Every marketing team that adopted AI writing tools in the last two years hit the same wall. Drafts got faster. Approvals got slower. The expert reviewer, the attorney, the physician, the CFO, became the bottleneck, because every draft arrived needing the same corrections: a claim that was not sourced, a term used loosely, a line that a regulator would flag.
In a regulated professional-services firm I watched articles go through six drafting passes before publication. That was not an AI problem. It was a process problem that AI made visible. This is the SOP that replaced it.
Stage 1: a brief with a source list
No article starts from a topic. It starts from a brief: the question the reader is asking, the search demand behind it, the terms it must define, the service it connects to, and, critically, the primary sources it may cite. For tax content that means the IRS, the statute, the court opinion. The writer, human or AI, works from those sources and nothing else.
Stage 2: a master prompt that encodes the house rules
The master prompt is the longest document in the process and the one that saves the most time. It holds the structure every article follows, the voice of the assigned author, the terms that are banned, the claims that always need a citation, and the reader we are writing for. It is maintained like code: versioned, reviewed when something slips through, never edited casually.
Stage 3: the verification memo
Before anything is formatted, every factual claim is listed with its source. Not "per the IRS" but the specific publication, section and date. This memo travels with the article. When the expert reviews, they review the memo first, and most of the trust is established there.
Stage 4: the compliance scan
This is the stage that cut the most revisions. In our case the rules were state attorney-advertising regulations: no guarantees, no comparative superlatives, no implied results. We built a self-check that scans the draft for banned terms and risky constructions before a human sees it. Zero flagged terms reach review. The expert never has to be the compliance officer, which is the role that made them slow.
Stage 5: the package, not the draft
The article ships for review as a package: the article, the verification memo, a video script adapted from it, and the JSON-LD schema that will mark it up. Reviewing a package takes about as long as reviewing a draft, and it means one approval covers four assets instead of one.
Stage 6: the calendar
Articles are briefed sixty days ahead and shipped at a steady rate, four new and four refreshed a month in our case. The steady rate matters more than the number. An expert who knows exactly when drafts arrive, and that each one comes verified and scanned, reviews in a day. One who gets drafts at random reviews when they can.
What changed
Six passes became one or two. Topic to publish-ready went from weeks to two or three business days. More importantly, the reviewer's comments changed in kind, from corrections to improvements, which is the only kind of review that makes content better.
The part that is not about AI
Every stage above would improve a human-only content team. AI just made the cost of skipping them impossible to ignore, because it can produce a plausible, unverified, non-compliant draft in thirty seconds. The SOP is how you keep the speed and lose the risk.
Put the verification and the compliance scan before the review, not after it. Experts will approve work that arrives already checked, and they will rewrite work that does not.
