Will proofreading be replaced by AI? What we discovered when we used AI for proofreading

A business wanting to save money on professional proofreading services may well be tempted to abandon hiring a proofreader in favour of artificial intelligence. But is that a wise idea? The short answer is no. Well, not yet anyway.

The rise of AI tools like ChatGPT and Google’s Gemini AI has sparked questions about the fate of many professions, and editors and proofreaders are no exception.

In fact, we were so curious that we enlisted an AI expert and tasked him with using ChatGPT 4, to set up our own Proof Communications AI Proofreading Bot. If you have a subscription to ChatGPT 4, it’s available to you for free: https://chat.openai.com/g/g-wBS5JwwP9-proof-communications-proofreading-bot

Using our AI Proofreading AI Bot is a great ‘first pass’ for proofreading even lengthy documents (over 100 pages, for example). However, be aware that right now, it’s not foolproof.

Our AI Proofreading Bot will point out inconsistencies, typos and grammatical mistakes, which is great. But there are details that only our human professional proofreaders can pick up, at the moment.

Whilst the prospect of AI taking over what is, let’s face it, often a very tedious task will be cause for celebration in some circles, most people still prefer to engage a professional proofreader to look after their important business communications. There are two reasons for that. There is a considerable care factor involved in proofreading, and bots do not possess it. And there is still a good deal of distrust about AI and just how reliable it really is.

Of course, AI is becoming more sophisticated all the time. But as Wired editor Alan Henry points out, AI can be a powerful tool but remains exactly that — a tool that needs to be guided, observed and managed by human hands.

Calling in the AI experts – what we discovered

With the help of our AI expert, we used Chat GPT-4 to proofread a PDF annual report that one of our proofreaders proofread last year for a major listed company. We wanted to see what AI picked up compared to our human proofreader. This is what we found:

Blind to formatting: AI failed to identify headings, titles and tables, often treating them as regular paragraph text. Checking formatting such as heading hierarchies is a very important aspect of proofreading, so this result was disappointing. Particularly for reports, including annual reports, the inability to read tables means it misses the full meaning of the document, which could lead it to overlook other inconsistencies.

Page-by-page analysis: AI reads each page in isolation, rather than as one coherent document. This means that it proofreads in the same way as if we’d split the document into 80 pages and given each to a separate proofreader. It makes it impossible to identify inconsistencies that occur across the document, such as ‘Non-Executive Director’ on one page, and ‘non-executive Director’ on another.

To issue a report with missed errors such as these gives an impression of a lack of professionalism, and of a disorganised business that lacks care. When reputation is everything, even small mistakes slipping through can make a big difference to how a business is perceived.

False positives and hallucinations: Out of 80 pages, the AI flagged non-existent issues on 15. In addition, sometimes it introduced new, unrelated content – a phenomenon known as “hallucinations.” These hallucinations can be difficult to spot, as the sentence reads smoothly, despite having random information inserted. The ramifications of an unnoticed hallucination creeping into a client’s annual report do not bear thinking about.

Table of contents: Some sections were formatted incorrectly, leading the AI to not recognise them and exclude them from the table of contents. All in all, a disappointing performance from AI in this test case.

These issues arose from just one report. There are, however, many more potential limitations to AI’s proofreading abilities, so it pays to be aware of them if you’re considering using ChatGPT, other AI tools are available to proofread your documents.

• Language settings: It uses US English by default. That means words such as ‘liter’, ‘organization’ and ‘labor’, and hundreds more, will be assumed to be correct.
• Structural issues: when proofreading, it won’t correct repetition or structural errors that will be obvious to the reader. Clunky wording like this is harder to read and understand, and carries a greater risk of ambiguity.
• Technical terms and jargon: Company-specific terminologies, acronyms, or industry jargon might be flagged as spelling mistakes
• Stylistic choices: Unless it has been trained on a brand’s specific tone of voice, it may “correct” deliberate stylistic choices
• Formality: In instances where a more conversational tone is used, such as a letter from the Chair, it might flag this switch as a mistake as it’s inconsistent with the rest of the document
• Wordcount: It often returns responses well below or above the specified limit.

How to use AI for proofreading

Despite these limitations, AI can still be useful for elements of proofreading. To get the most out of your professional proofreader, it’s always best to get a document in the best possible shape before handing it over, and this is where AI can help. AI is particularly good at identifying those blatant spelling errors that often become invisible to writers who have reread a document over and over. It can spot basic punctuation and grammar errors, provided it’s given guidance on preferences like American vs Australian English. When asked, it can also help identify overly complex or wordy sentences, however the replacements provided are sometimes unreliable.

If you’ve used the spell check function in Word, you’ll know that whilst it picks up some errors, it often makes mistakes or ignores them altogether. It’s the same with any AI program you might consider using for proofreading.

The human touch of a professional proofreader

Regardless of what parts of the proofreading you enlist AI to help with, a human should always review its corrections. A real-life proofreader meticulously reviews a document for grammatical, spelling, punctuation and formatting errors. This includes spotting those annoying typos, manager and manger, definitely and defiantly, you get the picture, rogue capitalisations, formatting inconsistencies, problems with heading hierarchies, and flagging content that is wordy, or doesn’t fit with the overall style and tone.

Think of AI as taking the “first pass” at a document, identifying glaring errors and making basic corrections. Then, the human proofreader comes in, reviewing the changes and refining the content, focusing on nuance, tone, and context. They’ll not only review the changes made by AI; they’ll pick up all that it missed.

Give our Proofreading Bot a go and see what you think. You just need a subscription to ChatGPT 4: https://chat.openai.com/g/g-wBS5JwwP9-proof-communications-proofreading-bot

The future of proofreading

So, will AI replace human proofreaders? In the immediate future although it’s possible, it seems unlikely. At least for the time being.

However, roles may shift whereby professional proofreaders focus less on simple errors and more on refining the content’s quality, coherence, and impact. So while AI will undoubtedly play an increasing role in the process, the intricacies of human language and the depth of understanding required mean that AI still has a long way to go to achieve the same outcomes as a human proofreader. At least for now.

Get in touch

Ready to enhance your document’s clarity and precision? Whether it’s an annual report, sustainability report, white paper or any other kind of business communication, our professional human proofreaders are here to help. Contact Proof Communications here or ring us on 1300 PROOFS (1300 776 637).