How-To Guide

How do I review a week of AI social media drafts?

A three-pass weekly review of AI drafts: check facts against a source, test the voice, set priority, and record one disposition per draft.

9 min read

By the Superdirector editorial team. The AI Career Lab draft-review guide, the Predis AI, Kontentino and FS Poster approval guides, FINRA Rule 2210, the FTC endorsement guides and Instagram's content-publishing reference were read November 15, 2026. Every worked example is a labelled hypothetical, not a customer account.

At a glance

How do I review a week of AI social media drafts?

Read the week as one batch in three passes, then write one disposition per draft. Pass one is facts: every number, date, quote, price and platform statement has to re-find its source, and anything that does not is sourced, marked or cut. Pass two is voice: compare the drafts with a few posts you already approved and look for repeated openers, recycled hooks and phrasing nobody at the brand would say out loud. Pass three is priority: decide which draft carries the week's most important message, give it the strongest slot, and let the rest earn their place or drop. Batch the review into one sitting, because a week read together shows repetition and drift that a single post hides. Write what changes as a rule the next batch can follow. On an account that publishes by itself, the weekly review is the correction loop, not a gate on each post.

  • Three passes in a fixed order: facts first, voice second, business priority third.
  • Source or cut: a specific with no stored source does not go out.
  • Batch the week: repeated hooks and brand drift only show when the drafts are read side by side.
  • One disposition per draft: keep, fix, hold or cut, written where the next batch can see it.
  • When the account publishes by itself, the review feeds corrections forward instead of holding each post.
Three cards read left to right: what to check in a draft, where its source lives, and what happens when the source cannot be found.

The fact pass reads left to right: check the specific, open its source, and cut or mark it when it cannot be found. Original diagram from this guide's fact-pass table.

Superdirector

Make it work for your brand

Find it. Remake it. Post it. Superdirector turns the viral formats in your niche into your brand's carousels, X posts and video scripts, and keeps the calendar full.

Why a week is the right unit

A single AI draft is easy to over-rate. It has a clean opening, a tidy close and no obvious error, so it passes a quick read. The same week, read together, tells a different story: the same question hook appears in draft one and draft four, the call to action is word-for-word in three of them, and the post that matters most is the shortest one in the middle. None of those defects is visible when you read one draft at a time.

That is why the review is batched. Kontentino's approval guide puts the rule bluntly, batch weekly, don't drip daily, and gives one review session of twelve posts every Thursday as the shape that beats twelve interruptions across a week. The number is their working example, not a measured limit, but the principle holds at any size: one sitting, all the drafts, one pass over the week.

Keep the number of decision makers small. Kontentino's rule is that committees comment; individuals approve, so let anyone who cares leave notes and let one person make the call. Predis AI makes the same point from the process side, saying to keep it to a minimum because we do not want unnecessary project delays. A weekly review with three approvers is not three times as careful; it is three times as slow.

Then keep the passes in order. Facts, voice, priority. Kontentino's guidance to add stages for risk, never for status applies inside the sitting too: the fact pass exists because a wrong price is expensive, the voice pass because a generic post wastes the slot, and the priority pass because a good post in the wrong week is still a miss. None of the three exists to make the process look thorough.

Pass one: check every fact against a stored source

Read the draft only for things that could have been invented: numbers, percentages, dates, deadlines, prices, names, citations and quotes. The AI Career Lab's review guide calls this the pass where mistakes are unrecoverable and gives the rule that should govern the whole sitting, if you can't source it, cut it or mark it. Not probably right. Sourced, or gone.

Two traps make this harder than it sounds. The first is that specificity reads as authority: a precise clause or a named section sounds checked even when it was generated. The second is confidence. Models are unreliable at this and completely confident about it, so the draft's own certainty is no evidence at all. A summary of a source is not proof that the source says what the draft claims; the source has to be opened.

Platform statements deserve their own line in the checklist, because they are facts about a system that changes. If a draft promises that a carousel can carry any number of slides, check it. Instagram's content-publishing reference states that carousels are limited to 20 images, videos, or a mix of the two, that accounts are limited to 50 published posts within a 24-hour period, and that carousel images are all cropped based on the first image in the carousel. Each of those is the kind of detail a draft can get plausibly wrong.

Money and regulated claims get a higher bar than the rest of the week. A price, a discount or an availability promise publishes a fact the customer will be asked about, and in a regulated industry the review duty is written into rules, not left to good practice. FINRA Rule 2210 requires that an appropriately qualified registered principal of the member must approve each retail communication before it is used. When a post carries a sponsored relationship, the disclosure has to be checked as copy, not as an afterthought: the FTC says the disclosure should be in the same language as the endorsement itself and rejects vague shorthand.

What to checkWhere the source livesWhen it cannot be found
Numbers, dates, percentages, calculationsThe original report, product page or brief, opened in fullCut the number or mark the draft as held with the specific named
Prices, discounts, availabilityThe page the customer will see, checked the same dayHold the draft; a price claim is never sent on memory
Quotes and citationsThe stored page text, read around the quote, not just the sentenceRemove the quote or replace it with a sourced paraphrase
Platform limits and formatsThe platform's own current documentationCut the promise or soften it to what the documentation supports
Regulated or sponsored claimsThe rule text or the disclosure guidance for that industryEscalate to the named reviewer; do not resolve it in the weekly sitting

One more rule closes the pass: approve the version that will publish. FS Poster's approval guide names approving a draft, not the final post as one of the mistakes that breaks a process, because an edit after the check means the checked text is no longer the live text. If the fact pass changes the words, the voice and priority passes happen on the changed words.

Pass two: test the voice against approved posts

Voice is easier to judge against examples than against adjectives. Before the sitting, collect a few posts the brand has already approved and would happily publish again. Put the week's drafts beside them and read for the same things: how sentences open, how much the brand explains, whether it uses the reader's words or its own, and where it lands a joke. Predis AI puts the written standard at the centre of its process, asking for a document with your brand's font, colors, tone, voice that every reviewer can share.

Then look for repetition across the week. The same opener in more than one draft is a style choice; in most of them it is a tell. The same rhythm in every caption, the same rhetorical question before every point, the same closing call to action: these are the habits that make a week of AI drafts read as one template. The AI Career Lab's editing pass names the cures directly, cut the throat-clearing, cut what the reader already knows, cut the stacks of hedges, and expect to remove a good share of a first draft. If you are removing nothing, you are not reviewing.

The fastest single test is the one that guide ends on: could this exact document have been sent by anyone else, about anyone else? If yes, keep editing. For a social post, FS Poster's tone line is a good companion check: tone: it reads like the brand, and it would still read fine next to tomorrow's news. The first test catches sameness; the second catches a post that is on brand but badly timed.

Read the changed lines aloud before moving on. Stumbles are useful: a sentence that trips you is usually a sentence that is doing two jobs, or one that nobody would say to a person. Fix those lines in the draft, not in your head, so the version that reaches priority is the version that will publish.

Pass three: set the week's business priority

A clean week can still be a wasted week. The third pass asks a different question from the first two: if only one post from this batch were seen, which one should it be? Name the week's job first, whether that is a launch, an offer, an announcement, a recruitment push or steady presence, and then sort the drafts by how directly they do that job. The post that carries the job gets the strongest slot, the day and time the audience actually looks. The rest either support it or give up their place.

This is where a batch review earns more than a per-draft one. Read alone, every draft is reasonable, so every draft gets a slot. Read together, the calendar shows what the week is actually about: three posts circling the same idea, one post carrying the only real news, and two posts that exist because the calendar had a gap. Cut the gap-fillers or move them to a quieter week. A business-priority pass is not a demand for a campaign every week; it is a demand that each slot have a reason.

Set the timing rule before the week starts, especially when someone else has to read the drafts. Kontentino's guide gives 24 hours for internal review and 48 hours for client review as a sane default. FS Poster adds the failure rule: an approval has a deadline: if the approver does not act in time, the post is delayed, not published by default. A priority pass that ignores the clock just moves the scramble to the day the post was due.

Write the week's decision down in one line before you close the review. Something like: this week is about the launch, the launch post goes Tuesday morning, the two how-to posts support it, and the rest move to next week. That line is what makes the next sitting faster, because it shows whether the week did what it said it would.

A ranked ladder of the week's drafts, with the post carrying the week's job at the strongest slot and filler pushed down or off.

The priority pass ranks the batch: the week's job takes the strongest slot and filler moves down or off the ladder. Original diagram from this guide's priority section.

Record one disposition per draft

The review is not finished until every draft has one outcome and that outcome is written where the next batch can see it. Four are enough: keep, fix, hold, cut. Keep means the draft is ready. Fix means a named change, written as a correction, not a mood. Hold means the draft is sound but a specific is missing. Cut means the draft should not run, and the reason matters more than the decision.

  1. Keep: the draft passed all three passes and carries a slot.
  2. Fix: write the exact change and the reason, then make it before the next pass.
  3. Hold: name the fact that is missing and who can supply it.
  4. Cut: keep the post in the list with its reason so the next batch does not repeat it.

Turn recurring fixes into rules for the next batch. If three drafts this week invented a price, the rule is that no price appears without a link to the page it came from. If every caption opened with a question, the rule is that two questions in a row is one too many. Predis AI's process advice is useful here: with written guidelines, you do not have to take up every feedback. The point of the record is to choose which notes become rules instead of relitigating the same taste every week.

Keep the versions straight. Kontentino's rule is that approval binds to a version, so any change after a sign-off reopens the decision. In plain terms: if you edit a draft after the person who signed it saw it, the sign-off no longer covers the words that will publish. FS Poster's declined posts are never deleted rule belongs in the same record, because a cut draft with a reason is the cheapest training material the next batch will get.

Two numbers make the record useful over time. Kontentino calls revision rounds per post is the single most useful health metric your approval process has, and FS Poster's process sets a limit of two rounds before the loop stops. Track which drafts took a second round and why. A week that needs three rounds on every post is telling you the brief is vague, not that the writers are careless.

What review means when the account publishes by itself

On an account that publishes by itself, the weekly review cannot be a gate, so it has to be something else: the correction loop. Superdirector's Growth plan at 199 dollars a month is one AI-run operated account, and extra accounts are 99 dollars each. You turn posting on for each account after one warning. From then on, its eligible Instagram carousels and single X text posts publish by themselves. No post asks for approval after that; turn it off any time. The review reads the batch, edits the drafts, and records the corrections that should shape the next one.

That makes the written disposition more important, not less. When a person approves every post, an unwritten note dies with the approval. When the account publishes on its own, the only thing that travels to the next batch is what you wrote down. Edit any draft. Corrections guide the next one is the product's own description of the loop, and it is the reason a weekly sitting beats a stream of one-off edits: the sitting produces the rules.

The scope decision still belongs to the customer. Our guide to which posts should require customer review sorts the classes that should stay close, regulated claims, money and price claims, third-party commitments and synthetic imagery presented as proof. The weekly review assumes that split has already been made. It does not re-open it for every draft, and it does not turn the account-level permission into a queue.

The plans stay distinct. Starter, 29 dollars a month, covers live social data, ideas, scripts and downloads in Claude or ChatGPT with no publishing, so on Starter the review is entirely before the fact. Growth carries the publishing permission and 2,000 monthly credits against Starter's 300. Whichever plan runs the account, the weekly batch review is the same three passes, in the same order, with the same one-line record at the end.

A four-step correction loop drawn beside a separate account-level posting switch shown as on, not a per-post gate.

On an account that publishes by itself, the weekly review feeds corrections into the next batch, separate from the posting switch. Original diagram from this guide's account-level section.

Frequently asked questions

How do I review a week of AI social media drafts?

Take the week as one batch and run three passes in order. Facts first: numbers, dates, quotes, prices, names and platform statements each have to re-find a stored source, and a specific that cannot be found is cut or marked. Voice second: compare the drafts with posts you already approved, cut the phrases nobody at the brand would say, and read each one aloud. Priority third: name the week's job and give the strongest slot to the draft that carries it. Then write one disposition per draft and turn the recurring correction into a rule for the next batch. The AI Career Lab describes its own review pass as three passes and about six minutes for one document, which is a useful floor for a single draft, not a target for a whole week.

Which check comes first, facts or voice?

Facts. The AI Career Lab's guide puts verification in a separate first pass and says to read only for things that could be invented, because that is the pass where mistakes are unrecoverable. Voice edits are reversible; a wrong price, a fabricated quote or a made-up date is not. Check the specifics before you change a single sentence for tone, or you risk polishing a claim that should have been removed.

What should I do with a draft whose claim I cannot verify?

Cut it or mark it, and do not publish it while the mark is open. That is the AI Career Lab's rule in plain words, if you cannot source it, cut it or mark it. Marking means the draft moves to a hold with the specific named, so the next reader knows exactly what is missing. FS Poster's process keeps a declined post in the list with its history instead of deleting it, which is the same principle: a removed draft with a reason teaches the next batch, a deleted one teaches nothing.

How many drafts belong in one review sitting?

Only as many as one sitting can read closely. Kontentino's guide gives one review session of twelve posts every Thursday as the batch that beats twelve interruptions across a week, and it presents that as a working example, not a measured maximum. If the week is larger, split it by account or by campaign and keep the passes intact, because a batch skimmed at speed loses the repetition check that made batching worth doing.

How long should a weekly review take?

Scale the single-draft time by the number of drafts. The AI Career Lab times one document at three passes and about six minutes, split across verification, cutting and putting the voice back. A week of drafts multiplies that, so the honest planning move is to cap the batch at what the sitting can finish and to move the largest claims into the fact pass first, where a single catch saves the most.

Does Superdirector ask me to approve each post?

No. On Growth, once posting is turned on for an account, its eligible Instagram carousels and single X text posts publish by themselves, and no post waits for a separate approval after that. The switch follows one warning that names the account, the public effect and the cost, and it can be turned off at any time. The weekly review therefore works as a correction loop: read the batch, edit the drafts, and let the corrections guide the next batch, with no post held for a yes.

What is the fastest test for whether a draft sounds like the brand?

The substitution test from the AI Career Lab: could this exact document have been sent by anyone else, about anyone else? If yes, it is not yours yet. For social posts, add FS Poster's tone line, whether it reads like the brand and would still read fine next to tomorrow's news. If either answer is no, the fix is usually a missing specific or a phrase nobody at the brand would say, not another round of adjectives.

Related practical guides

Run the week once, then keep the record

Pick the next week's drafts, read them together, and run facts, voice and priority in order. Write one disposition per draft and turn the recurring correction into a rule before you close the sitting.