Why I prompt AI to write at a 7th-grade level — Anatoliy Babushka
Why I prompt AI to write at a 7th-grade level
June 25, 2026 · 6 min read
I prompt AI to write at a 7th-grade level by pinning it to ELA7,<br>the US Common Core language standard for grade 7, the writing<br>expected of a twelve-year-old. It forces precise, concise sentences<br>and strips fancy vocabulary, so tickets and design docs come out<br>scannable and easy to act on.
I could pick the AI-written tickets out of our backlog without<br>checking who filed them. They were the ones stuffed with words<br>nobody says in a standup ("leverage," "utilize," "robust") around<br>sentences you had to read twice.
Learning the standard taught me something I now believe about all<br>writing: the clearest document I produce is also the best prompt I<br>can hand an AI. Most people work the other way around. They polish<br>prose to sound intelligent and feed the machine sloppy<br>instructions, then wonder why the output reads like a brochure.
Here is the assumption worth challenging: that smart readers want<br>sophisticated prose. They don't. The stakeholder skimming your<br>design doc wants something concise and scannable. The engineer<br>opening a user story wants to grasp it without spending real focus<br>to decode it. Big words spend the reader's attention without buying<br>anything back.
What ELA7 actually says
ELA stands for English Language Arts. The Common Core standards<br>define, grade by grade, what students in the US should be able to<br>read and write. One line in the<br>grade 7 language standard<br>does most of the work for me. L.7.3 asks writers to "choose<br>language that expresses ideas precisely and concisely, recognizing<br>and eliminating wordiness and redundancy." That is a sharper editing<br>instruction than most style guides manage in a chapter.
Why grade 7 and not grade 5? When I was learning something new, I<br>used to ask AI to "explain like I'm 5" (ELI5). It works, but it<br>strips out too much. ELA7 sits between baby talk and an academic<br>paper. It keeps the real content and drops the ornament.
When I switched my learning prompts from ELI5 to ELA7, the<br>explanations stopped feeling like cartoons and started reading like<br>a sharp colleague explaining the thing over coffee.
One detail matters for the objection I get to below: the same<br>standard (L.7.6) expects students to use "domain-specific words and<br>phrases" accurately. ELA7 does not tell you to delete technical<br>terms. It targets the sentence around them.
Why "sounding smart" costs your reader
The mechanism is plain. Every rare word and every nested clause adds<br>parsing cost; linguists call it dependency distance. The reader holds<br>the start of your sentence in working memory until the end arrives to<br>complete it, and fancy vocabulary plus long wind-ups stretch that gap.
I have paid that tax twice over. Refinement sessions ran long because<br>the team had to stop and decode what a fancy PBI was actually asking<br>them to build. And my proposals came back covered in comments from my<br>managers, asking me to clarify points that only read as complex<br>because I had written them that way. Both burned time we did not have.
There is a motive worth naming too. A lot of fancy writing is a<br>status move. We reach for the bigger word because it signals<br>expertise. To anyone who knows the topic it signals the opposite,<br>and to the stakeholder who just needed the decision and the date it<br>signals nothing at all.
How I use ELA7 with AI
I have run backlogs in kanban and scrum since 2015, writing the PBIs<br>(product backlog items) and acceptance criteria that fill them. Since<br>I moved into management at Unity Technologies in 2021, the documents<br>got bigger: technical proposals, engineering strategy, product<br>requirements docs drafted with product managers. I came up as a<br>software engineer first, so I know how it feels to open a fancy PBI<br>and have to decode it before I can start.
When ChatGPT became usable in early 2023, I handed that backlog work<br>to it. The drafts came back too fancy. Identifiable on sight. The<br>trouble was never the length of a ticket; it was the wording. Once I<br>learned the standard, I added one instruction to those prompts.
Rewrite this ticket using ELA7 (US Common Core grade 7) language:<br>precise, concise, no wordiness, no rare words used for flavor.<br>Keep all domain terms. Then check the result against ELA7 and<br>flag any sentence that fails.
The change was immediate. Tickets got shorter and plainer.<br>Acceptance criteria read as steps a person could follow instead of<br>clauses in a contract. The backlog became something any engineer on<br>the team could pick up and act on, rather than a document only the<br>author and one or two specialists could parse.
Naming a concrete grade level beats telling the model to "write<br>simply." In my own prompts, "simple" barely changed the output; the<br>model had nothing to aim at. "ELA7" gave it a specific target, and<br>the drafts came back shorter and plainer, usually clean enough to<br>ship without a second pass.
Isn't this just dumbing...