Skip to content
Article Every Sep 2026

Every: what working writers who use AI actually do

Published in Every on September 9, 2026, this piece by Laura Entis is part of the “How We Write Now” series. It interviews seven writers who use AI regularly — including journalists, nonfiction authors, and newsletter writers — and asks them to describe their actual workflows rather than their opinions about AI.

What the article is about

The central finding is that AI writing workflows among serious writers are more varied, more deliberate, and more human-centered than the public debate around AI and writing tends to suggest. Most of the writers interviewed use AI heavily for research, transcription, organization, and fact-checking, and write their own prose. The exception is Dan Shipper, who uses AI to generate sentences and then iterates heavily until the text reflects what he actually means.

Context and contributors

The contributors are working writers with established editorial standards:

Alexandra Samuel used AI to turn research into podcasts she could listen to during exercise, had ChatGPT interview her to capture her argument aloud, and used Claude Code for gap analysis on her manuscript. She also built a fact-checking workflow using subagents. “AI does none of my writing,” she said.

Kevin Roose built a pipeline for his book that transcribed interviews, suggested sources, and flagged related material. He described AI as “alarmingly bad” at subjective editorial calls and used a board of AI reviewers that he found about 95% not useful. He never used AI to generate the book’s text.

Maggie Appleton dictates brain dumps and asks agents what literature she has missed, which surfaced a relevant anthropologist she had overlooked. She uses AI to suggest rephrasing of her own sentences but writes the final prose herself.

Emilia David built a custom GPT that evaluates her pitches from the perspective of her target reader. It identified a lock-in concern in an Anthropic announcement that became the angle for a story.

Katie Parrott runs her drafts through multiple AI reviewers, saves what she learns about her own patterns, and applies those lessons to future pieces.

Laura Entis writes her outline and rough draft herself first, then uses AI for structure, refinement, and fact-checking. She notes she can lean on it too much when tired or rushed.

Key takeaways

AI is strongest for transcription, research discovery, gap analysis, and first-pass fact-checking. It is weaker for structure, pacing, and editorial judgment. Several writers noted that feedback from AI requires heavy filtering — it can be useful but rarely applies wholesale. The writers who found AI most valuable had built specific tools calibrated to their own needs, rather than using general-purpose prompting.

Who it is useful for

Writers, journalists, and editors who are skeptical of both extremes in the AI and writing debate — the claim that AI will replace writers and the claim that it has nothing to offer. The article describes working practices that treat AI as a serious tool while keeping editorial judgment in human hands.