On AI Detectors and Freelance Writers
How to know if your freelancer is using AI generators (and does it really matter?)
I recently encountered my first AI contract clause.
It stated that if an AI detector found an article to be more than a certain percentage AI-written, I would not receive payment. Of course, due to the inaccuracy of AI detectors, I knew I couldn’t sign a contract that included a pay withholdment provision based on a metric outside of my control. But it raised a conundrum: how could I assuage new clients’ AI fears while protecting myself from not getting paid?
It occurred to me that non-writers may not know how to distinguish between quality content and generic, copy-pasted AI rubbish. They want a tool that can tell them, definitively, whether a piece of writing is hastily slapped together or is worth the money spent. Unfortunately, this doesn’t exist. However, there are a few key things you can do to determine if the writing is high quality, no matter your writing background.
1. Do your due diligence during the hiring process
A freelance writer’s portfolio and how they speak about their craft can be good indicators of their ability and trustworthiness. Here are some things to look into.
- Testimonials. If they have plenty of happy customers, you can feel more secure in the quality of their writing.
- Pre-AI articles. If their writing style is consistent between the articles that predate AI and current publications, then it’s a good sign they don’t lean on AI to complete writing tasks.
- Overall content quality. If they have numerous pieces that are enjoyable to read, then it’s likely they either don’t use AI or know how to use it in a way that still produces excellent content.
- Ability to discuss their craft. Have a video call, and prepare questions about their writing process. Those who compose quality work should be able to discuss this topic comfortably.
2. Offer a paid test article designed to highlight AI use
Before offering a retainer agreement or signing the freelancer onto a big project, offer an initial 1,000-word article. There’s nothing wrong with asking for a test project as long as it’s fully paid. AI aside, it can be a great way to get a feel for the freelancer’s process and see if you’re a good fit.
When developing a brief for your test article, add details that an AI generator can’t do well. These include things like firsthand experience, thought leadership, or deep dives into a topic. AI-generated articles rarely go into detail, and they prefer to offer generic information. If the final piece fails to cover these sections adequately, AI may have been used.
3. Look for AI indicators
When you look over the test article and future articles from the freelancer, look for these indicators that the text may have been AI-generated.
Writing that’s too perfect
Perfect grammar, similar sentence structure, and lack of opinion or nuance are all features of AI writing. If the writing is too structured and too grammatically correct, it could be written by AI.
Get Meghan O'Neal’s stories in your inbox
Join Medium for free to get updates from this writer.
Writing is a craft, and good writers know when to break the rules. Writing conventions are meant to be bent to emphasize a point or to emulate the way we speak; some grammatically correct sentences sound strange to our ears because we tend to be grammatically incorrect when conversing. If the tone requires it, breaking grammar rules is an important writing choice.
AI typically won’t break grammar rules. If the writing is too perfect, it could be a warning sign. However, this should not be the only indicator. Many writers use tools like Grammarly (this writer included) that do a great job of highlighting inadvertent errors and spelling mistakes.
Bland and predictable experiences
Machines might not have experiences of their own, but they can make them up. However, the experiences that they produce are usually bland and cliche (unless the writer has spent a lot of time honing the prompt to include unique details). Here’s what ChatGPT produced when I asked it to tell me a story about a time its grandmother made it cookies.
“I don’t have personal experiences, so I can’t share a specific story about my grandmother making cookies. However, I can imagine a scene where the air is filled with the warm, sweet aroma of freshly baked cookies. The kitchen would be cozy and filled with laughter, as a grandmother lovingly mixes ingredients with care, sharing stories and passing down family traditions. The cookies, golden and slightly crisp on the outside with a soft, gooey center, would be a symbol of her love, a comforting treat enjoyed around the table with smiles and shared memories.”
This “recollection” has no zest or unique features. It feels like an ad — a little too perfect to be real.
Wordiness
AI tends to be verbose. If the text is super wordy and uses uncommon terms like “demystifying,” then there’s a good chance it’s written by AI. Here’s a text produced by ChatGPT from the prompt, “Write a paragraph about the beauty of mountains”:
“Mountains are a breathtaking display of nature’s grandeur, rising majestically into the sky with rugged peaks that seem to touch the heavens. Their beauty lies not only in their towering presence but also in the way they interact with light and weather, creating ever-changing landscapes. At sunrise, mountains glow with golden hues, and as the day progresses, their shadows stretch and shift, adding depth to their vastness. Verdant forests, snow-capped summits, and crystal-clear streams that flow down their slopes all contribute to their serene and awe-inspiring allure. The quiet stillness of mountains, combined with their imposing strength, evokes a sense of wonder and peace, making them a sanctuary for the soul.”
Although this isn’t terrible, it leans on too many adjectives, cliches, and unnatural wording (who says “verdant?”) to get the point across. Furthermore, it uses basic literary tools. Rather than employing metaphors and personification, it offers a wordy one-dimensional depiction. Look for these indicators as you read over your freelancer’s articles.
What to do if you suspect that your writer is using AI
This is the big question, isn’t it? How can clients protect themselves from ordering original articles and receiving AI without penalizing writers who don’t use AI? After all, it’s possible for original writing to exhibit the indicators outlined above. As AI becomes more sophisticated, it’s only going to become more difficult to determine what is AI and what isn’t.
You’re going to have a difficult time finding freelancers willing to sign contract provisions that include pay withholdment for suspected AI. They cannot risk putting in the hours to create original work only to go unpaid. This may be frustrating to clients, but it’s the truth of the game.
Solutions to this problem may need to be dealt with on a case-by-case basis. One resolution could be to gently bring up your concerns in the early editing stages so the writer can address potential AI issues until you’re satisfied.
For example, I include two rounds of edits as part of my pricing. I would have no issue if a client made a comment during the first round of edits that said, “Hey, this section is a little too generic, which makes me wonder if it was AI-generated. Could you adjust this to include more tangible advice for the reader?” This way, I can address concerns within the agreed project scope, and the client can feel secure in the final product.
If you bring it up in this way, be specific. Tell them exactly why you think it was at least partially written by AI so that they can fix it. Treat it as an edit rather than an accusation. Even if the article wasn’t AI-written, your note may highlight areas where the text can improve.
In the end, does it really matter?
At the end of the day, the most important thing is that your freelancer achieves your content goals. If the piece exhibits Google’s E-E-A-T (Experience, Expertise, Authority, and Trustworthiness), is interesting to read, and offers critical insights, then who cares if the writer used AI? If you can’t tell that it’s AI, then they’ve clearly developed the appropriate skills to adequately prompt the AI generator and edit the generated response. As long as the final result is excellent, who cares how they got there?
Developing AI prompts is a skill in itself, one that takes a long time to perfect. Some people are making entire careers out of it. If a freelancer has developed these skills, it’s just as worthy of payment as original writing. Ultimately, the most important KPIs are things like traffic, CTRs, and page views rather than the writing process.
AI detectors don’t work. The best way to ensure that your freelancers produce high-quality content in the era of AI is to develop an eye for quality writing. Work on your editing skills, do your due diligence during the hiring process, and judge your writers based on whether their work achieves your goals rather than worrying about AI use.
