
Give yourself an hour to prepare a small calendar you can review, rather than a month of captions nobody has checked. AI can help sort source material and draft versions. It cannot supply a confirmed launch date you have not decided or turn an unfinished customer story into approved evidence.
The one-hour limit here is a working-session plan, not a promise about how quickly every team can finish. It assumes your accounts are connected, you have usable source material, and one person can make routine editorial decisions. If those conditions are missing, use the hour to identify the gaps.
Aim for a short planning horizon, such as the coming week, with fewer posts if necessary. A calendar with four defensible ideas is more useful than fifteen entries that all need research before anyone can release them.
Minutes 0–10: collect facts and constraints
Open the material you already have: confirmed announcements, published articles, recurring customer questions, approved photos, and the current offer. Record where each fact came from. Mark dates or claims that still need someone else’s answer.
Write down your audience and the action each post should support. “People considering their first repair appointment” is specific enough to guide a draft. “Everyone interested in our industry” gives the writer little help deciding what to explain.
Then list practical constraints. Who reviews? Which assets are ready? Does a client need to approve pricing? When is the business actually available to answer responses? These are planning inputs, even though they are less interesting than brainstorming headlines.
Do not spend this opening block searching for a perfect strategy document. Use a compact note with links to source material and a visible list of missing facts. You need a dependable starting point for this session, not a finished brand manual.
Minutes 10–20: choose a few distinct jobs
Ask the assistant to group the material by the question it answers for a reader. A timetable announcement, an explanation of how booking works, and a repair tip have different jobs. Several captions that all say “book now” remain the same idea with different wording.
Use a prompt such as:
From these sources, propose four posts for next week. For each, give the reader’s question, the factual source, the needed asset, and the next action. Flag missing information. Do not invent events, statistics, testimonials, or deadlines.
Choose the ideas yourself. Remove a slot if its only purpose is to fill Tuesday. You can leave room for a timely update without promising that something newsworthy will happen.
At the end of this block, you should be able to explain why each entry deserves to exist. If two entries answer the same question with the same evidence, combine them or choose a different source.
Minutes 20–35: draft from the selected sources
Give the writer your chosen ideas and a few examples of your normal language. Ask for a complete caption rather than a headline that someone still needs to expand. Keep facts and writing preferences separate so the model can see what must not change.
Start with the channel that needs the fullest explanation, then adapt the idea for other destinations. Do not require one identical paragraph everywhere if a network needs a shorter version or the audience needs different context.
Postbean’s hosted MCP server provides create_post_draft for saving drafts and update_post for revisions. It supports per-network copy through overrides. The local @postbean/mcp package instead uses create_post; set draft: true explicitly if you use that transport.
Save the IDs you receive next to your planning entries. If you change an opening later, edit that existing draft. Creating a new object for each revision can leave a discarded version in the calendar.
If you use Postbean’s AI-generation tools, account for their credit consumption. Having a connected assistant write text and saving that text is a different operation from calling a Postbean generation tool.
Minutes 35–45: review the substance
Read every draft against its source. Check the date, conditions, link, and any numerical claim. Remove unsupported urgency and invented praise. A model may produce a smooth sentence that introduces a promise nobody intended to make.
Then inspect the opening. Does it tell the reader what changed or answer a real question? “An exciting new chapter” gives less information than the actual change in booking hours. Keep concrete details that help someone decide whether the post is relevant.
Review assets alongside the copy. A correct caption attached to an old price card is still wrong. If you have not checked permission to use a customer photo or quote, leave that entry unresolved rather than inventing a substitute story.
Give unresolved entries an owner in your planning note. “Waiting” is less useful than “Sam must confirm the workshop capacity.” This ownership is part of your working process; do not assume the scheduling application has created an approval assignment merely because you wrote one in a prompt.
Minutes 45–55: assign times with a real clock
Look at your proposed week as a whole. Avoid placing several similar posts together just because the assistant returned them in that order. Consider the business’s actual dates and the time someone can respond to questions.
Postbean has a calendar and supports both shared and per-network schedules. Choose one shared time when that matches the plan. Use separate network dates when there is a concrete reason; extra scheduling detail also creates more things to review.
On hosted MCP, call get_workspace before setting times. It provides the workspace time zone and current local time. An offset-free value is read in that zone; a timestamp with Z identifies a UTC instant. Ask the assistant to return the saved dates with their offsets.
Do not turn “AI suggested 09:00” into a claim about your audience’s best posting time. For this first session, pick defensible operational times. You can revise future schedules using actual results and your team’s experience.
Only schedule the entries that passed review. A scheduled post is authorized to go out later. It is not a draft that will automatically pause and request a client’s approval before sending.
Minutes 55–60: read back what exists
Ask for a compact report containing each saved post ID, target accounts, state, and scheduled time if present. Compare it with your planning note. Confirm that unresolved entries remain drafts and that no discarded version is scheduled.
Postbean reports text-limit problems through publish_issues. Fix the affected network version before scheduling. A draft can exist with a problem; saving it is not the same as confirming that it is publishable.
If a URL upload is still processing, inspect that media rather than uploading it repeatedly. If an entry depends on missing video or a required platform choice, keep it out of the released batch until those pieces are resolved.
What to do when the hour runs out
Stop with a clear record of what is ready and what is blocked. You may have two scheduled posts and two drafts instead of four scheduled posts. That is an acceptable result when the missing information is real.
Write down the smallest next action for each blocked entry. A factual confirmation, asset export, or short client response is easier to finish than “improve content.” Schedule the next planning session around collecting those inputs, not around generating more captions from the same incomplete brief.
Postbean’s Approvals tab is currently a coming-soon placeholder, so keep any required approval record in your existing process. Do not treat the calendar as proof that someone reviewed the content.
After publication, inspect account-level outcomes and the actual responses that matter to the post’s purpose. Bring those observations into the next session. You can start with the calendar, improve the source material, and use the caption editing method when a draft sounds interchangeable.
Sources
- Postbean MCP guide — checked November 22, 2026