Hazel Bird
“Draftsmith respects the way professional editors work.”
Hazel Bird is an artsy spreadsheetoholic who has been a full-time freelance editor and editorial project manager since 2007. She publishes The Edit Better Letter, a monthly newsletter for editors, proofreaders and other wordy people who set no limits on editing – and running their editing businesses – better. Her course Handling Large Editorial Projects offers tools and mindset tips to help editors avoid overwhelm and handle these projects’ unique challenges with methodological assurance.
Originally published on Word Stitch Editorial
Copyeditors vs AI: harvesting the harvester
The great insult of AI is that it wants us to thank it for the ease and efficiency with which it supplants our imperative to think. Copyeditors must constantly be vigilant against this persuasive pillaging. But many of us cannot afford to disengage from AI entirely, and we need tools that genuinely work for us, in our environment. AI has harvested from us – but what if we could harvest it right back?
Our primary relationship with AI is not technological – it’s emotional. And one of the major emotions brought up by AI is fear. Fear of losing our jobs. Fear of losing our privacy. Fear of bias proliferation. Even fear of catastrophic harm on a planetary scale.
But for those of us who work with ideas or in creative industries, there’s another – in some ways deeper – fear not so often spoken about because it’s harder to quantify.
It’s the fear of losing our ability to think for ourselves.
We might also think of it as a fear of losing our intellectual autonomy – or our ability to ideate.
It’s a fear that despite our best intentions and stalwart efforts, we will cave to the endless whispering enticements to let AI do the thinking for us (what Polly Hudson ingeniously calls ‘the smartphone equivalent of Edgar Allan Poe’s tell-tale heart beating away under the floorboards’). We fear that click by click, draft by draft, idea by idea, we’ll become hooked. Dependent. Hollowed out.
Why I value my intellectual autonomy
For me, delegating my ability to think is a firm no.
It’s also an imperative if I wish to continue doing the job I enjoy and at which my clients tell me I excel.
If I don’t stay ahead of AI, it will take my job. Nobody is going to pay me to edit if they can get the same level of service from AI. So I have to continue to use all those cognitive processes that AI doesn’t have: judgement, reasoning, empathetic connection and genuine creativity.
But habits are extremely hard to build and easy to lose. We are already seeing what is being called digital dementia. And I don’t want to be like the humans in the Pixar movie WALL-E, gradually sliding into dependence on AI and only realizing at the last moment what I have lost.
The closer you look, the less you fear
Given this fear of losing the ability to ideate, you might think I’d have been horrified to hear the team behind the AI tool Draftsmith saying that ‘a lot of people will just use it for ideas’. And I should have been even more horrified when I realized that when I gave the tool a try, that’s exactly what I was doing.
However, setting aside the generalized background horror of having to interact with AI at all, I actually felt remarkably unthreatened.
It helped that it soon became abundantly clear the tool is no threat to editors’ livelihoods (as I’ll explore below). And this is by design: Draftsmith is not specialist editing software. It’s primarily aimed at writers to help them refine their drafts, and any use that professional copyeditors get out of it is secondary.
It also helps that Draftsmith’s founder and CEO, Daniel Heuman, comes from the editing world and, as he says on the Draftsmith blog, ‘care[s] deeply about the craft of editing and the importance of editors’. I don’t believe that Draftsmith is out to replace us. And indeed PerfectIt – made by the same company – is a non-AI tool that has become indispensable for professional editors worldwide.
But the most important reason I didn’t feel threatened was that I felt Draftsmith left me firmly in the driving seat. It meets professional editors in their own environment (Microsoft Word). Its changes can be tracked just like any others. And it gives us tools that we can use to implement, ignore or modify specific suggestions as we see fit.
I’ll delve into these topics further below. First, I’m going to dispel some potential misconceptions about what Draftsmith does and doesn’t do – specifically in relation to professional copyediting. I’ll then suggest how editorial professionals might use Draftsmith, based on how I found it helpful in a real project. Finally, I’ll offer my thoughts on how Draftsmith sits within the whole complex, messy and evolving relationship we copyeditors are (grudgingly) learning to have with generative AI.
What Draftsmith doesn’t do – and why these ‘deficiencies’ don’t matter for professional editors
This isn’t a detailed technical review of Draftsmith – you can find that elsewhere. Rather, here I will offer an overview of Draftsmith’s features in terms of how they interact with the philosophy and goals of professional copyediting.
(Note: most of the time I deliberately refer to ‘copyediting’ and not ‘proofreading’ in this article. Although Draftsmith does have a Remove Typos feature, this is not a tool for professional proofreading as traditionally defined.)
Draftsmith won’t do your job for you
With Draftsmith, as the team said in an early-access tutorial, the user judges whether an AI-suggested change is good or not, ‘which is exactly the right way round’. And indeed, Draftsmith has barriers designed into it to prevent you simply accepting all its suggestions. You review each paragraph individually and changes are suggested at the sentence level, with each one shown in mock-up form outside your document (see image below). You could simply whizz through clicking Accept on everything, but the interface at least tries to make you slow down and consider every suggestion.
Why this matters. The default position in professional copyediting is not to make changes – every change has to be justified. Whereas at the writing stage authors might be throwing ideas around and trying things out, that’s not what copyeditors do. Many other AI interfaces (such as ChatGPT) will simply show you the suggested changes already integrated, putting the onus on you to find the ones you don’t want. In contrast, with Draftsmith, no change makes it into your copy without your explicit approval. In short, Draftsmith doesn’t take away the requirement to think.
An example of Draftsmith’s interface, which appears down the side of the document in MS Word. Here I’m using the Remove Typos tool on one of the paragraphs above with typos added for demonstration purposes.
Draftsmith doesn’t replace judgement, professional resources or knowledge of the client’s goals
You see how I said above that you could rush through accepting every change? Well, that would be a very, very bad idea. There is the potential for an inexperienced, untrained, undisciplined, unscrupulous (or just tired) editor to make an unholy mess of a manuscript by indiscriminately accepting Draftsmith’s suggestions. This is because Draftsmith has no true concept of context (more on this below) or editorial style, and can’t know the client’s goals – all guiding lights for professional editors.
Why this matters. We professional editors still need our style guides, our grammar sources, and our connection with the author and their goals, not to mention our expertise. Arguably, we need them more than ever, to protect us from the temptation of seemingly quick fixes that may not be the right choice for a project. Gen AI is simply another tool in our belt.
Draftsmith isn’t psychic
Draftsmith contains numerous options for refining the kinds of suggestions it offers. For example, some help you trim word count. Some help with implementing plain English or calibrating the reading level. And the Empathy Tuner has interesting options for dialling friendliness and emotionality up and down. But the one I found most helpful was the Fluency Enhancer – specifically the Improve English tool. In the project that I used to test Draftsmith (more on this below), I was focusing on smoothing out translated English, and Fluency Enhancer made some useful suggestions.
But it wasn’t psychic.
When a sentence was garbled or incomplete, Draftsmith offered potential rewrites. But it always assumed there was some sense to be gleaned from the original sentence (Dr Rachel Baron explains this ‘aim to please’ and its consequences in an article for Science Editor). Often there will be some sense to the original sentence, but sometimes not. Sometimes a sentence needs to be deleted entirely. And sometimes it’s so unclear that the only option is to ask the author what they meant to say.
Human editors obviously aren’t psychic either. But, unlike AI, we have intuition coupled with a sense of the person that wrote the original words. This gives us a kind of insight that will always be beyond AI.
Why this matters. Draftsmith doesn’t give explanations or caveats on its suggestions. A request to edit a mangled sentence in ChatGPT might elicit a suggestion with a caveat or explanations for different options (see the example from ChatGPT below). In contrast, in Draftsmith, there is no way of ascertaining the software’s certainty or reasons for the suggestions it offers. This isn’t a problem for professional copyeditors, who can judge for themselves whether a suggestion is appropriate. But it is always vital to remember that Draftsmith (like any other gen AI) doesn’t have a psychic hotline to the author’s brain. An enticingly tidy rewrite may be grammatically clear but utterly wrong.
An example, from ChatGPT-4-turbo, of gen AI’s aim to please. Although I’m tickled by the AI’s interpretation of my offhand nonsense (and technically it did obey my instruction), a human editor would clearly have identified that this sentence almost certainly needed more than basic copyediting.
Draftsmith is generally overkill for well-prepared manuscripts
Some projects reach copyeditors still needing significant development. But in many cases, by the time a manuscript reaches copyediting, it’s already had a vast amount of work done on it. On this kind of project, the author has carefully chosen and reviewed every word. They (and, where relevant, the publisher) are happy the text says what it should, the way they want to say it, and that it will reach the intended audience. At this point, any concept of drafting or improving is firmly off the table. A copyeditor who offered anything other than essential refinements – carefully judged, with detailed knowledge of the project’s goals and limitations – would be egregiously overstepping the bounds of their brief.
Why this matters. Draftsmith can’t replace professional copyediting because, as I’ve said, that’s not what it’s designed to do. If I used its more interventionist tools (which is almost all of them) for many of my projects, my clients would be right to withhold payment and never hire me again. However, that doesn’t mean it can’t be useful for some projects, as I’ll explore below.
Draftsmith has no true concept of context
As I mentioned, Draftsmith’s suggestions are now offered at the sentence level. I found this an appropriate and helpful design choice that allows you to really focus in on the details of a sentence. However, the idea that you could actually finalize a sentence in isolation would clearly be absurd. As any professional editor will tell you, context is key.
To take one example, in my test project, Draftsmith didn’t address the fact that a sentence in one paragraph referred to a ‘coalition agreement’ and a sentence elsewhere referred to ‘coalition agreements’. This isn’t a criticism of the product – the different phrases could both have been valid. But it’s important to understand that Draftsmith leaves consistency issues of this kind to the user.
Why this matters. For professional editors used to Intelligent Editing’s other major product, PerfectIt, it’s vital to understand that Draftsmith doesn’t look for issues with consistency outside a sentence. Whereas PerfectIt’s entire raison d’etre is to process (potentially vast) quantities of text and find tiny stylistic inconsistencies, Draftsmith takes a magnifying glass to a tiny portion and ignores the rest.
Welcome to our playground
The project I used to try out Draftsmith had been machine-translated into English (itself a massive issue that I don’t have space to do justice to here). Unsurprisingly, the client wasn’t happy with the quality. I’d been asked to improve the translation where possible, using AI to suggest alternative wording. It’s more usual for editorial clients to ban editors from using AI, so this explicit request to use it was rare.
Keeping our data safe
The client didn’t even require me to turn on the setting that prevents your data being harvested for AI training (although I always have this turned on by default). But here’s where Draftsmith goes one better: there’s no setting to turn on, because Draftsmith immediately deletes your data after processing. The company doesn’t use your data in any way (Daniel has written about this in more detail).
This is a big deal to me (protecting my clients’ data is point eight in my copyediting and AI manifesto). It’s also mentally liberating to feel that with Draftsmith, I’m editing without an insidious presence lurking to siphon off the good bits for its own benefit. I appreciate that I can relax in my space, on my computer, with my projects and edit without fear of my skill and creativity being exploited.
Doing things our way
This willingness to meet us where we are, and to fit into our existing ways of working, is another big plus. From custom ribbons to macros to add-ins to language and font choices and even page background colours, every freelance editor’s set-up is different. Asking us to edit in another environment isn’t just irritating. Removing access to our professional toolkit fundamentally changes the nature, speed and quality of what we can do (imagine asking a professional mechanic to work with what you have lying around in your garage).
So I very much like that Draftsmith respects the way professional editors work. It functions as an add-in within the software that almost all of us already use (eliminating the need to transfer text elsewhere and then wangle the amended version back into Word). And it works with Track Changes, so any edits we make with the software are blended seamlessly with our own.
Draftsmith also understands that professionally editing a document is about more than changing text. It has barriers built in to prevent things like formatting, footnotes and hyperlinks being lost or messed up. These are all huge plus points that, for professional editors, may make Draftsmith a much more appealing prospect than the standard gen AI offerings.
Room to grow
There are improvements that could be made. For example, changing the language setting (which governs UK vs US spelling) is currently rather awkward, and I’d love to see Draftsmith able to honour the document’s existing editorial styles (punctuation, capitalization etc). Also, I did spot some glitches in the offerings, such as incorrectly decapitalized European names, capital letters added mid-sentence and faulty changes from singular to plural. I also noticed some missed errors and skewed meanings in the suggestions offered.
But many of these are general pitfalls of gen AI (editor Adrienne Montgomerie has documented some great examples across various models). Draftsmith’s clear interface gives users far more control in spotting and avoiding these issues than the conventional chat-based interfaces. Conceptually, too, I feel the software is starting from the right place. And I’ve already witnessed that the team is keen to receive and learn from feedback, and it plans to make further enhancements in the future.
How copyeditors might use Draftsmith
The way I ended up using Draftsmith was to leave it open down the side of my screen (where, incidentally, it sits quite unobtrusively, keeping any plans for world domination to itself and politely minimizing if you need to access the Styles Pane). I then referred to it for inspiration as needed. If Draftsmith had a useful suggestion, I would either make the change manually or (more usually) implement my own change inspired by the suggestion.
I found the Improve English tool gave me the most helpful suggestions for my test project. But I do plan to get to know two of the Editing Helper tools (Polish and Redraft) better. (I like the idea of the Remove Typos feature, but the sentence-by-sentence processing means it wouldn’t be feasible across the large documents I commonly work on.)
I did try out the feature where the software makes the changes for you (following your approval) and found this works well. And the facility to modify Draftsmith’s suggestions before they are implemented in the text is clever as it leads to tidier tracking. But most of the time, because I was using the software to get ideas that I could adapt to make them suitable for my specific document and client, I found it quicker to implement my version directly.
This to me is what AI is supposed to be: an assistant kept on the sidelines if needed. And, crucially, this way of working means I tackle the sentence first. I only used Draftsmith if I got stuck, for a beat or two, with identifying the best way forward for a sentence. Even in these cases, I know I could have solved the issue without Draftsmith (after all, I’ve done so on dozens of millions of words over the past 15 years). The power of Draftsmith was its ability to speed up my thinking in especially tricky cases.
By working with gen AI in this way, I continue to use, preserve and enhance my skills.
Turning the tables: ‘harvesting’ gen AI
As I continued to use Draftsmith, this pattern of use turned into something of a game. When I encountered a tricky sentence, I’d deliberately not look at Draftsmith and craft an edit. I’d then see what Draftsmith suggested. If I felt my edit was better, then I’d allow myself a moment of satisfaction and move on. Alternatively, if I saw something I liked from Draftsmith, I’d work its suggestion into my solution. But, crucially, I’d also think about why its offering had been useful.
By working with the AI in this way – consciously and with self-control, and using it as an assistant rather than relying on it – I can get useful suggestions for suitable projects. And I’m also spurred on to be a better and better editor.
You could almost call it ‘harvesting’ gen AI.
It felt rather like being able to consult a thesaurus for whole sentences. When you use a conventional thesaurus, there are lots of options, many of which are utterly unsuitable. But one of them might be right, or you might be inspired down a path of lookups that leads you to something better. In the same way, I was able to see a range of different suggestions for ways to edit a problematic sentence. And sometimes this nudged me towards a solution more quickly than I might have identified one on my own.
I also found the software useful when I was on the fence about making a change. Would Draftsmith suggest an edit and, if so, would its suggestion be a genuine improvement? Sometimes it takes implementing a change to realize that, actually, the sentence was fine as it was. Having suggestions arrayed before me (with no effort on my part) sometimes let me bypass that process of experimentation and therefore move on from a sentence more quickly.
One final way Draftsmith may be helpful to professional copyeditors comes, ironically, from AI’s bias towards making changes. As I observed above, because it’s primarily a writing tool, many of Draftsmith’s suggestions are wholly unsuitable for copyeditors, whose imperative is to intrude as little as possible on a text. However, sometimes clients require or authorize heavier levels of intervention. An AI tool with a bias towards change may be helpful in jolting diffident copyeditors towards a more experimental or even playful approach.
No revolutions, thank you: a way forward for copyediting and gen AI?
Draftsmith’s team explicitly say that the software is not intended to be revolutionary.
That’s a good thing. We professional copyeditors don’t need or want a revolution. We’re doing just fine.
What we need are thoughtfully designed, appropriate tools that support our adaptation to a world with gen AI in it. Draftsmith isn’t such a tool and it doesn’t aspire to be, because it’s primarily designed for writers. But it, and the approach it represents, may come to be an important part of the picture.
I like what the team behind Draftsmith are doing because they’ve created a brake on AI. Yes, AI is inevitable. Yes, we have to learn to live with it. But no, we don’t have to give ourselves over to it wholesale and lose our human agency. And the Draftsmith team understand what is lost when AI is embraced for its efficiency without thought to the quality that only humans can bring.
I retain – and will probably always maintain – a deep fear of AI as a thief not only of our existing ideas but also of our future ability to think for ourselves. AI manufactures what I call ‘facsimiles of novelty’ (reformulations of existing human ideas), but it does not truly create and it does not think. The more we let AI do for us – or the less intellectually autonomous we become – the fewer ideas we will generate ourselves as a species.
But I do recognize that AI can be a helpful support to human endeavours, when kept firmly under human control and used with restraint.
As to whether I will ever truly relax around gen AI? I hope not. And in fact I firmly intend not to.