"Can AI do my social media?" Owners ask it when the company page has not posted since the summer, when the one person who looked after it has left, or when a vendor promises a month of posts in an afternoon. The honest answer is that a tool can draft, schedule, resize and suggest from a written voice and a calendar, and a person still decides what the company says, approves anything that names a customer, an employee, a price or a result, and holds the logins. This article is for the owner of a company with fifty to five hundred people and a LinkedIn page, a Facebook page or an Instagram account that someone is supposed to be looking after. It sets out what a tool can take over, what a person keeps, five checks before a tool touches the company's channels, and the mistakes that repeat. Where money is handled it is not accounting or tax advice, and what each platform allows is a question for its own rules and your advisor, not for this page. For the company's search presence, can AI do my SEO covers what a tool drafts and checks and what a person decides the company claims.
I was COO and then President of a public company that closed more than forty acquisitions, and I ran the integration on most of them. In almost every one, the acquired company's social pages were the first public thing we inherited: a few were well kept, most were a year stale, and one or two were still being posted to by someone who no longer worked there. I am a lawyer by training. Now I write fixed-fee AI plans for CEOs of private-equity-backed and owner-led mid-market companies through 013 Labs.
The short answer
For a company of this size, social media is a small publishing operation: a calendar, a voice, a handful of kinds of post, and a public inbox that strangers and customers write into. A tool is good at the operation. Given a written voice and the material you already have, it can draft a post about a finished project or a new hire, make the version each channel wants, resize the photo, queue it for the right morning, suggest an answer to a comment, and count what was seen. What it cannot do is decide what the company stands for, whether a claim is true, whether a photo may be used, or what to say on a bad day. Every post is a public statement by the company, and a comment answered badly is a public statement too. So the useful question is not "can a tool do our social media" but "which posts are routine enough to draft from a written voice, and who approves anything that carries the company's name".
What a tool can often take over
Drafting from your own material. A finished job, a new hire, a trade show, a plain answer to a question customers keep asking. A tool can turn a few lines of notes and a photo into a post in your voice, and offer two or three versions to choose from.
The version for each channel. The same news reads differently on LinkedIn, on Facebook and on Instagram. A tool can produce each version from one approved draft, so one approval covers all three.
The calendar and the queue. A tool can hold the plan for the month, keep the posting times steady, avoid holidays, and show the gaps. The habit is where most company pages fail, and a queue is what keeps it.
Resizing and captions. The sizes each channel wants, a caption, and the short description a screen reader uses. Routine work, done in a minute.
Suggesting answers to comments and messages. A question about opening hours, a thank-you, a request for a quote. A tool can draft the answer from your own information and put it in front of the person who will send it. Can AI respond to my Google reviews draws the same line for the public review, and Can AI answer customer emails for the inbox.
Counting. Which kinds of post get looked at, answered, or bring someone to the website. Can AI help promote my business covers where the social pages sit in the rest of the marketing.
Keeping the record. Who approved which post, in which version, when it went out, and which comments were answered by whom. That record matters the day someone asks who said that.
What a person must keep
What the company says. Positions, opinions, anything about a competitor, anything about the news. A tool asked for "something timely" will produce something timely, and the company will own it.
Every claim. The fastest, the largest, the only certified, a result for a customer, a safety record. A claim in a post is checked by the person who owns the fact before it is published. A tool does not know which award lapsed or which figure is out of date.
Everyone who appears. An employee in a photo, a customer in a story, a child at the company picnic. A person checks that there is written consent and that the person still wants to be there. A tool must never invent a quote, a testimonial or a result, and must never produce an image of a real person who did not agree to it.
Money. A boosted post, an advertising budget, a giveaway, a discount announced in public. These are decisions about cash and about promises, taken by a person, and never by a tool on a schedule. Not accounting or tax advice.
The logins. The company owns the accounts, in the company's name, with two named administrators, a recovery address and a phone number that belong to the company and not to whoever runs the page this year. Access ends the day someone leaves. No tool, and no employee, runs a company channel from a free personal account.
The bad day. An accident, a public complaint, a layoff, a death in the company. On that day the queue stops and a person decides whether the company says anything at all. Can AI answer my business phone draws the same line for the other public moment.
The data rule
A social media tool asks for the keys to your channels, and often for your photo library, your customer stories and your calendar of what is coming. Write one sentence before any tool gets that access: which tool is approved, under which terms, what it may see, what it may post without a person, and what never goes into it at all: a customer's name without consent, anything under a confidentiality agreement, anything about a deal that has not closed. What happens to your data explains what to ask a provider before a single photo is uploaded, and Do I need an AI policy for my small business shows how short the written rule can be.
Five checks before a tool touches the company's channels
First, write the voice file and the never-list. Three or four sentences on how the company sounds in public, and a short list of what no post may contain: a customer's name without consent, a price, a promise, a claim nobody has checked, a political opinion, a face that did not agree, a competitor's name.
Second, decide who approves, by kind of post. Routine posts drafted from approved material can be approved by one named person. Anything with a claim, a name, a face or money goes to the owner or a named manager. Write the two lists down; a tool can route by them, but the people must be named.
Third, put the accounts in the company's name. Two administrators, a recovery address and a phone that belong to the company, a written list of who has access, and a rule that access ends on the day someone leaves. Do this before any tool is connected; the tool inherits whatever access it is given.
Fourth, run draft-only mode for a month. Every post is drafted by the tool and approved by a person, free to change any word. At the end of the month, count how many drafts went out unchanged and how many were stopped. Those two counts tell you how far to trust the queue.
Fifth, write the stop rule. Who can pull every scheduled post within the hour, on which kinds of day, and who decides when the company speaks again. A tool can pause a queue in a second; the person who may press that button has to be named in advance.
Check yourself
Seven yes or no questions. Each no is a thing to fix before a tool touches a company channel.
- Does the company own every social account, with two named administrators and a recovery address that nobody has left with?
- Is there a written voice file a new hire could draft a post from?
- Is there a written list of what a post may never contain?
- Does every post that names a customer or shows a face rest on written consent you could find?
- Could you say who approved last week's posts, and in which version?
- Has every claim about a product or a result in this year's posts been checked by the person who owns the fact?
- Could someone stop every scheduled post within the hour on a bad day?
Mostly no? That is normal, and it is the real work. The AI readiness scorecard asks ten questions like these across the whole company.
Where it goes wrong
The page that reads like a machine. The same cheerful cadence, the same adjectives, a stock picture for every post. Customers and recruits notice, and the page tells them nobody at the company is there.
The claim nobody checked. "The only certified installer in the region", copied from an old website line, when the certification lapsed last spring. A competitor notices first.
The face without consent. A team photo that includes someone who left on bad terms, a customer's child at a site visit, a picture with another company's logo in the background.
The invented quote. Asked for "a customer testimonial" to go with a photo, a tool produced one. It sounded right, it was signed with a first name, and no such customer exists. What is hallucination explains why a tool does this and why the raw material matters more than the tool.
The post on the wrong day. A bright post about the team's safety record, scheduled ten days earlier, going out the morning after a serious fall on a job site. Nobody had a stop rule.
The account in a personal name. The person who ran the page for years left, and the page went with them: the recovery phone, the administrator seat and the tool's connection were all theirs.
Starting safely
Start with the calendar and with drafts from material you already have, where a person approves every post for a month. Then let the tool make the per-channel versions. Then let it suggest answers to comments, with a person sending each one. Keep the approver list and the never-list; what changes is how much of each post arrives already drafted. Picking your first AI project is the general version of this order, and How to train employees to use AI covers the person who will run the tool, who needs more than a login.
Picture it going wrong
A made-up example, to show the shape. A family-owned commercial roofing company, about a hundred and twenty people, four crews. The owner's nephew ran the social pages for years from his own accounts and left in the spring. In the autumn the owner turns on a tool, points it at the website and the photo folder, approves a month of posts in one sitting and lets the queue run. For three weeks it looks like the company has woken up: steady posts, good photos, a few new followers. Then a post celebrating "our team" goes out with a photo that includes a foreman who was let go in the summer, and he notices before the owner does. The next week a post calls the company "the only manufacturer-certified installer in the state", a line copied from an old page, when the certification had lapsed and a competitor now holds it. The week after, a bright post about the crews' safety record goes out early one morning, scheduled days earlier, a few hours after a serious fall on a job site. When the owner tries to pull it, nobody has the administrator login: it was the nephew's, and so was the recovery phone. Nothing was wrong with the tool. The accounts were in the wrong name, nobody had written what a post may never contain, nobody was approving, and nobody could stop the queue. The fix was the accounts moved into the company's name with two administrators, a voice file, a never-list, one approver for routine posts and the owner for anything with a face, a claim or money, a month of draft-only, and a stop rule with a name on it.
Mistakes that repeat
- Connecting a tool before the accounts are in the company's name.
- Approving a month of posts in one sitting, then not looking again.
- Publishing a claim from the website without asking whether it is still true.
- Posting a face without written consent, or a quote nobody said.
- Boosting a post, or announcing a discount, on a schedule nobody reviewed. Not accounting or tax advice.
- Letting a tool answer comments in public because the backlog was embarrassing.
Frequently asked questions
Can AI do my social media?
A tool can draft posts from your own material in a written voice, make the version each channel wants, keep the calendar, suggest answers to comments and count what was seen. A person decides what the company says, approves every claim, name, face and spend, and holds the logins.
Can AI run my company's LinkedIn page on its own?
Some tools will post without anyone looking. That is the setting to turn off. Draft-only mode, with a named approver and a queue a person can stop, gives you the steadiness without the public mistake.
Can I use AI to write social media posts for my business?
Yes, from material you give it and in a voice you wrote down. The failure is asking a tool to write from nothing: it will produce something plausible, and plausible is not the same as true about your company.
Can I use AI-generated images on my company's social media?
For illustrations, with care and with a person deciding. Never an image of a real person who did not agree, never a picture that implies a project or a product that does not exist, and never a photo that could be mistaken for a real site or a real customer.
Can AI answer comments and messages on our pages?
It can draft the answer from your own information and put it in front of the person who sends it. A question about price, a complaint, a safety matter or a job applicant always gets a person, and the harder conversations move off the public page.
Is it safe to give a social media tool our photos and customer stories?
Only under written terms that say what the tool may see and what it may do with it, through company accounts, never a free personal account. A customer story without consent, anything confidential and anything about an unfinished deal stay out of the tool altogether.
Where should I start?
Put the accounts in the company's name, write the voice file and the never-list, name the approvers, run the tool in draft-only mode for a month, and write the stop rule before the first post goes out.
Not sure where you stand? Take the 3-minute AI readiness scorecard.