Why do editors reject AI writing pipelines, and what shape do they accept? One-shot generation fails for three reasons that have nothing to do with the model. The fix is a workflow, not a prompt.
An editorial team producing long-form marketing content had tried one-shot AI drafting and thrown it out. Their reasons were specific: no sources, no way to fix one paragraph without regenerating everything, and no way to see why the model wrote what it wrote.
What they run now, built on Make.com, Airtable and OpenAI:
1. Five stages, five scenarios: research, outline, draft, score, polish. Each is a Make.com scenario triggered by a status change in Airtable.
2. A human gate between every stage. A piece cannot move until a person sets the status. Airtable is the control panel and the editor owns it.
3. Every stage stores what the model was given and what it returned, on the record. An editor can read why a draft says what it says.
4. Notes carry forward. What the editor writes at outline is in the brief at draft.
5. A scoring stage grades the draft against the brief before a human reads it, so the editor opens it knowing where it is weak.
6. Single-stage re-runs. A wrong section means re-running that one stage, never the whole article.
7. Prompts live in Airtable next to the stage, so the team changes tone or structure without opening Make.com.
Ten small background automations do the plumbing: Slack notices, a source check that flags claims with no source before polish, formatting and hand-off to the CMS.
Output: 1,800 to 4,000 word articles with a source list per piece, run by the team without an engineer. The client reports production time per article fell from six to nine hours to sixty to ninety minutes, their estimate.
The diagram shows the five stages and the gates. The full case study is on my profile, and the same shape applies to proposals, reports and product descriptions, anything where a human has to sign off.
Prem Patel, Nex Automations. Make.com Level 5 certified, official Make partner, Zapier Certified Expert.
Why do editors reject AI writing pipelines, and what shape do they accept? One-shot generation fails for three reasons that have nothing to do with the model...