Brief to published piece with a review loop that has actual criteria — so "make it better" never enters the process.
Content Pipeline — 7 nodes · 7 connections. Labels on the arrows are the conditions that take each branch.
Content pipelines break in the review step. The writing gets done; then a draft enters an endless loop of "can you tighten this" and "something feels off", burning more time than research and writing combined. The reason is always the same: the review has no written criteria, so every pass invents new ones. That is exactly the failure mode an AI inherits when you hand it a prose instruction like "review the draft and improve it".
In this graph, the review is an evaluator with a checklist — brief match, structure, claims sourced, voice — and a hard revision limit: two rounds, then the piece goes to a human decision instead of round three. The brief node forces the two questions most briefs skip: who is this for, and what should they do after reading. Open it in nodalo, put your own voice rules into the review criteria, and hand it to your AI to run a real piece through the pipeline.
Each node carries a contract — what comes in, what happens, what goes out. This is the text your AI receives, node for node, when you hand the graph over.
Entry point, and the contract the whole pipeline is measured against. A brief is complete when it answers four questions: who exactly is the reader, what should they know or do after reading, which search intent or channel is this for, and what does the piece claim that competitors have not already said. A brief missing any answer does not enter the pipeline — it goes back to whoever requested the piece.
Collect the raw material: sources, data points, examples, quotes and competing articles on the same intent. Every factual claim that might end up in the piece gets a source link now — retrofitting sources after drafting is how errors survive. Output is a research note: findings with links, plus an explicit list of what could not be verified.
Turn brief plus research into a section-by-section skeleton: heading, the one point the section makes, and which evidence it uses. The outline is approved by the brief owner before drafting starts — structural feedback is cheap here and expensive after 1,500 words. An approved outline is binding for the draft.
Write the piece following the approved outline. Deviating from the outline requires a note explaining why — silent restructuring is what makes reviews unpredictable. On a revision pass, change only what the review named; a failed criterion about sourcing is not an invitation to rewrite the intro.
The evaluator. Check the draft against exactly four criteria: (1) every question in the brief is answered, (2) every factual statement carries its source from the research note, (3) the structure matches the approved outline or documents its deviation, (4) voice follows the style rules — banned words, sentence length, reading level. Fails → back to draft with the failing criteria named, at most two rounds. After the second failure, escalate to the brief owner: re-brief, restructure, or kill.
Prepare the approved piece for its channel: title tag and meta description within length limits, headings carry the terms readers search, internal links to and from related pieces, images with alt text, schema where it applies. Packaging must not change meaning — if a change would, it goes back through review, not around it.
The result: the piece is live, submitted for indexing, announced on the channels named in the brief, and logged with its target query and publish date. The log entry is what the 90-day performance review reads — a piece nobody measures is a piece nobody learns from.
The structure carries over: brief, research, outline, draft, criteria-based review, format pass, publish. For other formats you rewrite the node contracts — the outline node becomes a shot list or storyboard, the SEO pass becomes platform packaging — while the loop and its stop condition stay identical.
Only things a reviewer can check, not taste. The template ships four: every claim in the brief is addressed, every factual statement carries a source, structure follows the approved outline, and voice matches the style rules. Replace the voice rules with your own — banned words, sentence length, reading level — and the loop stops producing "something feels off".
Because a draft that fails review three times has a brief problem or an outline problem, not a wording problem. The limit forces that diagnosis: after two failed rounds, a human decides whether to re-brief, restructure, or kill the piece — instead of paying for a fourth polish of the wrong article.
Open the template, adjust the contracts, and hand the graph to your AI — as a specification it can execute, not prose it has to decode.