🇨🇦Want to use AI for best effect?
How to avoid AI hallucinations while meeting company mandates to use ChatGPT or Copilot
Many companies are now mandating the use of AI for written communications, for everything from RFPs to inter-company emails, via blog and social media posts. It can save time researching and writing first drafts, and it can also, in the multilingual world, help with translation. It’s an excellent resource… for those who know how to use it to get useful results, but most especially, for those who know what a good outcome looks like.
A May 2025 article by UX specialists the Nielsen Norman Group, aimed at AI vendors, also served up lessons for AI users. It's a long article but the section headings (below) tell a clear story, or cautionary tale if you prefer:
🤖 Users Struggle to Verify AI Outputs Successfully
🤖 Reducing Errors Increases Interaction Cost
🤖 LLM Outputs Signal Authoritativeness
🤖 Users Are Not Building Expertise to Spot Errors
🤖 A Finished Product Is Harder to Evaluate
🤖 Designing Checkable AI Tools
The summarized version is that AI lulls those using it to generate texts into a false sense of security. They are not able to easily substantively edit (check for facts and flow), and they overly trust what is presented. And that means you end up with results that seem fine, but there’s something not quite right about them.
Four days after the article was published, the Chicago Sun-Times was caught publishing an unverified summer reading list, where 10/15 books featured did not even exist. And this kind of falsity is apparently quite common; it’s called “AI hallucinations”.
Call in a substantive editor
If your in-house content and communications teams are already hard-pressed meeting volume and time requirements, calling in reinforcements makes sense. An experienced substantive editor will ensure that:
🧭 Overall organization of the piece is coherent – for example, that a press release actually follows the pyramid structure, and isn’t just a collection of paragraphs.
🐾 Ideas follow on from each other.
🔍 Facts are checked and corrected – which might entail rewriting.
👍 The piece can be understood by the audience it is intended for – which might also entail rewriting.
They will also carry out copyediting, ensuring the language sounds human, uses the correct words in the context, and is spelled and punctuated correctly in accordance with your house style or the intended audience. A really good editor will help your teams understand why the changes were made, so they (or the AI) can learn at the same time.
What is an experienced substantive editor?
For you to get your money’s worth, the editor you choose has to not only have experience doing this type of editing, but also be able to spot inconsistencies and dubious facts. They need to know and have some understanding of your context. At the broadest level, this could be your industry, such as aviation or manufacturing. More specifically, they might know airlines or household appliances, finance or DEI. And they need to know your audiences.
Of course, your communications teams have a role to play here too. They need to be clear on exactly who their audience is and what they are concerned with, but also what the objectives of the communications strategy are, as well as for the piece. We write in a different way if the aim is to increase or maintain employee engagement levels than we do if it’s to prepare the workforce for mass layoffs. We speak with a different voice to investors than to customers.
In our experience, however, marketing and communications teams struggle with defining their audience and their objectives, and this might mean that the AI-written piece requires some considerable rewriting. Some editors may balk at assisting them with this definition work, as it is outside their remit, and this is important to bear in mind. If you don’t have the bandwidth to conduct a definition exercise, look for an external resource that can work with you to do this in an ad-hoc, agile way.
Editing bilingual communications
In Canada, with its two official languages, or if you’re putting out communications internationally as well as domestically, you’ll want an editor that is competent in both languages. This will help ensure that important messages don’t get lost in translation of course. But it will also ensure that appropriate sentence structures are used so that audiences can relate to what you are saying (as human beings), and fully understand it. In fact, you may want to have your English-language communications edited for both North American and British audiences, for example, particularly if they are employee-targeted.
In short: To ensure good communications, it’s essential to have your AI-generated content substantively edited by someone who understands your company’s context, your audiences and your objectives. Otherwise, you risk communicating false information, being misinterpreted, being misunderstood, and failing to reach your objectives. Do you have the skills to do this in house?