- Four jobs, four tools. A chat assistant (ChatGPT, Gemini, Claude) for planning and captions; ChatGPT’s image generator or Canva for graphics; Canva or CapCut for vertical video; your own head for the voice.
- Never post a first draft. Cut the AI’s throat-clearing, add one real detail only you know, and check any number, claim or product fact yourself.
- Repurpose, don’t restart. One long thing (a video, a customer question, a blog post) becomes a week of posts.
- Label it and stay honest. Most large platforms now ask you to disclose realistic AI-generated or altered content. No fake reviews, no using a real person’s face or voice without permission.
AI for social media comes down to four concrete jobs: planning what to post, drafting captions and hashtags, making images, and cutting short vertical videos. A free chat assistant plus Canva or CapCut covers all four. What you keep for yourself is the voice, the real details, and the fact-checking. That split is the whole trick.
What AI is good at, and what you should keep
Before touching a tool, it helps to know where the line sits. AI is strong at structure, volume and variations. It is weak at anything that depends on having actually been there.
| Hand to AI | Keep for yourself |
|---|---|
| 30 post ideas from one theme | Which three are actually worth posting |
| First-draft captions, five hook variations | The sentence about what really happened |
| Hashtag shortlists, alt text, translations | Prices, stock, opening hours, any claim about your product |
| Background removal, resizing, subtitles | Faces and voices of real people (consent first) |
| Turning one video into ten captions | Replying to comments and DMs as a human |
Step 1: Plan a content calendar in one sitting
Open any chat assistant and give it three things: who you are, who you are talking to, and how often you post. Vague input produces vague output, so be specific.
Something like: “I run a two-chair barbershop in Leeds. My customers are men aged 25 to 50 who book last minute. I can post four times a week on Instagram. Give me four content buckets, then a four-week calendar with one post per slot. For each, say the format (photo, carousel, 20-second vertical video) and the point of the post. No emojis in the plan.”
You will get something usable in about ten seconds. Now do the part that matters: delete half of it. Anything you cannot photograph or film this week is a fantasy. Anything that sounds like a marketing textbook will read like one.
Keep the plan somewhere boring and visible: a spreadsheet, a notes app, or Canva’s content planner if you already use Canva. The tool matters less than the habit of knowing what Thursday’s post is before Thursday.
Step 2: Captions and hashtags that don’t sound like a robot
AI captions have a smell: three-word opener, em dash, rhetorical question, emoji, “the best part?” Readers clock it instantly. You fix it in the prompt and in the edit.
In the prompt, ban the tics and feed it your own words. Paste two or three captions you wrote yourself and say: “Match this voice. Short sentences. No emojis, no rhetorical questions, no words like ‘elevate’ or ‘unlock’. Write three caption options under 60 words for a photo of a fade haircut we did this morning. The point is that walk-ins are possible before 11am.”
Then edit. Two rules cover most of it:
- Cut the first line. AI almost always warms up before it says anything. The second sentence is usually your real opener.
- Add one thing only you know. The customer who came in on his lunch break. The clippers that died mid-cut. That single detail is what makes the post yours, and no model can guess it.
For hashtags, treat the AI’s list as a starting point, not gospel. It does not know today’s volumes on any platform. Ask for a mix of broad and niche tags, then check a handful in the app’s own search to see whether real posts are using them. If you want a wider look at which writing assistants handle brand voice best, our comparison of AI writing tools goes deeper.
Modern assistants can hold a lot of material at once. A token is roughly a chunk of a word, and a context window is how much text a model can keep in mind in one conversation. Several current models are listed in Convly’s models database with context windows around one million tokens, which in practice means you can paste a year of past captions and ask for a voice guide. App subscriptions have their own separate usage limits, so check the vendor’s plan page for what your tier allows.
Step 3: Images
You have two realistic routes, and most people end up using both.
Generate from scratch. ChatGPT’s built-in image generator handles text-on-image far better than earlier tools did, which makes it genuinely useful for quote cards, simple announcements and illustrative backgrounds. Our walkthrough of how to use the ChatGPT image generator (Images 2.5) covers prompt structure and the aspect ratios that fit feeds and stories. If you want alternatives, see our roundup of free AI image generators.
Improve what you shot. This is the underrated option. A real photo of your real product, cleaned up, beats a synthetic image almost every time. Canva’s Magic Studio tools cover background removal, object erasing and resizing; phone editors do much of the same. Our guide to AI photo editing apps compares the main ones. Feature names and availability inside Canva change regularly, so check Canva’s own pricing page for what sits in the free tier versus Canva Pro today.
One caution: generated images still get hands, text and logos wrong. Look closely before posting, especially at anything a customer might read as a real photo of your premises or stock.
Step 4: Short vertical videos
For most small accounts, the fastest path is not a text-to-video model at all. It is your own footage plus AI editing.
Film 20 seconds on your phone. Drop it into Canva or CapCut. Let the app auto-generate captions (huge, because most people watch muted), trim the dead air, add a hook line in the first second, export vertical. That loop takes minutes and the result looks like a person made it, because one did.
Generated video is worth knowing about for B-roll, concept shots and product scenes you cannot film. Text-to-video is sold by the second through developer APIs, and the figures below are the API list prices recorded in Convly’s models database — not app subscription prices. Always confirm against the vendor’s own pricing page before budgeting.
| Model | Maker | Listed API price (Convly models database) |
|---|---|---|
| Wan 2.5 | Alibaba | from $0.05 per second |
| Veo 3.1 | from $0.05 per second | |
| Kling 2.5 Turbo Pro | Kuaishou | from $0.07 per second |
| Seedance 2.5 | ByteDance | from $0.097 per second |
Two things to note. First, most everyday users never touch an API — you reach these models through consumer apps or resellers, whose pricing is different again. Second, this space churns: OpenAI’s Sora 2 and Sora 2 Pro were withdrawn from the Videos API in September 2026 with no replacement, so treat any tutorial that assumes they are available as out of date. Our current picks live in the best AI video generators, and if you want to animate a still photo instead, see turning an image into a video with AI.
If you are heading towards ad-style content with a presenter, read how to make UGC ads with AI without breaking the rules first. The rules there are stricter than people assume.
Step 5: Repurpose one thing into a week of posts
This is where AI saves the most time and it is the step most people skip. Take one substantial thing you already made — a five-minute video, a customer email, a workshop you ran — and ask an assistant to break it apart.
Paste the transcript or text and ask for: five standalone captions, each on one idea from the source; three quote cards; one carousel outline with a line per slide; and a list of the questions this raises that you could answer in a follow-up post. Then pick the ones you actually agree with.
The discipline is one idea per post. AI will happily cram four into a caption, and each one gets weaker.
Phone or laptop: what changes
iPhone and Android
Fine for the whole loop. ChatGPT, Gemini and Canva all have phone apps, CapCut is phone-first, and you are already holding the camera. The weak spot is anything involving a lot of typing or comparing drafts side by side. Our list of the best AI apps for iPhone covers what is worth installing.
Windows and macOS
Better for planning, bulk work and editing longer video. ChatGPT and Canva both offer desktop apps for Windows and macOS alongside the browser version; CapCut has desktop versions too, though which features appear on desktop versus mobile varies by release. Everything also runs in a browser, so a desktop install is a convenience, not a requirement.
Linux and ChromeOS
No native desktop apps for most of these, and you do not need them. Canva, CapCut and the main chat assistants all run in Chrome or Firefox with no loss of core function. Video export can be slower on low-powered hardware because the work happens in the browser and on the provider’s servers.
What it costs
You can run a small account on free tiers indefinitely. Free plans of ChatGPT and Gemini will write captions and plans; Canva’s free tier covers plenty of design; CapCut’s free tier handles basic editing. What paid plans buy is usually higher usage allowances, faster or more capable models, and extra creative tools.
Because plan names, prices and limits change often, check the source rather than a blog: OpenAI’s ChatGPT pricing page and Google’s AI plans page are the primary sources. For a plain-English breakdown of what each tier actually gives you, see our guides to ChatGPT Free vs Go vs Plus vs Pro and whether Gemini is free. Developer API rates, like those in the table above and published on OpenAI’s API pricing page, are a separate billing system entirely and tell you nothing about what a subscription includes.
Labelling, honesty and the lines not to cross
Most major platforms now expect realistic AI-generated or significantly altered content to be disclosed, and several add a label automatically when they detect signs of generation. The exact wording and the toggle’s location differ by platform and change over time, so check the current help pages for Instagram, TikTok, YouTube and LinkedIn rather than trusting a screenshot from last year. Disclosing costs you nothing; getting caught not disclosing costs reach and trust.
Some lines are not about platform policy at all:
- No invented reviews or testimonials. A quote from a customer who does not exist is a false claim about your business, whoever typed it.
- No one else’s face or voice without written permission. That includes employees, customers and anyone recognisable in the background. Voice cloning is easy now, which is exactly why consent matters — see our piece on deepfakes.
- Check every factual claim. Assistants state wrong things fluently. Prices, dates, ingredients, legal and medical points: verify before posting.
- Watch what you paste. Customer names, addresses and order details typed into a chatbot may be retained or reviewed. Strip identifying details, or use whichever privacy settings your plan offers.
- Follow the rules where you are. Schools, employers and some client contracts restrict AI-assisted work. If disclosure is required, disclose.
Frequently asked questions
Can AI just run my social media for me?
No, and accounts that try are easy to spot. AI can carry the planning, the first drafts and the repetitive editing, which is realistically half the work. It cannot supply the experiences that make posts worth reading, and it should not be answering your comments or DMs as if it were you.
Will followers be able to tell my captions were AI-written?
If you post the raw draft, often yes — the rhythm and stock phrasing give it away. If you cut the opening line, replace the generic claims with specifics from your day, and read it aloud once, it reads as yours because the substance is. Do not chase tools that promise to fool AI detectors; write something real instead.
Do I have to label AI images and videos?
Assume yes for anything photorealistic or that could be mistaken for a real recording. Major platforms ask for disclosure and may label content automatically, and requirements keep tightening. Obvious illustrations, cartoons and abstract graphics are treated more leniently, but check the platform’s current guidance rather than guessing.
What is the cheapest setup that actually works?
One free chat assistant, one free design tool, and your phone camera. That covers planning, captions, graphics, editing and captions-on-video. Pay for something only when you hit a specific wall, like running out of usage on a free tier or needing a feature you now use weekly. Our list of 25 free AI tools is a good place to start.
Can I use an AI voice for narration?
Yes, if the voice is a licensed synthetic one or a clone of your own voice, and you follow the platform’s disclosure rules. Cloning someone else’s voice without their permission is off limits regardless of what a tool technically allows. Our guide to making an AI voice-over covers the practical side.
Which assistant should I pick for this?
Any of the main ones will write a decent caption; the difference shows up in tone control and how well they hold your brand voice across a long session. Try the same prompt in two and keep whichever sounds less like a press release. Our Claude vs ChatGPT comparison and list of ChatGPT alternatives lay out the trade-offs.
The accounts that use AI well look no different from accounts run by an organised, well-rested human. That is the goal. Let the machine handle the blank page and the busywork, and spend the time you save on the things it cannot fake.