A finished draft lands in your queue. You have five minutes, maybe ten. You know you should read every word, and you know you probably will not. This is the checklist to keep on the desk for when that happens.
The point of the human approval step is not to catch typos. Typos are the cheap part. Your job is to catch the things a model cannot check for itself: whether a claim is true, whether the post sounds like your business, whether it answers the question a real reader would ask, and whether the title matches the post that follows it.
Two passes, not one
Read the post twice. The first pass is a skim, the second is a line-by-line read of the parts that matter. Split like that, most posts take under ten minutes.
The skim pass answers three questions:
- Does this post do what the title promises?
- Does it sound like us?
- Is it laid out the way our posts are laid out?
The line-by-line pass is only for the parts where a wrong sentence would actually hurt: any sentence with a number, a name, a product capability, a price, a date, or a link. You do not need to read a linking paragraph word for word. You do need to read every sentence that makes a factual claim.
The skim pass
Open the draft and go from top to bottom without stopping. Do not fix anything on this pass. You are looking for shape.
Does the title match the post. The most common failure with a first draft is title tension: the title promises one thing and the post is about a wider, blander version of the same thing. If the title says "five" and there are seven, or if the title asks a question and the post never answers it head-on, mark it. This is a send-back to the editor agent, not something to fix inline.
Does the post sound like your business. Read the first two paragraphs and one middle heading out loud. If the voice does not match, the writing rules or the website profile is off, not this one post. Fix the profile, then rewrite the post from it. In RankMill a rewrite is a fresh write and uses 10 credits, so it is worth fixing the profile first. A single post patched to match the voice will drift back next time.
Is it laid out normally. Are the headings sentence case if that is your rule. Are lists used where you use lists. Are paragraphs the length you like. Structure drifts more than people notice, and it drifts fastest when nobody is looking.
The line-by-line pass
Now go back and read only the sentences with weight in them.
In RankMill, a post the writer finished has a "Key points" card above the editor: a short list, written by the writer alongside the post, of what the post says. It is marked for review only and is never part of the post, and it is flagged as stale once the body has been edited. Use it as a map of the claims to go and check, not as a substitute for reading them.
Claims about the product. Every sentence that says the product does something, integrates with something, supports something or costs something is a factual claim. Check each one against what you know is true. The writer has your website profile and your answers to its questions, but a draft can still overreach: it can turn "we plan to" into "we do", or turn a soft limit into an absolute. If the sentence is not something you would be happy defending in a support ticket, mark it.
Numbers. Any number in the post is either from a source or made up. Prices, percentages, ranks, dates, headcounts, plan tiers. If the number was not in the material the writer was given, it should not be in the post. A well-behaved writer will normally write around a missing number rather than guess one, but you are the last line.
Names. Customer names, competitor names, tool names, person names. Spelling and case matter for these, and a model will happily miscapitalise a brand it has seen a hundred different ways. This is worth ten seconds a name.
Quotes. If the post quotes anyone, the quote came from you or it is invented. There is no third option. Cross-check it against the source, or cut it.
Sources. If the post links out to back up a claim, click the link and read the page. The link should say what the post says it says. A broken link, or a link to something that does not actually support the claim, is worse than no link at all.
Intent match
This is the one people skip and the one that quietly ruins posts.
Read the title again and imagine the person who searched for it. What did they want? An answer, a comparison, a walkthrough, an opinion, a template. Now read the post and ask whether that person got it inside the first two hundred words.
A post can be well written, on-topic, factually correct and still miss. It misses when it is about the subject rather than doing the job the reader came for. If someone searched "how to X" and the post spends four paragraphs on why X matters before it gets to the how, the intent is wrong. That is a send-back, not a tweak.
Internal links
Every project builds up its own set of posts, and the newer ones should link back to the older ones where it makes sense. Two things to check:
- Every internal link points at a post you still publish, at the right slug.
- The anchor text reads like something a person would write, not like a keyword.
If you notice an obvious place where the draft should be linking to another post you have and is not, that is fine to add yourself. It is a one-word job in the block editor.
Voice fit
Voice fit is the vibe check. It is not one thing you can point at, but you know when it is off. The tell is usually one of these:
- A stock phrase you would never use.
- A sentence that hedges when your business does not hedge, or does not hedge when it should.
- A metaphor that belongs to a different industry.
- A tone that talks down to your reader.
If you find one or two of these, edit them out yourself. If the whole post feels off, do not sit there rewriting sentence by sentence. Send it back to the editor agent with a note that says why: "sounds too corporate", "reads like a listicle, we do not write listicles", "assumes the reader is new to this, our readers are not". The editor will do a better job in one turn than you will in twenty.
Fix it yourself, or send it back
The rule of thumb: if the change fits in one sentence you would type, fix it yourself. If it needs more than that, send it back.
Send back to the editor agent when:
- The voice is off across the whole post.
- The intent is wrong.
- A whole section needs to go, or a new section needs to appear.
- You do not know how to phrase the fix and you want the editor to try.
Fix inline when:
- A single fact is wrong and you know the right one.
- A link is broken and you have the right URL.
- A word is missing, or a heading is in the wrong case.
- You want to cut a paragraph.
In RankMill neither route costs anything. A hand edit in the block editor saves itself without running an agent, and a turn with the editor agent in the "Ask AI" panel costs no credits either. Pick whichever is quicker: if you are about to make the same tiny fix five times in a row, describing it once to the editor agent usually is.
The printable version
Keep this next to the queue. Two minutes to skim, five to read the load-bearing sentences, one to decide.
Skim (2 minutes)
- Title matches the post
- Voice sounds like your business
- Layout matches your usual posts
- Intent is answered inside the first two hundred words
Read line by line (5 minutes)
- Every product claim is one you would defend
- Every number came from a source you can point at
- Every name is spelled and capitalised correctly
- Every quote is real
- Every outbound link opens the page the post says it does
- Every internal link points at a live post with sensible anchor text
Decide (1 minute)
- Small fact wrong, link broken, wording off in one place: fix inline
- Voice off, intent wrong, section missing, structure wrong: send to the editor agent
- Everything checks out: approve and schedule
One last thing
The approval step is not there to make the post perfect. It is there to make sure nothing goes out that you would not have written yourself. A post you approved is a post you own. Read like that is true, because it is.