Social Media Marketing AI Prompts for a Month of Client Posts

Give ChatGPT or Claude five prompts in order: a brand voice brief built from the client's real posts, three to five content pillars, a four-week calendar, caption variants for each slot, and a final approval pass that checks facts, tone and banned claims. Saving the first two in a Project keeps every later chat on-brand.

The order matters more than the wording. Most people ask for "30 posts for a bookshop" and get generic cheer. A manager who briefs the voice first, then the pillars, then the calendar, gets drafts a client can approve with light edits. Each stage's output becomes the next stage's input, so a flaw caught in the brief costs one edit, while the same flaw caught in the captions costs twelve.

Follow me on Instagram@sagnikteaches

The client and the numbers

Our example: a small independent bookshop with one shop, an events room and a loyal local customer base. It posts on Instagram and Facebook, three times a week, and wants 12 posts for the month. The agency owner has 90 minutes to plan, draft and prepare approvals. Without a method that would be tight; with this sequence the time splits roughly as follows (illustrative):

Connect on LinkedInSagnik Bhattacharya

Set up a Project (ChatGPT Projects or Claude Projects) per client. Paste the brief in as project instructions or upload it as a file, and every new chat inherits it. On business plans these Projects can be shared with colleagues, so the person covering your holiday starts from the same voice. If you want the wider picture of what an agency should and shouldn't automate, read which social tasks an agent can take over first; the prompts below are the hands-on drafting half.

Subscribe on YouTube@sagnikteaches

Stage 1: the brand voice brief

Everything downstream depends on this document. Gather five things from the client before you prompt: three to five of their best posts, two posts they disliked or would never write, a list of words they use and words they avoid, facts that never change (opening hours, address format, what they sell, what they do not), and topics that are off limits. Then ask the model to turn them into a one-page brief.

I manage social media for a small independent bookshop. Below are four
posts the owner loves, two they hate, and notes on how she talks.
Write a one-page brand voice brief I can paste into every future prompt.
Include: 1) the voice in five adjectives with a one-line example of each,
2) sentence length and punctuation habits, 3) words and phrases to use,
4) words and phrases never to use, 5) emoji rules, 6) what the shop never
claims or promises. Do not invent facts about the shop. Where my notes are
thin, write [NEED FROM CLIENT] and list the questions at the end.

LOVED POSTS: [paste]
HATED POSTS: [paste]
OWNER NOTES: [paste]

A trimmed illustrative result:

VOICE: warm, dry, specific, unhurried, a bit nosy about what you're reading.
Short sentences, one idea per line. Full stops over exclamation marks.
USE: "on the table by the till", "staff pick", "come and browse".
NEVER: "must-read", "bestseller" (unless true and dated), "grab yours now".
EMOJI: none, or one book emoji at the end at most.
NEVER CLAIM: discounts, signed copies, stock levels unless in the weekly notes.
[NEED FROM CLIENT]: Do you reply to every comment? Do you name staff?

Read it as the client would. The "never claim" line is the most valuable part: it was derived from the hated posts, where the owner disliked an unsupported "bestseller" label. Send the open questions to the client and update the brief. When several people on your team write for the same client, how agencies keep brand voice consistent covers holding the voice steady across writers.

Stage 2: content pillars that are not "tips and promotions"

A pillar is a repeatable kind of post. Without them the model produces an even mix of vague ideas. Ask for pillars grounded in the shop's actual activity, and ask it to show its working.

Using the voice brief, propose four content pillars for a month of posts
for this bookshop. For each: a name, what a post looks like, which real
shop activity feeds it (events, new stock, staff recommendations, customer
questions), how often per month, and one risk to avoid. Total must be 12
posts. Flag any pillar that needs material I must get from the client.

Typical sample output, condensed: Pillar A "On the table this week" (4 posts, fed by new stock; risk: implying stock the shop does not hold). Pillar B "Staff pick of the week" (3 posts; needs a short note from a staff member; risk: the same person every time). Pillar C "Events and signings" (3 posts; needs dates; risk: promising an author before confirmation). Pillar D "Ask the shop" (2 posts, answering a real customer question; risk: personal data in screenshots).

What you fix: the model often over-weights promotion. If three of four pillars exist to sell, change one. A useful rule of thumb is that at least half of the posts should give something away (a recommendation, an answer, a behind-the-scenes look) rather than ask for a sale. Note also that each pillar names the material the client must supply. That list becomes your weekly request email.

Collecting the raw material

Create a single form or email template for the client: three photos from this week, one staff pick with a sentence on why, event dates and times, one question a customer asked. Models write well from specifics and badly from nothing. If the client supplies nothing, say so in the calendar rather than letting the model invent events.

Stage 3: the four-week calendar

Now ask for the calendar in a table you can paste into a spreadsheet. Provide the dates, the closures and the confirmed events, and forbid invention.

Build a calendar of 12 posts for [month], three a week, posting Tuesday,
Thursday and Saturday at 10:00. Use the four pillars in the stated shares.
Confirmed facts only: author event Thursday the 14th at 19:00; shop closed
Monday the 25th (public holiday). Output a table with columns: date, pillar,
format (single image / carousel / short video / text), purpose, the
material needed from the client, and any fact to confirm. Mark anything
you assumed with [ASSUMED].

Sample rows (illustrative):

What to fix: look for clustering (two event posts back to back), formats the client cannot actually produce (a video when nobody films), and anything marked [ASSUMED]. Move the "Ask the shop" post away from the day of the event, and replace any slot that needs material the client has said they cannot provide. If you work with a client that posts across Instagram and LinkedIn, run the calendar once per channel: the same slot rarely wants the same post. A realtor example with a different mix of pillars is in a month of posts from listings.

Stage 4: caption variants that do not all sound alike

Ask for two variants per slot, and make the variants differ in structure, not just in wording. Run the calendar one week at a time (three posts per prompt) so the voice does not drift.

Write caption variants for week 1 of the calendar (three posts). For each
post give Variant A (under 40 words, starts with a concrete detail) and
Variant B (50 to 80 words, starts with a question or a small story).
Follow the voice brief exactly. Use only facts in the calendar and the
brief; write [CHECK: ...] where a fact is missing. Add alt text under 125
characters and one hashtag set of at most four tags. List the first five
words of every caption at the end.

A sample output for the Tuesday post (illustrative):

Variant A: Three new novels on the table by the till. One is already
being passed between staff. [CHECK: titles]
Variant B: Which book did you finish last and wish you hadn't? We ask because
this week's table is built around the ones people finished quickly. Come and
see if you agree. [CHECK: opening hours]
Alt text: A wooden table by a shop counter with a neat stack of new paperbacks.
First five words: Three new novels on the | Which book did you finish

What you fix: Variant B's line "built around the ones people finished quickly" claims something the client never said. That is an invented rationale, and it reads well, which is why it slips through. Cut it or ask the client. A second common fix is length: models run long when a post is meant to be glanceable, so keep variant A as the default and use B for slower platforms. The first-five-words list is a cheap repetition detector; if three posts start "Come and", rewrite two.

For the image side of captions, writing Instagram captions with AI that sound like you has more prompts for turning a photo into a caption.

Stage 5: the approval pass

The last prompt is a different job: not writing, but checking. Run it in a fresh chat, with the brief and the drafts, so the model is not defending its own work.

You are a careful editor for a small bookshop's social media. Check these
12 drafts against the voice brief. For each post, return: (1) any claim not
supported by the brief or the calendar facts, (2) any word from the NEVER
list, (3) any sentence the owner would find unlike her, (4) a corrected
version if needed. Then list every fact the client must confirm before
publishing as a numbered list with the post it belongs to. Do not praise
the drafts.

The "do not praise" line matters: reviewers built on chat models are polite by default, and a polite reviewer misses things. The returned fact list becomes the approval email, which goes to the client as a numbered list with one question each: "Confirm: author event is Thursday 14th at 19:00, tickets free?" A short list with yes/no questions gets answered; a ten-page draft pack does not. For a full human sign-off routine once the drafts are clean, use the pre-publish checklist for AI-drafted posts.

A second client, to show the prompts bend

Swap the bookshop for a landscaping company and the structure holds while the content changes. The voice brief says "plain-spoken, practical, no hype"; the pillars become "finished jobs", "seasonal how-to", "meet the crew" and "ask the gardener". The calendar needs different material: job photos with permission from the homeowner, which you must request before the month starts. The approval pass adds a rule: never imply a guaranteed result for a lawn or planting, and never show a customer's house number. Notice that what changes is the list of "never" and "need from client" items, not the five-stage shape. That is why the sequence works as a template: the thinking sits in the brief, and the brief is where each client differs.

Turning client feedback into the next prompt

The month does not end when you send the approval email. Clients reply with comments such as "too salesy" or "we'd never say that", and the useful move is to feed those comments back into the brief rather than fixing the single post. Paste the client's message and the draft into a fresh chat and ask the model to explain the rule behind the complaint.

The client said this about the attached draft: "Too salesy, we don't do
urgency." Write the underlying rule in one sentence I can add to the voice
brief, and show the draft rewritten to follow it. Also list any other
phrase in the other 11 drafts that breaks the same rule.

An illustrative before and after. Before: "Only three copies left of this week's staff pick, so come in soon!" After: "This week's staff pick is a quiet, funny novel about a lighthouse keeper's letters. It's on the shelf by the window." The rule added to the brief: no scarcity wording, no countdowns. The second half of the prompt is the valuable part: it hunts for the same fault in the other eleven posts, so the client does not find it on page four.

Adapting one month for a second channel

Many clients want the same content on Instagram, Facebook and a business network. Do not ask for twelve more captions from scratch; ask for adaptations, and tell the model what changes.

Adapt the approved Instagram posts for week 1 into LinkedIn versions for
the shop owner's personal page. Keep the facts identical. Change: first
person singular, 80 to 120 words, one clear takeaway, no hashtags, no
emoji. Do not add any claim that is not in the Instagram version.

Illustrative result for the Tuesday post: "We put three new novels on the table by the till this morning. One of them has already done the rounds of the staff room, which is usually a good sign. If you are in the shop this week, tell me which book you finished last and wish you hadn't. It's the question that gets the best recommendations." The check is the same as before: the statement "done the rounds of the staff room" must come from the notes, and the instruction "tell me" must be something the owner is willing to answer.

Short video scripts from the same calendar

If a slot is marked "short video", ask for a shot list and a spoken script rather than a caption alone. A fifteen-second video is about 35 to 40 spoken words, so tell the model the word limit.

Write a 15-second video script for the Thursday staff pick slot. Maximum 40
spoken words. Give: a first-second on-screen text, three shots (what the
camera shows), the spoken line for each shot, and a one-line caption. The
speaker is a bookseller holding the book. No claims beyond the staff note.

The output is only as good as what the shop can film. If the client has one person who is comfortable on camera, write to that person's habits and say so in the brief. A script that asks a nervous owner to deliver a punchy line will get ignored. If a client's month has three or more video slots, write all the scripts first and ask for one filming session, so the person on camera records the month in an hour rather than being chased every week.

What the client approval email looks like

Keep the email short and specific, because the pack should be easy to approve quickly. An illustrative version:

Hi [first name], the 12 posts for next month are attached as one document.
Three things need your yes or no before I schedule:
1. Author event on Thursday 14th at 19:00, free entry, booking link in bio?
2. The staff pick on the 7th: is the book on the shelf and who recommends it?
3. Closed on Monday 25th: should the Saturday 23rd post say so?
Everything else follows the voice we agreed. Please reply by Friday and I
will queue them for the Monday.

A client can answer that in two minutes. Compare it with an email that says "Please see attached and let me know your thoughts", which is likely to wait a week. The fact list from stage five gives you the numbered items for free.

A quick sum on what the workflow saves

Take the agency's old routine for the bookshop: about 4 hours a month planning, writing and polishing 12 posts, plus about 1 hour chasing the client for details. With the five prompts and a saved Project, a realistic second-month figure is around 90 minutes of agency time (the table above) plus about 20 minutes of the client's. If your agency has ten similar clients, that is a saving of about 35 hours a month before you count the tidier approvals. It is an estimate: your first month will run longer because the brief takes the longest, and a messy client will take more time whatever you do. The sum becomes useful when you rerun it each quarter with your own timings.

When the client sends nothing

Some clients never return the weekly material. Decide in advance what the agency does, and write it into the brief rather than improvising each month. A workable fallback is a short list of "evergreen" slots that need no new facts: a shop-front photo you already hold, a reminder of regular opening hours the client has confirmed in writing, a recommendation drawn from the staff picks archive, or a repost of a past post that did well, with a fresh caption.

Ask the model to build that reserve once, in the first month, with a prompt such as: "Using only the facts in the voice brief, draft eight evergreen posts that need no new information from the client, each with a note on what photo already in our library would suit it." Store them in the Project. When a week arrives with no material, you pull two from the reserve and flag to the client that you did so. An illustrative failure to avoid: a landscaping client sent no photos for three weeks, the agency asked the model for "seasonal tips", and the output told followers to prune a shrub that the client's own customers had planted with a different growing season. Evergreen posts built from the client's own facts prevent that kind of confident, wrong generic advice.

Finally, log each month whether the reserve was used. A client who needs it every month is a client whose retainer should include a content day, and that conversation is easier with the log in hand.

Four mistakes that make a month of posts feel generic

1. Skipping the brief. Asking for captions with only the client's name and industry yields interchangeable copy. If a caption could be posted by any bookshop, the brief is thin.
2. Letting the calendar invent events. A model will happily schedule a "Meet the author evening". Insist on [ASSUMED] flags and check each one.
3. One chat for everything. By the third week of captions the model is imitating itself. Fresh chat per week, brief loaded from the Project.
4. Reviewing your own drafts in the same chat. The model tends to approve what it just wrote. Use a fresh chat and the editor prompt.

Keeping the workflow running month to month

After the first month, the brief barely changes; you update it with what the client approved, rejected and edited. Save the final approved posts in a file called "approved examples" and swap the oldest three into the voice brief every quarter, so the voice tracks what the client actually likes rather than what they liked on day one. Keep a running list of the edits clients make most, such as "removes exclamation marks" or "always adds the booking link", and add those as rules. After three months, most of your corrections turn into permanent instructions and the edit time per post falls.

Track two numbers per client: minutes from brief to approved month, and the number of posts the client changed. If minutes fall and changes fall, the system is working. If minutes fall and changes rise, you are saving your time by spending the client's patience, which is a bad trade. Use scheduling tools such as Buffer or Metricool for the actual queue; both have free or low-cost entry plans, though you should check the current pricing pages. For a view of the time you might recover across an agency, how much time AI can save on social media management gives a sense of the range.

What to keep for yourself

Do not hand the model the part of the job that clients pay you for: knowing what the client should and shouldn't say this month. Choosing the pillars, killing the post that sounds off, and noticing that a fixed Saturday slot clashes with the client's closing day are judgement calls. The prompts compress the typing; they do not replace the strategy. If you want a second pair of eyes on how to build this into your agency's routine, I cover that in my AI implementation consultation, which you can request through the contact page.

Questions about prompting for client calendars

Should I use one long chat or several separate ones?

Use a Project per client with the voice brief and pillars saved in it, then start a fresh chat for each stage. Long single chats drift: by caption 20 the model is copying its own earlier phrasing instead of the client's voice.

How many example posts does the voice brief need?

Three to five approved posts is enough. More than eight tends to make the model imitate them too closely. Choose examples that differ in length and purpose so it learns the voice rather than one format.

Can I paste a client's customer comments into ChatGPT or Claude?

Only on a plan whose terms you have checked. Business plans from both vendors do not train on business content by default, whereas free consumer plans may. Strip names and contact details from anything you paste regardless.

How do I stop every caption starting the same way?

Ask for a different opening device in each variant (question, number, statement, scene) and tell the model to list the first five words of every caption so you can see repeats. Then reject any opener used twice.

Further reads

• Personal Trainer Social Media Posts: A Month of Content With AI — A month of content for a different kind of client.
• Interior Design Instagram Captions: A Month of Posts From Photos With AI — Captions from photos, a useful variation on stage four.
• AI Marketing for Plumbers: A Worked 12-Month Posts and Offers Calendar — Seasonal calendars with offers, for trades clients.
• AI Social Media Monitoring: A Monthly Insight Report for Clients — Turn the month's results into a client insight report.
• How Much Does It Cost to Automate Social Media With AI? — Work out what the tools cost against hours saved.
• How to Label AI-Generated Images and Video on Social Media — Labelling rules when the posts include AI imagery.
• How to Build a Brand Voice Guide That AI Can Follow — Build the full voice file that stage one starts.
• Coffee Shop Social Media With AI: Captions, Reels and Rhythm — How one photo session feeds a month of posts.
• Social Media Client Onboarding: A Template and AI Welcome Pack — Collect the right client inputs, fill a practical onboarding record and turn approved facts into a reviewed welcome pack.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: OpenAI help article (Projects in ChatGPT), ChatGPT Business pricing page, Claude pricing page and Claude help article (What are projects?), Buffer and Metricool pricing pages, read 4 November 2026.