How to Write for Humans
On AI, literacy, Plain Language, and why sentences still need to make sense.
A few weeks ago, I clicked on a Fast Company article and promptly grew irate.
I wish I could say this was due to the subject matter — how the workplace fails older women — but no.
It was the writing.
The article started out fine, with an anecdote from the author about caring for her elderly mother. But I soon became distracted by the language1.
The first tip for designing workplaces to support older women was clear: “[T]reat caregiving as a normal feature of working life rather than a one-time emergency.”
But the second lost me entirely: “[A]udit who gets the future and who gets the past.”
“Who gets the future”? “Who gets the past”? When have we ever described generational inequity as such?
Using context clues from the ensuing graf, I understood what the author was trying to say. “Audit who is receiving stretch assignments and promotions” would’ve done the exact same work, with less cognitive lift.
Then, a few days later, I saw a subhed on a daily LinkedIn newsletter I enjoy skimming: “The tariff you woke up to this morning came with a date on the calendar circled”
I’m sorry. WHAT? Just… [rubs eyes, shakes head, blinks dramatically] WHAT?
Maybe: “Tariffs continue to be a headache — pay attention to the timing.”
The next day, I directed my ire at a line in a Substack by a man who fired his AI CEO: “I was shocked, and then I was embarrassed, in the specific way you’re embarrassed when a stranger’s slide deck knows your product roadmap.”
Last time I checked, a stranger’s slide deck can’t know anything about my product roadmap because it is an inanimate object2.
A simple edit: “I was shocked, and then I was embarrassed. How did my product roadmap end up in a stranger’s slide deck?”
(Maybe the AI CEO was now his executive communications manager.)
Look, I don’t want to stop reading the internet. I like reading the internet.
But I also don’t want to edit the internet.
We shouldn’t have to edit the internet.
We deserve better — on the internet and our intranets.
In This Issue
Clarity is Kind. And Trauma-Informed
But First, Nuance
These days, if I’m not distracted by this specific type of clunky construction (for shorthand, I’ll use “AI slop”), then I’m distracted by the discourse around it — and further, by my own thinking on language and power.
Over the past five-plus years — thanks to several writers, editors, scholars, and a few podcasters — I’ve worked to unlearn the rules around grammar policing and language purity.
I believe it’s always worth remembering how proper grammar and language came to be and who benefits from enforcing what is correct; that language is constantly evolving; and to interrogate why, how, and when we critique someone else’s word choices.
Generally, I’m of a mind with Sarah Khan when she writes:
The whole point of language is to communicate clearly. That’s why there are “right” and “wrong” spellings of words and why we have rules about when to use a comma and when you use a semicolon. But if the message gets across despite any errors, then the rules are irrelevant because the point of the language — to get a message across clearly — has been made.
I also believe that AI can genuinely help people who have struggled for whatever reasons to express themselves. I’ve grappled with whether being annoyed by the proliferation of generative AI writing betrays my own classism or ableism.
Lastly, I have spent a lot of time wondering whether the reason I care so much about any of this is because I began my career as a professional writer and editor.
Words are my kinda thing.
I recall an exchange on LinkedIn earlier this year with my friend and colleague Regine Nelson:
Regine: 🌶️ Hot take: The comms pros worried about AI are the ones who were only good at writing.
Ellen: [P]erhaps the only comms pros celebrating AI are the ones who struggled most with writing. (My spice is mid, I know. 🫑)
I share all of this because it’s impossible for me to tread lightly with most discourse concerning AI slop and AI in general. Nuance is critical. So is examining our biases.
On that note, I’m a professional communicator.
I believe coherence is sacred.
Not Good Enough
I personally haven’t seen too much talk of upending internal communication strategies over The End of Reading. But I spent enough time researching and developing video-forward editorial strategies earlier in my career3 to assume there are those among us who will read this essay in light of the current framing of literacy in America and think so what, who cares?
So, just in case, please know: clickbait gonna clickbait.
Trevor Aleo (and other teachers and researchers he quotes) frames the current literacy discourse as a moral panic that prevents us from addressing deeper issues:
My hope for this thesis is to remind folks that literacy has always evolved with technology. The difference today is that our informational ecosystem has outgrown our definitions and models of assessment. If we cling to a purely cognitivist view of literacy, we risk misdiagnosing the problem: blaming individual students decoding and reading fluency for much more complex and pernicious shortcomings in our educational imagination, curricular design, educational policy, political will and collective ability to design new social futures.
My fear is that the moral panic will get thrown in the corporate spin-cycle, where CEOs will hear “nobody reads,” a CCO will say “let’s do videos!” and the comms teams will increasingly outsource writing to AI because it’s good enough.
Good enough is “who gets the future.” “Came with a date on the calendar circled.” “When a stranger’s slide deck knows your product roadmap.”
Good enough is what my friends and colleagues at CRA wrote about in their last Substack.
Today’s AI models are competent enough to do a “good enough” (or better) job on a wide range of tasks that historically would’ve been handled by junior employees.
But as the authors point out, it’s important — for both the quality of the work and for professional development — to keep going. “Can you defend it?” is a better assessment than “Is it good enough?”
Can the person who produced the output, with or without AI assistance, defend and explain the myriad choices that together make up the whole: the flow of the argument, the inclusion or omission of certain details, why the body of work pursued one line of inquiry but not others, and so on?
If you can defend it — if you can explain what your words mean — then good for you. Gold star. Keep going.
(And then read the story out loud and see if you stumble over the words. Always. Revise if you do.)
If you can’t defend it, start over.
I mean it.
Plain and Simple
A few weeks back, professor and author Alison Taylor called out Fast Company for publishing an AI slop article, quoting this clunky graf, among others:
Here’s the distinction that matters. Disengagement framed as an effort problem invites effort solutions: incentives, perks, a return-to-office mandate, a new recognition program. And companies are trying it all. But you can’t pay someone to care, because what they’ve withdrawn was their fire: the creativity, the initiative, the sense that who they are and what they do are connected.
Could the author defend what “effort solutions” means? (And how RTO fits in the same sentence as incentives, perks, and recognition?)
Likely not — as Taylor noted recently, the publication took it down.
But still: If this is how AI is writing for Fast Company, how is it going inside organizations?
Employees deserve better than good enough — and customers do, too.
In her essay “Clarity is kinder: Designing content for bereaved and dying people” in Designed with Care, Michelle Keller writes, “Family members who have been caring for their loved ones through death need honest, clear, straightforward information.
“They don’t want to sift through jargon or flowery sentiment… I want families to be able to find what they need quickly and to understand the resources they find.”
The same is true inside organizations, except it’s employees searching for their benefits, a policy, an update on a change initiative, or business results.
When people can find information and understand it on the first pass, it saves time and money.
(Dare I say shareholders deserve better than good enough?)
Anyway: A simple way to do this is by writing in Plain Language.
In the excellent Accessible Communications, authors Lisa Riemers and Matisse Hamel-Nelis, citing Plain Language Association International, write:
Plain language is the language users understand the first time they read it, presented in a logical structure. It’s about choosing the right words that express your meaning, using the vocabulary your audience understands, with a simple sentence structure.
Concerned this will dumb things down? Don’t be: Riemers and Hamel-Nelis cite research that shows even experts prefer reading information in plain language. One reason:
Shorter sentences are easier for everyone to read and understand, particularly when reading on a screen.
The authors’ research suggests sticking to a maximum of one idea per paragraph, writing shorter sentences4 (max. 25 words, aim for 15-20) with “familiar words” and removing “unnecessary complexity.”
That said, watch out for cognitive bias, which happens when “when working with a subject that we’re familiar with [and we] forget how much we know about.”
This is where I find AI gets particularly tripped up, truncating context into meaningless shorthand (i.e., ”Who gets the future”).
Clarity is Kind. And Trauma-Informed
Plain Language also serves readers with differing cognitive abilities, including burnout and traumatic stress. Both can impact information processing, memory recall, and decision making.
This is where centering our employees’ capacity — editing for cognitive load — becomes critical.
In another Designed with Care essay, “Trauma and technology,” kon syrokostas explains:
Cognitive load refers to the load placed on working memory by a range of cognitive processes. In short, it’s how much “thinking” we need to do. … Trauma can impair our cognitive abilities, which means that we might have a hard time concentrating, focusing, or making decisions. Emotional activations are all too common with trauma. When we are activated, the thinking part of our brain shuts down. This can make engaging with information harder.
Beyond this, “[n]o one is reading your content in perfect conditions,” write Riemers and Hamel-Nelis. “Very people have uninterrupted time to sit back and give 100 percent of their attention to your information, however engaged and interested they are.”
In syrokostas’s list of recommendations — everything from avoiding large blocks of text and long sentences, video, or audio files to avoiding asking people to make multiple decisions at once — I found this especially relevant for our work:
“[Don’t ask] people to remember something that’s not right in front of them.”
Stay tuned for an entire Substack on this topic.
Coda
AI supposedly writes better than humans, and maybe that’s becoming more true. My colleagues at CRA routinely use AI in creating their excellent Substack and I can’t recall a time I ever stopped reading in disgust or frustration.
Meanwhile, I’m finding myself using AI increasingly less these days. Where, say, in the past I might typically have run a pass of my newsletter through Claude to ask for feedback, I don’t.
This is largely due to the repulsion I feel when I encounter work I suspect has been written with AI.
It’s not unlike how I feel about other things I don’t care for — misogynists, pattypan squash, overly drone-y songs, Tim Robinson’s comedy if I’m not in the right mood.
Just like, “yup, not for me.”
Except that these days it’s so pervasive I can’t seem to avoid it.
Everyone needs an editor, is a favorite phrase of mine.
I never imagined it’d be this true.
Currently
Building: My presentation for Flyover Festival! Between that and BrightSide, I’m going to be meeting so many of my favorite people in the next eight weeks, I’m absolutely giddy. (Need a discount for BrightSide? I’m your gal.)
Enjoying: Writing. Lifting heavy weights. Went for my first solo standup paddle since moving to Oregon and got very high on my own supply of endorphins.
Listening: This Quincy Jones cover of “Summer in the City” on repeat. Also, pretty sure I’d be real-life besties with Flo Dill from NTS Radio.
Reading: Breezed through Maria Semple’s Go Gentle. More fiction about women over 50, yes, please. Finally finished The Art of Gathering, which I found both thought-provoking and more grating than I expected.
Watching: Written on the Wind on Criterion. The color! The melodrama! Very fun. Excited to chase it with Giant ahead of an upcoming trip to Marfa, Texas.
Wondering: About November.
I’ve read descriptions of this particularly clunky writing that unfortunately I didn’t have the presence of mind to screenshot, but every time I read something like this I think of a “bit” my partner sometimes does where he greets me when I walk in the room by pretending he’s an alien.
Yes, I understand how personification works.
These, sadly, never came to fruition. Want to use my ideas? DM me.
I am trying.
When I taught survey design I'd tell my students they needed to write all their consent materials for 6th grade reading level at the max, regardless of their target population, because you can't assume people will understand what you're talking about if you use jargon or overly large words. It's odd to run into the same issues again with leaders... this weird tendency to attempt to hide what you're trying to say (or maybe you don't even know what you're trying to say) with overly complicated nothing phrases. Maybe people think it will make them sound smarter or protect them from delivering hard messages but it always does the opposite.