Introduction
Can AI replace human writing skills? That’s the question I set out to explore this week. I’m facing off against ChatGPT in a Plain Language SmackDown. We’ll go head to head in three rounds to see who can take down beefy, bulked-up paragraphs to the lowest reading level.
(You must read the rest of the introduction in the voice of a WWE announcer. It’s important.)
Currently in the ring, from the outskirts of Oklahoma City, Oklahoma, weighing in at 30 years of professional experience and a degree in English Education, is the challenger…
Kayleen HOOOOLT.
Now stepping into the ring, from OpenAI in San Francisco, California, weighing in at 200 million active users and all the knowledge of the Internet is…
Chat. GEE. PEE. TEEEEEEE.
Are you ready to RUMBLE?
(I make no apologies for the cheesiness of this post.)
Round 1
Original Text
In round one, we have a wordy paragraph from an actual design document I was given to revise. The original paragraph scores a Flesch-Kincaid grade level score of 17.2. That’s a post-graduate level.
The educational design of the course should be a student-centered or a flipped classroom learning environment that shifts the focus of the educational outcomes from the instructor to the student. The intended educational outcome is to develop learner autonomy and independence by putting responsibility for the learning path in the hands of the students and not the instructor.
I suspect some people may be proud of a higher reading level score because they think it makes them sound smart or perhaps more professional, but the result is actually the opposite. Everyone prefers plain language, and even academics judge information as more credible when it’s easy to read.
My Revision
For each round, I rewrote the text myself before plugging it into ChatGPT, to make sure its wording didn’t influence mine.
My revision knocked the reading level down to 9.5, according to Microsoft Word’s document stats.
The course uses a student-centered design. This allows learners to be more self-directed and independent. It also uses a flipped classroom approach, which shifts responsibility for learning to the student.
The reading level score is based on sentence length and word complexity.
I was aiming for an 8th grade reading level, but 9th isn’t bad. (For anyone outside the U.S., 9th grade students are around 14 years old.)
ChatGPT's Revision
Next, I asked ChatGPT to rewrite the original paragraph using plain language at an eighth grade reading level or lower. Its revision scores a reading level of 7.5. (In the U.S., 7th graders are about 12 years old.)
The course should be designed to focus on the students instead of the teacher. This means students take charge of their own learning. The goal is to help students become more independent by letting them control how they learn, rather than the teacher deciding everything.
Round 1 Winner
ChatGPT wins this round—yikes! Will we be bowing down to our AI overlords soon?
No worries. I can use ChatGPT’s revision to help me take down the reading level even further. Replacing longer words like “self-directed” and “responsibility” with shorter phrases like “take charge” helps to simplify the language. However, The client specifically requested that we use the language “student-centered” and “flipped classroom.”
Here’s my second rewrite:
The course uses a student-centered design. This focuses on the learner rather than the instructor. The goal is to help students become more independent. The course also uses a flipped classroom approach so students can take charge of their own learning.
The second revision scores a reading level of 6.5. But did you notice how choppy this version sounds? I probably would not want to simplify the paragraph this much, because it’s boring to read. Language is a little like music. We need to listen to the rhythm of our sentences and vary their length so they’re more engaging.
A more interesting revision would look something like this:
The course uses a student-centered design that focuses on the learner rather than the instructor. This helps students become more independent. The course also uses a flipped classroom approach so students can take charge of their own learning.
This revision scores a 7.9 grade level. I’m happy with that.
Round 2
Original Text
Often in learning and development, we are tasked with creating training based on policies that contain a lot of legalese. So for round two, our hulking paragraph comes from a policy written by the U.S. Office of Personnel Management.
This paragraph has a 14.3 reading grade level score. That’s undergraduate university level—in other words, difficult to read.
The majority of position classification standards are developed by OPM and are applicable to occupations common to many or all Federal agencies. Standards for positions existing in one or a few agencies may be developed by OPM, or by a lead agency, under the guidance of OPM. All position classification standards are formally issued by OPM and distributed by the Government Printing Office, through subscription, to Federal agencies and other subscribers. Published standards remain in effect until they are abolished or replaced by OPM.
Word’s document stats also tell me that 75% of the sentences in this paragraph use the passive voice. No wonder it’s so difficult to read! (See this post for more about the passive voice.)
My Revision
My revision took the grade level score down to 10.1. Again, I was aiming for 8th grade, so I’m not happy with this one yet.
OPM writes most position classification standards for jobs that are common in Federal agencies. For less common jobs, a lead agency might write the standards with OPM’s guidance. OPM formally issues all position classification standards. The Government Printing Office distributes them to Federal agencies and other subscribers. Once published, the standards stay in place until OPM abolishes or replaces them.
ChatGPT's Revision
Again, I asked ChatGPT to rewrite the paragraph in plain language at an 8th grade reading level or lower. It must not have understood the assignment this time, because its revision comes in at a grade level of 11.3.
Most job classification standards are made by the Office of Personnel Management (OPM) and apply to jobs in many or all Federal agencies. For jobs in just one or a few agencies, OPM or a lead agency, with OPM's help, might create the standards. All job classification standards are officially issued by OPM and are available through the Government Printing Office to Federal agencies and other subscribers. These standards stay in effect until OPM changes or replaces them.
Notice how ChatGPT replaces the word “abolishes” with “changes.” Those two words don’t mean the same thing. It also changed the name of the “position classification standards” to “job classification standards,” which could cause some confusion. These are examples of why we need to use caution when using AI tools to summarize or revise content for us.
Round 2 Winner
I won this round! (Go me!) But like I said, I wasn’t happy with the 10.1 grade level score. So I took some notes from ChatGPT’s revision and rewrote it again.
The Office of Personnel Management (OPM) writes most position classification standards for common Federal jobs. For less common jobs, a lead agency might write them with OPM’s guidance. OPM issues all these standards. Then the Government Printing Office distributes them to Federal agencies and other subscribers. Published standards stay in place until OPM abolishes or replaces them.
My second revision scores a grade level of 9.3. That’s not quite to my goal of 8th grade, but because several necessary words have multiple syllables, it would be difficult to reduce it further.
Round 3
All right folks, it’s a close match! ChatGPT won the first round, and I won the second. It all comes down to this one. Who will take the championship?
Original Text
For the final round, I’m using some text from a chemistry course my team revised last year for a federal government client.
This paragraph comes from the original course materials. Its grade level score is 13.2 (university level).
More or less the same variables affecting physical states and state changes also apply to vapor pressure. Molecular weight within the hydrocarbon or derivative family plays a role. Low-molecular weight members have the highest vapor pressure. High-molecular weight members have the lowest vapor pressure. Polarity within the hydrocarbon or derivative family also affects vapor pressure. Intermolecular attraction caused by polarity acts to reduce the degree of molecular movement and thus also reduces vapor pressure. Thus, polar substances have lower vapor pressure than nonpolar substances of the same molecular weight. Molecular weight within a given polar family is the primary predictor of relative vapor pressure within the family.
My Revision
My rewrite scores a grade level of 8.7. (Beat that, ChatGPT!)
The same things that affect physical state also affect vapor pressure. One factor is molecular weight. Smaller molecules have the highest vapor pressure. Larger molecules have the lowest vapor pressure. Another thing that affects vapor pressure is polarity. This is because intermolecular attraction reduces movement of molecules. Less movement means less pressure of the vapor in contact with its liquid or solid form. So, when the size of the molecule is the same, a polar substance has lower vapor pressure than a nonpolar one.
Rewriting complex information so it’s understandable to others means we have to first understand it ourselves. That’s where a good relationship with SMEs and good internet search skills are important.
ChatGPT's Revision
Like before, I asked ChatGPT to rewrite the original text at an 8th grade level. Its revision scores a grade level of 10.1.
The same things that affect physical states and state changes also affect vapor pressure. The weight of the molecules in hydrocarbons or their derivatives matters. Hydrocarbons with low molecular weight have the highest vapor pressure, while those with high molecular weight have the lowest vapor pressure. The polarity of hydrocarbons or their derivatives also affects vapor pressure. The attraction between polar molecules makes them move less, which lowers their vapor pressure. So, polar substances have lower vapor pressure than nonpolar substances of the same weight. In a group of polar substances, the molecular weight is the main factor that predicts vapor pressure.
Round 3 Winner
In your face, ChatGPT! I mean… oh my gosh, I won!
Even though ChatGPT failed to revise the text at an 8th grade reading level, I was still able to take some of its wording to improve my revision. Here’s my second attempt, which scores an 8.2 grade level.
The same things that affect physical state also affect vapor pressure. One factor is molecular weight. Smaller molecules have the highest vapor pressure. Larger molecules have the lowest vapor pressure. Another thing that affects vapor pressure is polarity. This is because intermolecular attraction makes the molecules move less. Less movement means less pressure of the vapor in contact with its liquid or solid form. So, when the size of the molecule is the same, a polar substance has lower vapor pressure than a nonpolar one.
I also let ChatGPT take another stab at the paragraph, this time prompting it to rewrite it at a 6th grade level. That time, it produced a paragraph with a Flesch-Kincaid grade level score of 8.6. I won’t bore you with yet another rewrite of that paragraph, but I mention it to give this tip. If you’re using ChatGPT to help you reduce the grade level of your text, try telling it a couple of grade levels below your target.
For highly technical content like this, be sure to run your revisions past a subject matter expert. It can be easy to accidentally change the meaning if you’re not careful.
Summary
In this match, human writing skills bested ChatGPT in two out of three rounds. But it’s probably more productive to look at AI as a partner in the writing process rather than a competitor.
Here’s what I’d love for you to take away from this post:
- AI is not a replacement for good writing skills but can be used to help us hone our craft.
- No one’s first draft is pure gold. To be really good at writing, we need to embrace the process of rewriting.
This calls to mind the Ernest Hemingway quote, “The only kind of writing is rewriting.” Writing is an iterative process, and we can use AI tools like ChatGPT to help us in that process.
Want to Learn More?
Here are some other posts about using plain language and other writing tips:
- How to Use Plain Language to Improve Learning Outcomes
- More Plain Language Tips for Instructional Writing
- Writing Skills to Improve Accessibility
- Writing Secrets to Make Learning Stick
If you’d like a deep dive into improving your writing skills, my writing course for instructional designers (and others in L&D) is kicking off this week. Participants will get lots of hands-on practice, expert guidance, and personal feedback. We still have some spots open, but hurry!
Comments