Here is a post Autopilot wrote for my own account, in my own voice, during its first week running unsupervised:
Consistency is the secret to social media growth. Show up every day, provide value, and watch your audience grow. What's one thing holding you back from posting consistently? Let me know below.
It is grammatical. It is on topic. It uses my vocabulary. And it is completely, fatally dead. Nobody in the history of the internet has stopped scrolling for "provide value, and watch your audience grow." I would never post it. The system was a few hours from posting it for me.
The worse part is that nothing flagged it. The low-confidence hold, the mechanism that is supposed to catch bad output, stayed silent, because the model wasn't unsure. It was confident and boring. That is the gap every "we'll run your social media" tool has to reckon with: confidence and quality are different axes, and a machine can be very certain about something very flat.
I killed it in review, then went and rebuilt how much the system is allowed to do on its own. That rebuild is what this guide is actually about: three modes, the guardrails, the setup, and the situations where you should leave Autopilot switched off entirely.
The reason I care at all: I build social media software and I still don't want to do social media. The creative part is fine. The mechanical 80%, resizing one image five ways, scheduling, pasting a caption into five apps, again tomorrow, is what burns people out. 90% of content creators report burnout; over half blame the everywhere-every-day grind. Autopilot is for that 80%. The dead post is what taught me to fence off the rest. (Engineering side: building the Autopilot feature. The wider automate-versus-human question: social media automation guide.)
The three modes
There are three because people trust a machine with their name to very different degrees, and after the dead post I stopped assuming anyone should trust it much at first.
Individual Review holds every post for your approval. Each approve, edit, and reject also trains the voice model, so it doubles as the fastest way to teach the system what you sound like. New accounts belong here.
Batch Review generates a week at once and hands you the stack to approve, edit, or cut in one sitting; the survivors schedule themselves. Seven posts a week across three platforms is 21 separate scheduling actions, and this collapses them into a single half-hour. Most people land here and stay, which tracks with the finding that AI saves marketers 5+ hours a week on content work.
Full Autopilot runs the whole loop unattended and tunes itself to what performs. It also produced the dead post, so the only responsible advice is to reach it slowly, never to start there.
What the dead post changed
After that week I stopped trusting "it'll probably be fine" and built the parts that assume it won't. Three of them earn their keep every week.
Engagement-drop detection watches each account against its own baseline and can pause publishing when the numbers fall off a cliff, so one bad run doesn't quietly become ten. Image approval is mandatory even in Full Autopilot, because a flat caption is a flat caption but a bad image is a screenshot that outlives you. And a full audit trail logs everything generated, published, and edited, so nothing the system does is hidden from you. There is more underneath, trending-sound approval, repetition checks, the low-confidence hold from earlier, but those three are the ones I would refuse to ship without.
What none of them catch is confident-and-boring. That stays your job, and it is the whole reason to keep reviewing until the voice model is genuinely sharp.
When to leave it off
Some situations are wrong for automation, and I would be doing you a disservice to pretend otherwise. Below roughly 30 of your real posts, the voice model writes like a competent stranger instead of you, so don't automate before you've fed it that much. Don't put your primary, rent-paying account on it until you've watched it behave on a smaller one. And during a launch, a crisis, or any day where one wrong post is genuinely expensive, run it by hand. Automation is built for the ordinary days, not the ones that matter most.
Setting it up
This part is plain configuration, and the output is only ever as good as the attention you bring to it. The topic step in particular is where people hurry and regret it, so the first run deserves more than the fifteen minutes the wizard suggests.
- Accounts. Which connected accounts it runs. Instagram and TikTok on, LinkedIn manual, any mix.
- Brand Voice. The profile it writes in; the wizard makes you build one if you haven't. Aim for a quality score of at least 50 before turning it loose. (How Brand Voice works.)
- Goal. Engagement, reach, traffic, or balanced. It changes the writing: engagement gets conversational, reach gets shareable, traffic leans on the call to action.
- Topics. Up to 20, specific. "Marketing" is useless; "Instagram Reels for e-commerce brands" is workable. Set the off-limits list here too.
- Frequency. Posts per week within your plan (Free: 1 account, up to 7/week; Pro: 5 accounts; Agency: 20; Enterprise: 50 at up to 14/week each). Consistent publishers pull 3.5x more inbound traffic than the once-a-fortnight crowd, but choose a number you can actually hold.
- Mode. Individual, Batch, or Full. Switchable anytime.
- Posting windows. Days, time ranges, weekends or not. Optimal Times picks slots from your own history, tied to a 25-40% engagement lift over fixed scheduling.
- Formats. Single images, carousels, text, Reel and short-video captions, Stories.
- Muse (optional). Wire in Sydium Muse and set a trend ratio (default 30%) for how much rides what's trending now versus your planned topics.
What a few weeks does
Nothing here writes in a vacuum. Every approve, edit, and skip is a signal: this format works, you always trim the second sentence, you keep single images and reject carousels. Over a few weeks that feedback sharpens the voice, the topic mix, the formats, and the timing, until the system starts catching its own dead posts before you have to.
The usual arc: two weeks of Individual Review while you edit everything and train it, then Batch Review around week three as the edits taper, then Full Autopilot only on a secondary platform, and only once you're waving most posts through untouched. A lot of people simply stay on Batch Review because they like the weekly pass, and that is a completely fine ending. Begin with Individual Review, and let Autopilot earn each step before you give it the next.
Related free tools
Free, no signup, runs in your browser.
- Best Time to Post Calculator - Find the optimal posting times for each platform based on engagement research.