Plan one 90-minute session around a shot list of about 40 stills and 12 short vertical clips, log each shot with the facts only you know, and have AI draft captions and reel ideas from that log in your voice. Schedule four weeks in Buffer or Later at a rhythm you can keep, and approve every post yourself.
Coffee shop social media is easiest when you work backwards from the calendar, not forwards from the camera. A shoot with no plan produces the same eight good-looking lattes forty times, and then the month runs dry in week two. The shot list below is built from the posting rhythm, so every frame already has a slot, and the AI has facts to work from instead of a mood.
Work out how many pieces the month needs
Pick the rhythm first. Say it is three feed posts, two reels and four stories a week for four weeks. That is 12 feed posts, 8 reels and 16 stories, 36 pieces in all. Nine of the feed posts are single images and three are carousels of four images each, so the feed alone needs 21 stills. The reels need about 12 clips, because most reels use one to three clips and a good clip can appear in two of them with different edits. The stories can borrow the 10 or so stills left over, plus the odd clip.
That makes about 31 usable stills and 12 clips. Shoot 40 stills, planning to lose about a quarter of them to blur, poor framing and duplicates. If your rhythm is lighter or heavier, change the first sentence and redo the sum; the shot list scales with it.
The shot list: forty stills and a dozen clips in 90 minutes
Group the shots by subject, not by the order they come to mind, and count them before you pick up the phone. This list adds up to exactly 40 stills and 12 clips:
| Group | Stills | Clips | What to capture |
|---|---|---|---|
| Drinks on the counter | 12 | 2 | Six drinks, each from above and at 45 degrees; clips of a pour and of milk swirling in the jug |
| The craft at the machine | 4 | 4 | Espresso running, tamping, steaming, pouring; hands only |
| Pastry case and counter | 8 | 2 | The tray from three angles, close-ups of texture, a hand taking one; a clip revealing the case |
| Space and light | 6 | 1 | Window seat, counter, door; empty of customers; a clip of the sign being flipped |
| Beans and kit | 4 | 1 | Grinder, scoop, bag, a clip of beans dropping into the hopper |
| People | 4 | 2 | A barista at the machine from behind or hands only, with their agreement |
| Menu and small details | 2 | 0 | Menu board with prices you have checked; a line-up of takeaway cups |
Run the 90 minutes in blocks. Spend the first 15 on setup: wipe every surface, pick one cup and saucer style, check the background for price tags, staff notices and personal items, and look at where the light falls. Use the next 30 for the machine clips while it is warm. Give the drinks 25 minutes, with the six drinks made in a row, then 15 for pastries and the room, and the last five for the menu board and any reshoots. Shoot at the quietest time you have, hold each clip for eight to twelve seconds and record a little more than you need, and shoot vertically, since most of this will end up in a vertical frame.
Photographing staff, and any customer who might be recognised, needs their agreement. Keep customers out of the frame unless you have asked. For an honest look at what works when you consider generating images instead of shooting real ones, the cafe test of AI-generated marketing photos reports what looked believable and what did not.
Log every shot before the AI writes anything
The log is the step that stops generic captions. An assistant can see a flat white; it cannot know that the milk is whole, that the tulip took two goes or that the croissants arrive at seven and are usually gone by two. Within an hour of the shoot, rename the files and fill one row per usable shot with the facts only you know, and with the people in it. Here are five rows:
| File | What it shows | Facts the caption may use | People |
|---|---|---|---|
| flatwhite-top-01 | Flat white, tulip art, on a saucer, from above | Whole milk, double shot, second attempt at the tulip | None |
| croissant-tray-03 | Tray of six butter croissants on the pass | Delivered at 7 am; usually gone by 2 pm | None |
| steam-hands-02 | Close-up of hands steaming milk | Jug and machine only | Barista's hands, agreed |
| window-seat-01 | Window seat in morning light | Open from 7 am | Empty |
| iced-vanilla-02 | Iced vanilla latte with condensation | House blend, three pumps of vanilla | None |
If typing every row is slow, let a vision-capable assistant draft the first two columns from the photos and fill the facts column yourself. Upload ten photos at a time and keep the instruction narrow.
These are photos from my coffee shop. For each one, in a table, give the file name and a plain
description of only what is visible: drink type if clear, cup, light, angle, anything in the background.
Do not guess about taste, freshness, ingredients, milk type or where anything comes from.
If you cannot tell, write "unclear". Flag anything in the background that should not be public.Illustrative output (contains slips)
flatwhite-top-01: oat milk flat white with rosetta art, ceramic cup on saucer, soft window light
croissant-tray-03: freshly baked croissants warm from the oven on a metal tray, six visible
steam-hands-02: barista steaming milk in a steel jug, hands only, steam visible
window-seat-01: empty window seat, wooden table, daylight, a sticky note on the window edge
iced-vanilla-02: iced latte in a clear cup, condensation, straw, background blurredTwo slips and one catch, all typical. The assistant called the milk oat, which no photo can show, and the art a rosetta when it is a tulip. It wrote "freshly baked ... warm from the oven", a claim that would end up in a caption if you let it, when the tray was delivered, not baked on site. And it spotted the sticky note on the window, which is exactly why the last line of the prompt asks. Correct the log from your own memory, and only copy the "visible" facts. Add a line of alt text for each photo to the log at the same time, since the assistant can draft it from the visible description.
Captions: a voice sheet first, then batches of eight
Put a short voice sheet where the assistant will always see it: in a ChatGPT Project, a Claude Project or a Gemini Gem (becoming a skill from November 2026). Include how you sound, what you never say and what facts it may use. Keep it under 150 words, since a long sheet gets skimmed by people and models alike. Writing Instagram captions with AI that sound like you goes deeper on building one from your own past posts.
Voice sheet for [shop name], a neighbourhood coffee shop.
We sound like a friendly barista talking across the counter: short sentences, specific details.
We never say: delicious, perfect pick-me-up, treat yourself, foodie. No runs of exclamation marks.
We always name the drink as it is on the menu, give one concrete detail, and end with a plain
invitation or nothing at all.
Use only facts from the shot log. Never state prices, hours, ingredients or sourcing unless the log does.
Hashtags: at most 4, specific (drink name, shop name). Length: 10 to 30 words.
Give three options per photo.Then feed the log in batches of eight rows and ask for three captions per row. Compare a default caption with one edited from the log. The first is what an assistant writes with no facts; the second used the log row for the first flat white.
| Version | Caption | What is wrong or right |
|---|---|---|
| Default | Start your day the delicious way! Our freshly made flat white is the perfect pick-me-up. Come visit us today! #coffee #coffeelover #flatwhite #latteart #cafe #barista #coffeetime | Generic hook, an unprovable claim, seven hashtags and no detail a customer could use |
| From the log | Flat white, double shot, whole milk. The tulip took two goes. Open from 7. #flatwhite #[shopname] | Names the drink, one true detail, a plain invitation and two hashtags |
Instagram announced in December 2025 that it was limiting captions to five hashtags per post or reel, and press coverage describes a gradual rollout, so check what your own account allows. Four specific tags do the job in either case. Here are three more illustrative captions for different kinds of post, written from log rows:
- Pastry: "Butter croissants arrive at 7. If the tray is empty, that was a good day." (from the tray photo and the delivery fact)
- Process, for a reel: "Eighteen grams of coffee, a double shot, and the milk steamed while it pulls. #flatwhite #[shopname]"
- Space: "The window seat before anyone is in it. Open from 7."
Using the voice sheet, write three caption options for each of these log rows. Use only the
facts column. If a row has no usable fact, write "needs a fact" instead of writing a caption.
[paste rows 1 to 8]Illustrative output for flatwhite-top-01
1. Flat white, double shot, whole milk. The tulip took two goes. Open from 7. #flatwhite #[shopname]
2. A freshly poured flat white with a tulip on top. Pop in today! #flatwhite #coffee #latteart #barista #[shopname] #cafe
3. Whole milk, double shot, one slightly wobbly tulip. #flatwhiteOption 2 breaks three rules at once: it is generic, it has six hashtags and it ends with a stock invitation. Drop it, keep option 1 or 3, and borrow the opening hours from option 1 if the post needs them. Ten minutes of this per batch beats an hour of writing from a blank page, and the log means you are editing, not inventing.
Read each caption aloud before it goes further. If you would not say it across the counter, rewrite it. For longer posts or product launches, add a rule that the assistant must ask you for any missing fact, not fill it.
Reel batching: eight reels from twelve clips
A reel idea is a clip, a hook for the first two seconds and a line of on-screen text. Give the assistant your clip log (what each clip shows and how long it is) and ask for eight reel concepts that reuse clips in different combinations, each with a hook and text. Then choose the ones you would actually watch. Making TikToks and Reels for a small business with AI covers scripting and editing in more depth; this is the batching view.
| Reel | Built from | Hook, first two seconds | On-screen text | Length |
|---|---|---|---|---|
| One pour, no cuts | Pour clip | "Watch the milk, not the cup" | The drink's name | 9 s |
| Bean to cup | Scoop, espresso, steam, pour | "Fifteen seconds, bean to cup" | The four steps, named | 15 s |
| The pastry case | Reveal clip, tray close-up | "What is in the case today" | Pastry names you have checked | 8 s |
| Before we open | Sign flip, machine warming | "Five to seven" | Opening hours | 10 s |
| Anatomy of a flat white | Overhead pour clip | "What is actually in it" | Grams and millilitres from your recipe | 12 s |
| Iced build | Ice, milk, shot clips | "Built in layers" | The drink's name | 10 s |
| Hands only | Two machine clips | "The 8 am rush from the machine" | None | 12 s |
| New on the menu | Seasonal drink clips | "New this week" | Name and launch date, confirmed | 10 s |
Edit in the phone's own editor or whichever app you already use, keep the cuts slow enough for the drink to be recognisable, and caption the audio, because many people watch with the sound off. One scheduling detail matters here. Buffer's Instagram page says third-party tools cannot add music, stickers, polls or links to a post, so those must be added in the Instagram app. If you want trending audio, use notification-based posting for those reels and add the sound yourself. Otherwise post with the clip's own sound or none at all.
A posting rhythm you can keep for three months
The rhythm from the first sum becomes a weekly pattern that repeats, with a theme rotating by week. One week looks like this:
| Mon | Tue | Wed | Thu | Fri | Sat | Sun |
|---|---|---|---|---|---|---|
| Feed: drink of the week | Story: pastry of the day | Reel | Story: behind the scenes | Feed: pastry or space; Story | Reel | Feed: people or carousel; Story |
That is three feed posts, two reels and four stories. The four themes spread the 36 pieces so the same drink does not appear twice in a row: week one introduces the menu, week two shows the craft, week three covers pastries and the seasonal drink, and week four features people and the shop's routines. Ask the assistant to map the shot log to the 36 slots, with a column saying which file or clip each uses and why. It is the quickest way to see that you have 14 flat white pictures and only two of the pastry case.
Here is my shot log (40 stills, 12 clips) and my weekly pattern. Fill four weeks:
3 feed posts, 2 reels and 4 stories a week. Themes: week 1 menu, week 2 craft, week 3 pastries and
the seasonal drink, week 4 people and routines. Use each still at most once in the feed and never
put the same drink in consecutive feed posts. Mark any post that depends on a date as "date-sensitive"
and do not schedule it. Output a table: date, slot, file or clip, caption option, status.As for when to post, pick fixed slots near your busiest hours and adjust after a few weeks by looking at your own insights. Nobody else's best time describes your followers, and you can type your numbers into a sheet by hand and ask the assistant to compare weeks. I would put effort into keeping the rhythm steady before chasing clever timing.
Stories are the least scheduled part
Sixteen stories a month sounds like a lot until you see that most are small. Two a week can be scheduled in advance, such as the pastry of the day (a still from the pastry group) and a drink of the day with its price once you have checked it. The other two are better posted live: a behind-the-scenes clip from the machine and a "what is left in the case" picture an hour before closing. Interactive stickers such as polls have to be added in the Instagram app, because Buffer says third-party tools cannot add them, so keep those for the live ones.
Scheduling in Buffer or Later: what each plan allows
Both tools can schedule a month of Instagram in one sitting, and a month is about as far ahead as it is worth planning, because menus, prices and launch dates move. The differences are in limits and price, so here is what their own pages say as of October 2026, with the prices in USD:
- Buffer Free connects up to three channels and holds 10 scheduled posts per channel at a time, which you can refill as posts go out. It includes an AI assistant and reminder notifications for Instagram, but not first-comment scheduling or custom video covers.
- Buffer Essentials is listed at $5 a month per channel billed yearly ($60 a year), with unlimited scheduled posts per channel, first-comment scheduling for Instagram and custom video covers. Two channels, say Instagram and Facebook, would be $10 a month.
- Later Starter is listed at $18.75 a month billed yearly. It covers one social set (up to eight social profiles), 30 posts per profile per month and five AI credits a month, where one credit makes three ideas or one caption. Growth is listed at $37.50 a month billed yearly, with 180 posts per profile per month and 50 credits.
Both can publish carousels, stories and reels. Buffer's Instagram page says automatic publishing needs a Business or Creator account linked to a Facebook Page, that personal accounts get notification-based posting only, and that some reels and stories fall back to notifications when Instagram's own limits require it. Later's page describes auto-publishing for single images, carousels, Stories and Reels. Prices and limits change, so confirm them on each tool's pricing page before you pay. Later's pricing page counts Instagram Stories and Reels within the same 30 posts a month on Starter, so the full 36-piece month will not fit; scheduling only the two planned stories a week brings it to 28, with the other stories posted live.
| If you are | Start with | Because |
|---|---|---|
| One Instagram account, no budget for tools | Buffer Free | Schedule two batches of 10 and refill as they go out; the AI assistant is included |
| Instagram and Facebook, scheduling a month at once | Buffer Essentials | No cap on scheduled posts per channel; $5 a channel a month on yearly billing |
| Planning the feed visually and happy to pay more | Later Starter | Visual planner, link in bio and media library; stories count towards the 30 posts a month |
Later's AI credits on Starter stretch to five captions a month, so keep writing captions in your own assistant with the voice sheet, and use the scheduler only to publish. Buffer versus Later for a cafe or bakery's Instagram compares the two plans in more detail.
Owner approval: two minutes before anything goes live
AI drafts; the owner approves. Keep the plan in one sheet with a status column (draft, approved, scheduled) and let only approved rows reach the scheduler. For each row the check takes two minutes:
- Facts. Drink names, prices, hours and pastry names match today's menu, not last month's.
- Claims. Nothing you cannot prove: fresh, local, best, healthy. No promises about allergens or diets.
- People. Faces and names appear only where the person has agreed.
- Rights. Any music, logo or photo from someone else is cleared.
- Tags and links. The handles and links are right, and hashtags are within the limit.
- Voice. It sounds like you at the counter.
- Dates. Anything that mentions a day, a launch or an event has been confirmed.
The longer version, written for any AI-drafted post, is in the social media pre-publish checklist. Keep anything date-sensitive as a draft until you know the date, and schedule it in a short window.
A month later: three numbers and one realistic mistake
After four weeks, type three numbers per post into a sheet by hand: reach, saves or shares, and profile visits, copied from Instagram's own insights. Add a column for the theme and the slot, then ask the assistant which themes and formats produced the most of each, and which should get more of the next shoot's frames. A month is a small sample, so use it to decide what to shoot more of, not to declare a winner.
Now an illustrative mistake. A seasonal drink post was scheduled four weeks ahead for the 12th, telling people to "come and try it today". The launch slipped by a week because a syrup delivery was late, and the post went out anyway; three customers asked for a drink that was not on the menu. The fix was procedural: date-sensitive posts stay as drafts, move to the scheduler at most a week before, and the Friday check includes the menu board. Rerun the shoot every quarter in 45 minutes, using the gaps in your shot log as the list.
Beyond Instagram: where else the same shoot goes
The 40 stills should not live in one app. Pick six to eight of the best for your Google Business Profile photos, one or two for the top of your menu page or newsletter, and a few for table cards or a screen in the shop. Photos of the pastry case and the window seat are good candidates, because they show what a stranger will find on arrival. AI-drafted menu, hours and photos for a Google Business Profile covers that side, and the shot log you built here supplies the descriptions.
Questions about coffee shop social media with AI
How many posts a week should a coffee shop publish?
No number suits every shop. Start with a rhythm you can sustain for three months, such as three feed posts, two reels and a few stories a week, and judge it by your own saves, shares and profile visits. A small schedule that carries on is worth more than an ambitious one that stops after a fortnight.
Do I need to label captions written with AI?
Platform labelling rules mainly concern AI-generated or heavily altered images and video that look real, not captions you drafted with help and approved. If you ever generate imagery with AI, follow each platform's current labelling prompts. Keeping your photos real is the safer choice when honesty about the room and the drinks is part of your appeal.
Can I use trending music in scheduled reels?
Third-party schedulers cannot add music, stickers, polls or links for you; Buffer says so on its Instagram page. Either use notification-based posting and add audio in the Instagram app, post with original sound, or use Instagram's own scheduling. Check the platform's music rules for business accounts before relying on any track.
Is the free Buffer plan enough for one coffee shop?
For one Instagram account posting a few times a week it can be. The free plan connects up to three channels and holds ten scheduled posts per channel at a time, so you schedule in batches and refill as posts go out. A monthly batch of twenty or more needs two rounds, or a paid plan.
How do I stop AI captions sounding generic?
Give the assistant specific facts for each photo, such as the drink, the milk and one detail from the day, plus a short voice sheet with phrases to use and avoid. Ask for three options and edit the best. Delete any claim you cannot prove, and read each caption aloud once; if you would not say it at the counter, rewrite it.
Further reads
- How to Plan a Month of Marketing Content With AI in One Hour — A one-hour method for planning any month of marketing content.
- ChatGPT Prompts for Coffee Shop Owners: 10 for Menus, Posts, Rotas — Ten ChatGPT prompts for menus, posts and rotas.
- How to Label AI-Generated Images and Video on Social Media — How to label AI-generated images and video on social platforms.
- Can a Small Cafe Use AI to Bring In More Regulars? — Use AI to bring more regulars through the door.
- Ice Cream Shop Marketing Ideas: Weather-Triggered Posts With AI — Skip the idea list. Use the forecast to decide when to post and what to offer, with templates, margin sums and an AI-drafted stock note.
- Yarn Shop Marketing: Fill Knitting Classes and Promote New Stock With AI — A class calendar, listing and caption prompts, a monthly newsletter template and a three-hour routine for a small yarn shop's marketing.
- AI Marketing Campaign Examples: Five Worked Ones for Local Businesses — Five worked local campaigns, each with a prompt, assets, schedule, cost, break-even point and a simple way to check whether it paid for itself.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Buffer's pricing page and Instagram page; Later's pricing page and Instagram scheduler page (all as of October 2026); press reports of Instagram's five-hashtag limit, which is rolling out gradually. Captions, counts and schedules in the examples are illustrative.