The ten-minute approval habit that keeps a machine blog honest

When a machine writes the posts, people assume the human job disappears. It does not. It changes shape — from production to judgment — and it gets much smaller. Ten minutes a week, in our experience, for a blog publishing weekly.

But it only stays ten minutes if you know what you are looking for. Approvers who try to edit machine prose the way they would edit a junior writer's draft end up spending an hour and improving nothing, then quietly stop approving at all. That is how an automated blog dies.

Stop reading for the things that used to go wrong

The habits of reviewing human drafts are miscalibrated here. Grammar, spelling, paragraph structure, transitions, consistent tense — these were where a hurried human writer failed, and they are not where a model fails. Reading for them is a waste of the ten minutes and it crowds out the reading that matters.

Style is a second trap. If a sentence is merely not how you would have phrased it, leave it. Voice is governed by the style profile, and one-off edits do not propagate — you will make the same edit next week. If you find yourself making the same correction twice, that is not an edit, it is a profile change, and it should be written down once rather than applied fifty times.

Read for four things

Is it true? Every factual claim about your product, your pricing, your customers, or your market. This is where errors actually live, and they are invisible to the machine, because the machine cannot know what your sales team promised last week.

Is it ours to say? Sometimes a post is accurate and still wrong to publish: it pre-announces something unreleased, it contradicts a position taken in a customer contract, it names a partner who would rather not be named. Only you can catch this.

Is it a claim, or is it filler? Does the post argue something specific, or does it merely circle a topic? Filler is the failure mode of automated content, and it is best caught by a human asking whether they learned anything.

Would we be happy if a customer read this? The blunt version of every other question.

Two of these take seconds. The truth check is where the time goes, and it is worth spending it there rather than distributing attention evenly across the page. In practice most posts need no change at all, one in five needs a factual correction, and roughly one in twenty should not go out in its current form. If your rejection rate is zero, you are not reading; if it is above a third, the subject list is wrong rather than the writing.

Give the approver a default and a deadline

An approval step with no deadline is a bottleneck disguised as diligence. Decide in advance what happens when nobody responds by the cutoff.

We recommend "hold, don't publish" for anything making product or pricing claims, and "publish" for general commentary — but the choice matters much less than making the choice explicitly. An undefined default means the answer depends on who is on holiday, which is the same as having no process.

One named approver, not a committee. Two approvers is not twice the safety; it is each one assuming the other read it properly.

Log the rejections, they are the most valuable output

Every rejection contains information the system needs. A rejected post tells you the subject list drifted, or a fact in the brief is stale, or the profile permits something it should not.

Rejections that get logged make next month's posts better. Rejections that get fixed silently in the draft make next month's post fail identically. This is the single highest-leverage habit in the whole workflow, and it is the one almost everyone skips.

What this buys you

The reason we insist on approval rather than offering full automation is not caution for its own sake. It is that the approval step is what separates a blog from a content farm — and, as we have argued before, that distinction is also the one Google's policies actually turn on.

Ten minutes a week is a real cost and we would rather quote it honestly than pretend it is zero. It is also, by some distance, the cheapest quality control any publishing process has ever had.