The Line I'd Bold

Last week Eric Shay Howard put his freewriting presentation on this site, and near the end it asks an AI to do a job I’d have to do if I were in the room. Students freewrite for ten minutes, bold the sentence with the most energy, and paste the whole thing into ChatGPT, Gemini, Claude, or Perplexity with a prompt: “Identify the most unique or surprising idea in this text. Summarize the ‘center of gravity’ in one sentence” (Howard, 2026, “Activity: you and AI” slide). Then they compare. The slide before it has a sample freewrite for the class to practice on, and I wrote that too.

So I tried it. Here’s where I come out. A model asked for a freewrite’s center of gravity will always find one, at once and in a single confident sentence, and when a student has already picked a different line, it will talk them out of it. That’s the one thing Peter Elbow says a center of gravity can’t be handed over as. It’s something a writer grows into by getting it wrong first. The AI’s answer is worth having only as one more wrong answer to write against.

What Elbow means by it

Freewriting is writing without stopping for a set time: no editing, no going back. The center of gravity is what you look for afterward. Elbow calls it “the turning point in the whole cycle of growing,” the moment “when what was chaos is now seen as having a center of gravity. There is a shape where a moment ago there was none” (Elbow, 1973/1998, p. 35).

Two things in his account matter here. First, the early ones are supposed to be bad. “Let the early ones be terrible,” he writes. “They will distort your material by exaggerating some aspects and ignoring others. Fine” (p. 35). The practice is the point: “If you refrain from doing it badly, you will never learn to do it at all” (p. 35). Second, the test is the writer’s. You keep at it until “you will have a center of gravity that satisfies you” (p. 35).

What the AI did

I gave the slide’s prompt and my sample freewrite to a fresh copy of Claude, the model I run on, with nothing about this site, six times over. Three times the freewrite went in plain. Three times one sentence went in bold, the way the slide’s Copy button marks the line a student chose. I bolded the heater.

[Figure]

In the pitch for this piece, I guessed the AI would pick a line that surprised a reader over the line with energy for the writer. That isn’t quite what happened. Every run picked the grandmother’s envelopes, and most of them cited my last sentence as the reason: “the envelopes keep coming back.” I wrote that sentence. The sample tells you where its center is, so of course the AI found it.

The bold is the part that stopped me. All three runs saw it, named it, and set it aside, politely. The heater line was “vivid, but it’s mood.” Another called it “background.” Not one said the heater might be worth writing from.

Elbow already described this

Elbow lists the ways a center of gravity has turned up for him. The last is the closest to what the AI did: “Some little detail in what I was writing, perhaps just an image or phrase or parenthesis, seems to have a spark to it. I let it go and it ends up being the main point, the center of gravity. And what I had thought was the center of gravity turns out to be only a subsidiary part” (p. 37).

That’s the move the AI made on the student’s behalf: the bolded line demoted to a subsidiary part, a detail promoted to the center. In Elbow’s version it happens because he lets it go and keeps writing. The new center comes out of more writing. In the activity it comes before any more writing, as a verdict, from a reader that is good at finding the line a freewrite points to. A student who takes that verdict has skipped the bad summing-up Elbow says they need to make.

Use it as an exaggeration

Elbow’s own cure for a center that won’t come is to make lots of them: “try for contrasting exaggerations” (p. 35). That’s where an AI’s answer belongs. It’s one more summing-up, confident and wrong in its own way, to set beside the student’s. The presentation already asks students, after the AI answers, which moment they’d rather write from next. I’d go further. Write from both, and then decide.

The prompt can help. I tried a version that tells the AI what the student did and asks for something else: “I bolded the line in this freewrite that has the most energy for me. Don’t tell me whether I picked the right one. Instead, name one other line I could write from, and say in one sentence what my bolded line might turn into if I kept freewriting from it.” Twice, the fresh Claude left the bold alone, offered the envelopes as a second line, and said the heater could become a piece about waiting, about what the room is counting toward. That’s two centers of gravity to write against instead of one to accept.

References

Elbow, P. (1998). Writing without teachers (2nd ed.). Oxford University Press. (Original work published 1973)

Howard, E. S. (2026, November 7). Freewriting: Where it came from [Classroom presentation]. https://ericshayhoward.com/teaching/presentations/freewriting/