The method
The Approval-First Content Method
How teams that publish under scrutiny — regulated industries, founder-led brands, agencies with clients to answer to — should run AI content. Written to be useful whether or not you ever use our product.
The short answer
The Approval-First Content Method treats sign-off — not drafting — as the unit of content work. Capture voice from real material instead of prompts, ground every claim in a source you can point to, run review as a designed path rather than an obstacle, and measure quality by how little reviewers have to change.
No fabricated proof. No generic AI slop. Human approval stays in the loop.
01
The bottleneck was never the draft
AI made drafts nearly free. For most teams that changed nothing, because the draft was never the constraint. The constraint is what you can defend: the founder's actual opinion, the compliance officer's sign-off, the client's approval. Generating ten times more drafts just piles ten times more weight on the same gate.
Teams in trust-heavy markets — finance, health, legal, B2B with long memories — feel this first. A wrong claim in a published post is not a typo; it is an incident. So the method starts by naming the real unit of work: not the draft, but the approved draft. Every process decision follows from that.
This inverts how most content tooling is built. Tools optimize the part that got cheap and ignore the part that stayed expensive. An approval-first workflow does the opposite: it spends its effort making review fast, legible, and safe — because that is where the time actually goes.
Optimize the approval, not the draft. The draft got cheap; the sign-off didn't.
02
Voice is captured, not prompted
"Write this in my voice" is a prompt, not a voice. What comes back is a generic register wearing your mannerisms — the em-dashes survive, the opinions don't. A voice is a set of stances (what you actually believe and would say out loud), a cadence (how your sentences move), and signature moves (the openings, analogies, and refusals that recur in your real speech).
Those can't be conjured from an instruction. They have to be captured from real material: recorded opinions, interviews, calls, and — most reliably — the copy you have already approved. Every piece that ships without edits is evidence of what your voice actually is, which means a working system gets more like you the longer you use it.
The test is simple: could a reviewer who knows the founder tell which draft came from the machine? If yes, the voice was prompted. If they hesitate, it was captured.
Build voice from evidence — recorded opinions and approved copy — never from an instruction.
03
Ground every claim in a source you can point to
Trust-heavy content dies by fabricated proof: the statistic that doesn't exist, the customer quote nobody said, the regulation summarized from vibes. Language models produce these fluently, which makes them more dangerous, not less — a confident fabrication reads exactly like a sourced fact.
The rule is strict on purpose: a draft may claim only what its source material supports, and the provenance travels with the draft into review. When a reviewer asks "where did this number come from?", the answer should be attached to the draft — not an archaeology project through someone's browser history.
This changes the reviewer's job from adjudicating truth to checking work. If a claim has no visible source, that is not a judgment call; it is an automatic revision. Reviews get faster precisely because the rule leaves nothing to argue about.
No source, no claim. Provenance travels with the draft into review.
04
Make approval a feature, not an apology
Most workflows treat review as friction to minimize, which produces the dishonest middle: an optional checkbox everyone quietly skips until the day they shouldn't have. The method replaces it with two honest paths, chosen per brand — not per mood.
The strict path is for regulated work: review is mandatory, the toggle is locked, and nothing goes out through a connected channel without an approved draft behind it. The fast path is for low-risk channels: publish directly, no queue — but hard compliance violations still block the send. Neither path pretends. That is the point.
On either path, every piece answers two questions at a glance: what state is this in, and what happens next? A reviewer should clear a queue in minutes, with the risky items impossible to wave through and the safe items impossible to get stuck on.
Two honest paths — strict or fast — chosen per brand. Never an optional checkbox.
05
Measure quality in edits, not vibes
Every AI writing tool claims quality. Only one signal survives contact with a real team: what reviewers had to change. Edit distance between draft and approved copy, the share of drafts approved untouched, and how often someone gave up and regenerated — these numbers can't be gamed by a nicer demo.
If drafts increasingly ship with light edits, the system is learning your voice and your standards. If every draft gets rewritten, you don't have a content workflow; you have an expensive way to produce first sentences. Either way, you know — and you know from your own production data, not a benchmark.
Measured this way, quality also compounds: the edits themselves are training signal. What reviewers change is the most honest description of what they want that they will ever write down.
Track edit distance, approval-without-edit rate, and regeneration. Ignore demos.
06
Keep a cadence you can defend
Publishing in bursts is how trust-heavy teams get hurt: the burst overwhelms review, standards slip exactly when volume peaks, and the archive ends up with orphans nobody remembers approving. A steady cadence keeps the approval gate honest because the gate is never the emergency.
Defensible also means auditable. Who approved this piece, when, with what compliance state, over which sources — a publishing operation should answer those questions from records, not recollection. For regulated teams that audit trail is the license to operate; for agencies it is the thing that ends the "who signed off on this?" email thread.
The archive this produces is an asset twice over: every approved piece sharpens the captured voice, and the record of process is itself proof — to clients, to regulators, to your own future team — that the operation works.
Steady cadence, full audit trail. The archive is both training data and proof.
The method in six lines
- 01
Optimize the approval, not the draft. The draft got cheap; the sign-off didn't.
- 02
Build voice from evidence — recorded opinions and approved copy — never from an instruction.
- 03
No source, no claim. Provenance travels with the draft into review.
- 04
Two honest paths — strict or fast — chosen per brand. Never an optional checkbox.
- 05
Track edit distance, approval-without-edit rate, and regeneration. Ignore demos.
- 06
Steady cadence, full audit trail. The archive is both training data and proof.
Contentelli is this method as software.
Voice captured from recorded opinions and approved copy. Drafts grounded in your connected sources, with provenance attached. Strict and fast approval paths enforced server-side, not by policy documents. Quality measured in edit distance on your own production content. You can run the method with any stack — this is the one built for it.