The Human Touch & the Limitations of AI Editing

Artificial intelligence (AI) is sneaking into all industries. Authors now find that powerful technology provides new tools (and new limitations) that are just a couple of clicks away. AI-supported editing tools like Pro-Writing Aid, Grammarly, and Hemingway App have been around for years, but now there’s ChatGPT and other generative learning AI software — the latter of which can generate not only suggestions to your writing but also entirely new content based on prompts and direction from the user.

In one scant calendar year, it can feel like we’ve gone through the looking-glass and entered a new age of AI-assisted expression — especially if you’ve been swept up by all the media hype around generative software like ChatGPT.

But what actual impact can these tools have on your writing? And how do they interact with the accredited human editors that authors and publishers have relied upon for centuries?

Today, we’ll explore 5 key editorial areas where the instantaneous (and often useful) feedback of AI software should be augmented with an author’s critical eye and a professional editor’s thoughtfulness:

1. Grammar Rules

Grammar and spell checkers aren’t new. Microsoft Word and Google Docs both provide a grammar and spell check feature that runs your text through a gamut of common grammatical and spelling errors and flags these for your consideration. These useful tools are a writer’s first line of defense during self-revision. Especially if you set your manuscript up properly with your language convention (Canadian English rather than American, for example) and autocorrect pesky issues like replacing straight quotes with curly quotes. 

However, there are still limitations to these tools. For instance, you’ve probably encountered words that aren’t recognized by the spell checker and yet, after a frantic few minutes of searching dictionaries and online, you discover you did spell the word correctly. You add it to your “personal dictionary” and continue on. Similar issues can happen when a writer is intentionally breaking a grammatical rule for effect. Your grammar checker will tell you it’s wrong . . . but is it? In fact, you’ve probably seen that rule broken in traditionally published books before.

Further, writing aids were built with a specific style guide and audience in mind. Grammarly, for example, is based on AP Style to cater towards business and journalistic writing. But commercial publishing in North America defaults to Chicago Manual of Style and — unless you’re writing a business book for business people — that formal style may not be the right tone for your work. 

A pair of educated eyes can look past these limitations and maintain a clear style for your work. A human editor can also ensure that the words you choose and the way you arrange them best suit your audience.

2. Author Voice

This leads us to your “author voice” — every author develops their own unique sound through the course of their writing. Some write exactly how they speak. Others slip into different voices depending on the character or genre they’re writing, but there’s still something quintessentially them in the plot devices or arrangement of the sentences. This is voice, and it’s the first thing a good editor strives to preserve. 

Many AI tools homogenize voice for more commonly used phrasing, or try to replace certain words or phrases with more “suitable” alternatives. You may have seen these tools warning you that your strongly worded dialogue may “offend some readers.” By implementing too many of these changes, the elements that made the writing sound like you get filed away and the content ends up sounding dry, generic, or — worst of all — cliché. 

A human editor looks for those sparkling turns of phrase, those unique metaphors, or unusual word choices that make a character come alive or a story feel fresh — and preserves them. When recasting sentences, they aim to retain the author’s meaning while making the context clearer for readers. They remove correctly spelled but incorrectly used words and replace them with the word the author intended all the time. 

And editors look for tone, delivery, and language choices to ensure they’re actually suitable for the intended audience. Your AI tool may not flag your content for audience-refining improvement without precise user input, where your human editor might note that a sentence is too complex for your Grade 5 readers, or has too much jargon for lay readers to easily comprehend. The sentence may have been perfectly correct but it wasn’t as effective as you’d like. 

3. Style Flexibility

Part of what makes text effective is delivering a reader’s expectations. Different genres have certain tones, vocabulary, even formatting conventions that help the reader orient themselves in a book. This includes levels of formality, but also includes how complex the prose is, the consistency of the points of view and tense used, and even specific word choices. 

In a Western, you wouldn’t look twice at the use of “varmint” or “smoking gun” but they’d feel out of place in a cosy baking Romance. Think of the gritty, crude language you’d find in a Noir Thriller, or the poetic elegance of a YA Fantasy. Even in nonfiction, a no-nonsense investment book will use very different language than a heart-on-your-sleeve self-help. 

AI-generative text will regurgitate something that sounds vaguely like the source it was fed, but it cannot do so in an intentional, cohesive way. It can’t measure the lyrical flow between sentences or notice repetitious sentence structure or word choice that will trip up readers. Likewise grammar tools lack the nuances to maintain style or to differentiate between different characters or speakers’ unique voices. 

A human editor can look at the work holistically, ensuring that specific characters (or contributors) sound consistently like themselves, that the narrative or framing text maintains a cohesive style, and that chronology is managed clearly with correct tense use from scene to scene.

4. Factual Accuracy

A growing problem with AI-generated text is that its factual accuracy is dubious. Because these tools are trained on content written by other people, they may mash together information in ways that are problematic. This includes perpetuating commonly believed but inaccurate factoids, incorrectly attributing works or words to people who never created them, and even providing wholly inaccurate summaries for works.

Traditional publishers and distributors require their authors to attest to the accuracy of any claims or research they present in their manuscripts, which means self-published authors are equally responsible for ensuring their books are accurate. You cannot rely solely on software to have your best interests at heart (after all, algorithms don’t have “hearts”). A human editor can help you weed out incorrect facts, names, attributions, or missing quotes. 

Note that human editors aren’t infallible either, but that extra level of critical assessment can help catch blind spots in the research, anachronisms that will take the reader out of the story, possible areas for copyright infringement, or outdated studies or language that may impede the longevity of your book.

5. Copyright Concerns

Using an AI editing tool like Grammarly does not count as authoring your text and shouldn’t impact your claims to copyright. Using AI-generative tools like ChatGPT could. In March 2023, the US Copyright office ruled that it will only assign copyright to works created by a human author(s). Distributors are requiring authors to identify when a work has been created through the use of generative tools. AI-generative tools each have their own requirements for how content produced from their models should be credited or cited in a published work. But this landscape continues to evolve quickly as creatives and companies assess how (and if) they want to incorporate these tools in their work.

If your end goal is the satisfaction of bringing your story or idea to print, we’d recommend using AI tools only as a support while drafting. You might use AI-generation to help you brainstorm ideas that you then rework or refine in your draft. You might also feed your content to an AI-generative tool to help kickstart promotional material or social media post content. However much or little you use these tools, be sure to self-edit before submitting your manuscript. If you use AI tools in your revision process, do so with engaged and proactive caution. Don’t just accept all suggestions without considering their impact. Accept those that you recognize were correcting errors. Critically consider leaving as-is those that don’t feel accurate to you, and outright reject those that change your meaning. 

Then you can submit this polished manuscript to your human editor to get their holistic and trained eye ensuring every word and line you’ve kept on the page is serving your needs. They can then catch the tricky aspects that an AI tool missed, implement a consistent style guide, and query those areas that might need further attention. 

By all means, play, explore, and test AI tools however much or little you choose to engage with them. But they’re no substitute for pure human creativity and critical editorial advice. At the end of the day, you are the author and the story is yours. Satisfaction comes from the doing as much as from the end result, so enjoy your writing journey!


Like what you just read?

Learn more in our Author's Guide to Successful Publishing - get your free copy: