency, and suggesting improvements. However, there are several areas where human copyeditors have distinct advantages.

[Long list omitted here as it essentially repeated earlier answers.]

AI will likely continue to serve as a powerful tool for copyeditors, helping them work more efficiently and effectively. It can handle repetitive tasks and provide a preliminary review, allowing human editors to focus on the more nuanced and creative aspects of their work. Rather than replacing professional copyeditors, AI is more likely to augment their capabilities, making the editing process faster and more accurate.

What dangers should people consider when using you for editorial support?

When using AI like me for editorial support, there are several potential dangers or pitfalls to consider:

1. Overreliance: Relying too heavily on AI for editing may lead to complacency or a decrease in critical thinking skills. It's essential to remember that AI can make mistakes and should be used as a tool rather than a replacement for human judgment.
2. Lack of Contextual Understanding: AI may not fully understand the context, tone, or intended audience of the text, leading to suggestions that are not appropriate or accurate.
3. Privacy and Security Risks: Sharing sensitive or confidential information with AI platforms may pose privacy and security risks, especially if the data is not adequately protected or encrypted.
4. Bias in Language and Recommendations: AI models can inherit biases from the data they are trained on, leading to biased language suggestions or recommendations. It's crucial to be aware of potential biases and critically evaluate AI-generated content.
5. Loss of Human Touch: Over-reliance on AI may result in a loss of the human touch and personalized interaction that comes with working with a professional editor or collaborator.
6. Misinterpretation of Ambiguous Text: AI may struggle with understanding ambiguous or nuanced language, leading to incorrect interpretations or suggestions.
7. Limitations in Creativity and Judgment: AI may lack the creativity and judgment skills of human editors, particularly when it comes to making subjective decisions or interpreting complex text.

To mitigate these dangers, it's essential for users to maintain a balanced approach when using AI for editorial support. Human oversight and critical evaluation are still necessary to ensure the quality and appropriateness of the final edited content. Additionally, users should be mindful of privacy and security considerations when interacting with AI platforms.