On this page 9
  1. The short answer: what AI should and shouldn't do for your social media
  2. Before you start: the one-time setup (about an hour)
  3. The weekly workflow, step by step
  4. Copy-paste prompts for each platform
  5. How to keep AI-written posts authentic
  6. Which AI tools to use (and what they cost)
  7. What to measure (and what to ignore)
  8. Common mistakes to avoid
  9. Your weekly AI social media checklist

The short answer

Use AI as a drafting assistant, not as the person who runs your account. Let it brainstorm, draft and adapt posts for each platform, working from a voice brief and fact sheet you write once. Keep the facts, offers, real photos and customer replies in your own hands. Once you're set up, aim for about 2–3 hours of focused work a week, plus a few minutes a day for replies.

The short answer: what AI should and shouldn't do for your social media

Good jobs for AI

  • Brainstorming post ideas from what happened in your business this week
  • Writing first drafts and two or three variations to choose from
  • Repurposing one idea into a Facebook post, an Instagram caption, a LinkedIn post and a video script
  • Adjusting length and tone for each platform
  • Drafting alt text and suggesting shot lists
  • Spotting patterns in metrics you paste in

Keep these human

  • Strategy: who you're talking to and what you want them to do
  • Facts, prices, dates and offers
  • Photos of your real work, team and premises
  • Replies to customers, and especially complaints
  • The final read before anything goes live

The core loop and how long it takes

The core loop: plan the week, batch-draft with prompts, edit for voice and accuracy, add visuals, schedule, engage, review.

Time: around 2–3 hours a week once set up. This is a planning target to box your time, not a measured result. Your first few weeks will take longer.

Before you start: the one-time setup (about an hour)

AI writes generic posts when it gets generic input. The setup below gives it the specific details it can't guess. You do it once, then update it as you learn.

Your one-page brand voice brief (template)

Copy this into a document and fill in the blanks:

  • Business: [name], [what you do], [town or area you serve]
  • Who we serve: [your main customers, e.g. busy parents within 5 miles, local trades, startup founders]
  • What we sell: [top 3–5 products or services]
  • Tone words (3–5): [e.g. warm, direct, a bit cheeky, never pushy]
  • How we talk: [e.g. first names, short sentences, no jargon, the odd joke]
  • Words and phrases we never use: [e.g. elevate, game-changer, unlock, delve, 'look no further', 'in today's fast-paced world']
  • Emoji and hashtags: [e.g. one emoji at most, up to 3 hashtags, always include our town]
  • Example posts we liked (2–3): [paste them, yours or others']
  • Local details: [street, landmarks, local events, suppliers, team names you're happy to use]
  • Things we never claim: [e.g. 'best in town', guarantees, health claims]

Let AI interview you to build it

If a blank template feels like homework, let AI ask the questions:

You are helping me write a one-page brand voice brief for my small business so AI drafts sound like us.
Interview me with 10 questions, one at a time. Wait for my answer before asking the next one.
Cover: who our customers are, what we sell, how we want to sound, words we'd never use, posts we've liked, local details and what makes us different.
When you have all 10 answers, write the brief as a bulleted list under these headings: Who we serve, What we sell, Tone words, How we talk, Words we never use, Example posts, Local details, Things we never claim.
Use my own words where you can. Don't add anything I didn't say.

Pick 3–5 content pillars

Pillars are the recurring themes you post about, so you never start from a blank page. For example:

  • Behind the scenes: how you make, fix or prepare things
  • Tips and education: answers to questions customers ask
  • Customer stories: real ones, used with permission
  • Offers and news: new stock, services, dates
  • Community and local: events, suppliers, neighbours

Choose 1–2 platforms

Go where your customers already are. Two platforms done well beat five done badly.

Quick decision

  • You run a café or salonyou may get most value from Instagram, Facebook and Google Business Profile.
  • You're a consultant or B2B freelanceryou may only need LinkedIn.

Write a fact sheet

List your prices, opening hours, services, booking links, current offers with end dates, and anything you're not allowed to say.

The weekly workflow, step by step

Here's the whole routine on one page. The days are suggestions, so move them to fit your week.

StepSuggested dayTime boxWhat AI doesWhat you do
1. PlanMonday20–30 minTurns your notes into a 3–5 post planCut what isn't true or relevant; add what only you know
2. Batch-draftMonday or Tuesday30–45 minWrites 2–3 variations per post and adapts each for the platformFeed it raw material; pick the best version
3. EditTuesday20–30 minCan flag stock phrases if askedCheck every fact; run the de-AI pass; read aloud
4. VisualsTuesday20–40 minSuggests shot lists, graphics, alt text draftsTake real photos; check images and alt text
5. ScheduleTuesday10–15 minOptional caption tweaks inside your schedulerQueue posts at times from your own insights
6. EngageDailyAbout 10 minSuggests reply drafts for common questionsSend and personalise every reply; log questions
7. ReviewFriday15 minSpots patterns in metrics you paste inSanity-check its conclusions; decide what to change

Step 1: Plan the week's content (20–30 minutes)

Jot down this week's raw inputs: upcoming events, seasonal moments, new stock or services, and questions customers asked. Then paste them in with your voice brief and ask:

Using my voice brief and content pillars above, plan [3–5] posts for this week on [platforms].
This week's notes: [events, new stock, customer questions, anything that happened].
Give me a table with columns: day, platform, pillar, hook (one line), format (photo, carousel, video, text), call to action.
Only use facts from my notes and fact sheet.

That table is your content calendar for the week. Now edit it. Cut any idea that isn't true, isn't relevant or that you wouldn't enjoy posting.

Add the one thing only you know about, like the supplier who turned up early or the question three customers asked.

Step 2: Batch-draft posts with prompts (30–45 minutes)

Draft everything in one sitting. Start the chat in ChatGPT, Claude or whichever assistant you use by pasting your voice brief and fact sheet.

Many assistants also let you save them as custom instructions or in a project, so you don't repeat yourself. Feature names and availability vary by tool and plan.

Use the platform prompts in the next section. Ask for 2–3 variations per post and don't settle for the first draft. Repurpose as you go: one strong idea can become a LinkedIn post, an Instagram caption and a short video script.

Step 3: Edit for voice, accuracy and honesty (20–30 minutes)

This is the step that stops AI posts going wrong.

  1. Check the facts. Check every fact, price, date and claim against your fact sheet.
  2. Run the de-AI checklist. You'll find it in the authenticity section below.
  3. Read each post aloud. If you wouldn't say it to a customer at the counter, rewrite it.

Step 4: Add visuals (20–40 minutes)

Real photos of your work, team and premises usually build more trust than a generated image. Use AI to plan them: ask for a shot list for the week or ideas for caption overlays.

AI image tools and the template features in design apps are useful for graphics, quote cards and backgrounds.

For accessibility, ask AI to draft alt text from your description of the photo, then check it describes what's actually in the image. Add it wherever the platform lets you.

Step 5: Schedule (10–15 minutes)

Queue the week's posts so they go out without you. Free native options include Meta Business Suite for Facebook and Instagram, and LinkedIn's built-in post scheduling. Third-party schedulers can handle several platforms in one place. Features and free allowances change, so check what's currently on offer.

Many schedulers now include AI caption helpers. They're handy for quick tweaks, but they don't know your voice unless you tell them, so your brief still matters.

Pick posting times from your own audience data in platform insights, not from generic 'best time to post' lists.

Step 6: Engage daily (10 minutes a day)

Replies and conversations stay human. AI can suggest drafts for common questions, like opening hours or how to book, but you send them and personalise them.

For public reviews, our human-approved workflow for responding to customer reviews with AI shows where drafting help ends.

Keep a running note of questions people ask. They're next week's content ideas.

Step 7: Review what worked (15 minutes, weekly)

Paste or export your post metrics and ask AI to spot patterns, such as which formats or hooks got the most saves or link clicks. Then sanity-check its conclusions, because it can over-read small numbers.

Keep one or two things that worked, drop one that didn't and feed that into next week's plan. The measurement section below covers which numbers to look at.

Copy-paste prompts for each platform

Each prompt uses [brackets] for the parts you fill in. Start with the master wrapper, then add the platform block underneath it.

Master prompt wrapper

ROLE: You are a social media copywriter for my small business. You write first drafts for me to edit.

VOICE BRIEF:
[paste your voice brief]

FACT SHEET (only use facts from here):
[paste your fact sheet]

TASK: Write [number] variations of a [platform] post about [topic] for the [pillar] pillar.
RAW MATERIAL: [voice note transcript / customer question / what happened today]

CONSTRAINTS:
- Length: [short / medium]. Plain English, mostly short sentences.
- Emoji: [none / at most one]
- Hashtags: [none / up to 3, specific and relevant]
- Call to action: [book / call / visit / reply / click the link]
- Do not invent prices, dates, statistics, reviews or customer stories.
- Avoid every phrase on my banned list.

If anything you need is missing or unclear, ask me before writing.

Why it's built this way: each section fills a gap the AI would otherwise fill by guessing. The final line makes it ask instead of invent.

Facebook

Platform: Facebook. Write like a friendly local business talking to neighbours.
Open with something specific from [this week / the raw material].
Mention [local detail] if it fits naturally.
End with one easy question that invites comments, then one clear call to action: [action].
Keep it to three short paragraphs or fewer. Give me 3 variations with different opening lines.

Why: a local business's Facebook page often works like a community noticeboard. A simple question gives people an easy reason to reply.

Instagram caption

Platform: Instagram caption for [describe the photo or video].
First line: a hook of about 10 words or fewer that makes someone stop scrolling. No clickbait.
Then 2–4 short lines with a line break between each.
End with the call to action: [action].
Add [3–5] hashtags specific to [your niche] and [your town]. Put them on a separate final line.
Give me 3 variations.

Why: captions are often cut short in the feed, so the first line has to work on its own.

Instagram carousel

Platform: Instagram carousel about [topic].
Outline [5–8] slides. For each slide give: on-screen text (one short sentence) and a note on the image or background to use.
Slide 1 is the hook. The last slide is the call to action: [action].
Then write a short caption that adds one detail not on the slides.

Why: carousels suit tips and step-by-steps, and asking for image notes gives you a ready brief for your design tool.

LinkedIn

Platform: LinkedIn, written in the first person as [your name, your role].
Turn this into a short lesson or story: [what happened / what a client asked / a mistake you learned from].
Structure: a one-line opening that states the moment or the lesson; 3–5 short paragraphs with one concrete detail (a number, a place, something I said); one practical takeaway; a genuine question at the end that I'd actually want answers to.
No more than 3 hashtags. No 'I'm thrilled to announce', no 'Agree?', no one-word lines for drama.
Give me 2 variations: one more personal, one more practical.

Why: LinkedIn readers tend to respond to a person's real experience. Banning the usual clichés up front saves editing time.

Short video script (Reels, TikTok, YouTube Shorts)

Platform: short vertical video for [Instagram Reels / TikTok / YouTube Shorts], [20–45] seconds.
Topic: [topic]. I'll film it on a phone at [location].
Give me:
1. A hook I can say or show in the first 3 seconds.
2. A script in my voice, with rough timings.
3. A shot list using only things I actually have: [list props, people, places].
4. Very short on-screen text for each shot.
5. A caption and call to action: [action].

Why: listing what you actually have stops the AI from planning shots you can't film.

X and Threads

Platform: [X / Threads].
Write one short standalone post about [topic]. Plain and direct, no hashtags or at most one.
Then write a thread version of 3–5 posts: the first must make sense on its own, each following post adds one point, and the last has the call to action: [action].

Why: the standalone version is quick to post. The thread version is there for days when you have more to say.

Google Business Profile post

Platform: Google Business Profile post. Type: [offer / update / event].
Details from my fact sheet: [what it is, dates, terms].
Write 2–4 plain sentences: what it is, who it's for, when it runs and how to get it.
Mention [town or area] once, naturally. No hashtags.
Suggest which call-to-action button fits best (e.g. book, call, learn more).

Why: for local businesses, your Google profile can be one of the last things people see before they call or visit. Clear and factual beats clever here.

Repurposing prompt

Here is my source material: [paste a blog post / FAQ answer / voice note transcript].
Pull out the 3 most useful ideas for my customers. For the strongest one, write:
- a Facebook post
- an Instagram caption
- a LinkedIn post in my first person
- a 30-second video script with a 3-second hook
- a Google Business Profile update (only if relevant)
Adapt each one to its platform rather than resizing the same text.
Use only facts from the source and my fact sheet. At the end, list anything you weren't sure about.

Why: one good piece of thinking can fill most of your week. The 'not sure' list shows you exactly what to check.

How to keep AI-written posts authentic

The aim is AI as a drafting assistant, not a ghost that runs your account. These habits keep posts sounding like you.

Feed it raw material only you have

Specific input gives specific output. Before you prompt, grab one of these:

  • A 60-second voice note about today's job, transcribed
  • A question a customer asked, in their words
  • A description of a photo you just took
  • A small detail: the supplier, the weather, the mistake, the fix

The de-AI editing checklist

Run every draft through this before it goes anywhere:

  • Cut stock phrases and buzzwords (elevate, unlock, game-changer, 'look no further', 'we're excited to share')
  • Cut extra emojis and exclamation marks. One of each is usually plenty.
  • Replace every vague claim with a specific detail ('high quality' becomes 'pears from the farm shop up the road')
  • Keep at least one sentence that sounds personal, something only you would say
  • Vary sentence length. AI drafts often fall into a steady, even rhythm.
  • Remove anything you'd never say out loud
  • Check facts, prices and dates against your fact sheet
  • Check hashtags are real, relevant and not borrowed from an unrelated trend
  • Read it aloud once, start to finish

Worked example: the rough note and the first AI draft

Rough input (the owner's voice note, transcribed):

New thing this week, pear and ginger buns. Fri and Sat only, £2.40. Sam's been testing them for three weeks, finally stopped frowning at them on Tuesday. Pears from the farm shop up the road. Test batches went by 11. Happy to put one aside if people comment.

An unedited first AI draft:

🍐✨ Get ready to elevate your weekend! ✨🍐 Introducing our delicious new Pear & Ginger Buns, the perfect blend of sweet and spice that will tantalise your taste buds! Made with love and the finest ingredients. Don't miss out, visit us today! #Bakery #FoodLover #Delicious #Weekend #Yum

What's wrong with it: stock phrases, too many emojis, generic hashtags, and no price, days or story. 'Visit us today' is also wrong if the post goes out midweek and the buns are only on sale Friday and Saturday.

Worked example: the human-edited final

New this weekend: pear and ginger buns.

Sam has been tweaking the recipe for three weeks and finally stopped frowning at them on Tuesday. The pears come from the farm shop up the road.

£2.40, Friday and Saturday only. The test batches were gone by 11, so come early, or comment below and we'll put one aside.

#[YourTown]Bakery

Every detail in the final version came from the owner's note. The AI draft dropped most of them, and the edit put them back.

The line about Sam is the kind of detail only you have, and it does more for the post than any rewrite.

Never invent testimonials, reviews or results

Use real customer stories and quotes only, with permission. Don't let AI write a 'what our customers say' post from nothing.

Disclosure and platform labels

There's generally no blanket rule that AI-assisted text must be labelled. Realistic AI-generated or heavily altered images, video and audio are different.

At the time of writing, Meta, TikTok and YouTube all have policies that label this kind of content or ask creators to disclose it. These policies change, so check the current version in each platform's help centre, including the Meta Business Help Center, before you post AI visuals.

Which AI tools to use (and what they cost)

You don't need an all-in-one platform to start. Here are the main types of tool and where each fits.

Tool typeExamplesBest forLimitsCost model
General AI assistantsChatGPT, Claude, Google Gemini, Microsoft CopilotPlanning, drafting, variations, repurposing, reading pasted metricsCan make up facts; generic without your brief; free tiers have usage limits; doesn't post for youFree tier plus paid subscriptions
AI inside schedulersCaption helpers in native and third-party schedulersQuick tweaks where you already scheduleLess room for your full brief; quality and features vary by planSometimes free, often on paid plans
AI in design toolsCanva, Adobe ExpressGraphics, templates, resizing, backgrounds, quote cardsGenerated images can look generic or misleading; check usage termsFree versions; paid plans unlock more AI features or credits
Video and caption toolsBuilt-in editors in Instagram, TikTok and YouTube; dedicated video appsAuto captions, trimming, turning scripts into editsAuto captions need proofreading; AI voices and avatars can feel fake and may need labelsFree basics; subscriptions or credits for more

A sensible starting point

Start with one general assistant plus the native or free scheduler for your platforms. Run the full workflow for a few weeks before paying for anything.

If you're choosing between assistants, our guide on how to choose an AI writing assistant for your business covers the criteria and includes a 5-day trial plan. One week of social posts makes a good test.

A note on data

Check each tool's data and training settings, and turn off training on your chats if the option exists and you'd prefer it.

What to measure (and what to ignore)

The real question is whether social media brings in business for an amount of time you can sustain. Likes alone won't tell you that.

MetricWhat it tells youWhere to find itHow often
Link clicks / website clicksWhether posts send people to your site or booking pagePlatform insights; Google Analytics with UTM linksWeekly
Enquiries (messages, calls, forms)Whether posts start real conversationsYour inbox, call log, Google Business Profile performanceWeekly
Bookings or sales from socialWhat you actually earnedUTM conversions, promo codes, asking 'how did you hear about us?'Monthly
Saves and sharesContent people valued enough to keep or pass onPost insightsWeekly
Comments and repliesReal conversation, not just a tapPost insights and notificationsWeekly
Engagement rate per postHow each post did relative to how many people saw itPlatform insights, or work it out yourselfMonthly
Reach and follower growthContext: how many people you're reachingPlatform insightsMonthly
Hours spent per weekWhether AI is actually saving you timeYour own notes or a timerWeekly

Plain-English definitions

  • Engagement rate: interactions (likes, comments, saves, shares) divided by the number of people who saw the post, times 100. It lets you compare a post seen by 200 people with one seen by 2,000. Platforms and tools calculate it slightly differently, so pick one method and stick to it.
  • Reach vs impressions: reach is roughly how many accounts saw a post. Impressions (some platforms now say views) count how many times it was shown, including repeats. Labels change, so check each platform's definitions.
  • UTM parameters: short tags added to a link so Google Analytics can show which post sent a visitor. Google's help page on collecting campaign data with custom URLs explains how to build them.

Free tools and a monthly review

Free tools are enough: Meta Business Suite insights, LinkedIn page analytics, Google Business Profile performance and Google Analytics.

Common mistakes to avoid

  • Posting AI drafts without editing. Unedited drafts tend to sound generic, and regular customers notice when you stop sounding like yourself.
  • Letting AI make up facts. Unchecked prices, statistics, dates or hashtags can mislead customers and cost you trust.
  • Covering too many platforms. Spreading thin leads to rushed, generic posts everywhere.
  • One caption everywhere. Each platform has its own style. Adapt, don't copy-paste.
  • Chasing likes instead of enquiries. A post with ten likes and two bookings beats one with a hundred likes and none.
  • Automating customer replies. Use AI to draft, never to send, and never for complaints.
  • Misleading AI images. Don't show fake products, fake customers or results you didn't deliver.

Your weekly AI social media checklist

Once a week

  • Plan 3–5 posts from this week's events and customer questions
  • Batch-draft with the master prompt and platform templates; get 2–3 variations each
  • Fact-check every price, date and claim against your fact sheet
  • Run the de-AI checklist and read each post aloud
  • Add real photos where you can; check AI graphics and alt text
  • Schedule posts at times taken from your own insights
  • Spend 15 minutes reviewing what worked and note one change for next week

Every day

  • Spend about 10 minutes replying personally to comments and messages
  • Log new customer questions as content ideas

Once a month

  • Compare clicks, enquiries, sales and hours spent with last month
  • Change one variable and keep the rest the same

Frequently asked questions

Can I use ChatGPT or other AI tools to write my business's social media posts?

Yes, as a drafting assistant. Give it your brand voice brief, your fact sheet and some specific raw material, such as a customer question or what happened today. Ask for two or three variations, then edit and fact-check before posting. Keep strategy, offers, real photos and replies to customers human.

Do I have to disclose that a social media post was written with AI?

There's generally no blanket requirement to label AI-assisted text. Realistic AI-generated or altered images, video and audio are different: platforms such as Meta, TikTok and YouTube have policies that label this content or ask creators to disclose it. Advertising and endorsement rules, like the FTC's guidance in the US, apply however a post was written. These rules change, so check the current platform and regulator guidance before publishing.

How many posts a week should a small business create with AI?

There's no universal number. Choose a cadence you can sustain on one or two platforms, such as a few good posts a week, and stick to it. Then adjust based on your own enquiries, clicks and the time you have. Consistency and quality matter more than volume.

Will social media platforms penalise AI-generated posts?

No one outside the platforms can say exactly how their algorithms treat AI text. Broadly, platforms reward content people engage with. Generic, low-effort posts tend to do poorly whoever writes them, so original posts with specific details are the safer bet. Labelling rules for realistic AI images and video are a separate issue to check in each platform's help centre.

What's the best free way to start using AI for social media?

Use the free tier of a general AI assistant for planning and drafting. Pair it with your platform's native scheduler, such as Meta Business Suite for Facebook and Instagram, and the platforms' built-in insights for measurement. Free tiers and features change, so check each vendor's current page.

How do I stop AI posts sounding generic?

Give it specific raw material only you have, such as a voice note, a customer's exact question or details of today's job. Use a voice brief that includes a banned phrases list. Then run a de-AI edit: cut stock phrases and extra emojis, add one concrete detail and one personal line, and read the post aloud before publishing.

Can AI create images for my social media posts?

Yes, for graphics, backgrounds, quote cards and concept images. Use real photos for your products, team, premises and results. Never use AI images that misrepresent what you sell or the results customers can expect. Check each platform's rules on labelling realistic AI-generated imagery.

How do I know if using AI for social media is actually working?

Track business outcomes first: link clicks, enquiries, and bookings or sales from social (use UTM links, promo codes or ask customers how they found you). Add engagement quality, like saves, shares and comments, and the hours you spend each week. Compare month to month and change one thing at a time.