How to run a social media content calendar with an AI agent

A content calendar is where most social media plans quietly stall: ideas pile up, drafts wait, and nobody knows which posts worked. An AI agent can do the planning and drafting; a person keeps the judgment.

An AI-run content calendar is a weekly loop. The agent proposes next week's posts from your content pillars, recent results and this week's facts; a person approves them in one batch; the agent writes a version per network and schedules it; and tagged links let your CRM show which posts brought customers. The next plan starts from those results.

This guide gives you the prompts for each step, a pillar grid, a sample week and the review routine.

What should a content calendar contain?

At minimum: the date and time, the network, the pillar, the format, the draft, the link with UTM tags, the approver and the status. Everything else is optional.

A spreadsheet works for one person. Once an agent writes drafts and two people approve, a calendar inside your publishing tool saves the copying.

How do you choose content pillars and themes?

Pick three or four pillars that match how your buyers decide, give each a share of the week, and set one theme per month that all pillars point at.

A monthly theme ties them together. If November's theme is "month-end close", the problem posts cover close mistakes, the proof posts show a customer's close time, the product posts demo the report, and the opinion post argues for a change in how teams close.

The best inputs for pillar topics are the questions people ask you. Sales call notes, support tickets and comments are better sources than keyword tools for B2B. A social listening routine adds the questions buyers ask each other.

How often should you post on each network?

Choose a cadence you can keep for three months, then adjust from results. For LinkedIn, Buffer analyzed more than 2 million posts from more than 94,000 accounts and found that "posting 2 to 5 times weekly on LinkedIn is the sweet spot for improving reach and engagement" (Buffer, August 2025; a vendor study).

A starting cadence for a small B2B team, as a suggestion to test:

What prompt should you use to plan the week?

Use one standing prompt that gives the agent the pillars, the month's theme, last week's results and the facts it may use. Ask for a table first, so you approve the plan before anyone writes.

You are planning next week's social posts for [company].

Audience: [who buys, role and company size].
Pillars and weekly share: customer problems 35%, proof 25%, product 20%, point of view 20%.
This month's theme: [theme].
Networks and posts per week: LinkedIn 4, X 6, Threads 4, Instagram 2, YouTube Short 1.

Facts you may use this week (do not invent others):
- [launch, number, customer quote with permission, event date]
- [three questions customers asked this week]

Last week's results:
- [top 3 posts by contacts or clicks, with one line on why]
- [bottom 3 posts]

Plan 10 to 12 post ideas. For each: day, time, networks, pillar, format,
one-line angle, the fact it uses, and the link it points to.
Leave two open slots for timely posts.
Do not repeat an angle from the last four weeks: [paste last month's angles].

The "last week's results" block is what makes the calendar improve. Without it, the agent plans the same mix forever.

How do you batch a week of posts with an agent?

Batch in one sitting: approve the plan, have the agent write every post, review all drafts together, then schedule. A team of one can finish the week in about 90 minutes.

Edit the master draft before the network versions exist. One fix there saves the same fix in six places. Our guide on LinkedIn posts with AI that don't sound like AI has the editing checklist we use at step 4.

Who should approve AI-drafted posts?

One named person approves each week's posts before they are scheduled. A second person checks only posts with legal, pricing or customer-name risk.

Rules that keep approvals fast:

• Approve the week in one session. One review beats 20 interruptions.
• Approve the versions as well as the master draft. A fine LinkedIn post can turn into a misleading 280-character X version.
• Keep a short list of "always ask" topics. Pricing, customer names, competitors, security, hiring.
• Track edit rate. If most drafts need heavy edits, fix the prompt, pillars or brand rules before adding more volume.

The approval workflow guide covers queues, roles and what agents should never approve.

How do you repurpose a blog post into per-network versions?

Ask the agent to pull several separate ideas from the article, then write each idea for the network where it fits best. One summary posted everywhere wastes the article.

Here is a blog post: [paste or link].
1. List the 6 most useful standalone ideas in it, one line each, with the
   supporting fact from the article.
2. Map them:
   - LinkedIn: the idea with the strongest story, 800-1,300 characters, link in first comment
   - X: a 4-part thread of the step-by-step idea, each part under 280 characters
   - Threads: one idea under 500 characters with one topic tag
   - Instagram: a 6-slide carousel outline plus a caption with at most 5 hashtags
   - YouTube Short / TikTok: a 45-second script for the most visual idea
   - Reddit: a discussion post written as a member, method first, product last, disclosed
   - DEV: the full article with up to 4 tags and canonical_url set to the original
3. Every link uses utm_source=<network>, utm_medium=social, utm_campaign=<post slug>.
4. Keep this fact identical everywhere: [fact].

The limits in that prompt come from each platform's docs as of November 2026. Our cross-posting guide has the full table for 13 networks, with sources.

How do you measure which posts brought customers?

Tag every link, match CRM contacts back to the posts, and review monthly. Plan from contacts and deals, and treat likes as a secondary signal.

1. Tag links. Use utm_source for the network, utm_medium=social, utm_campaign for the post or theme, and utm_content to tell versions apart, per Google's UTM definitions.
2. Match contacts. New CRM contacts are matched to posts by UTM, click ID or original source. Our guide to UTM parameters for social media posts has a naming scheme that survives copy and paste.
3. Review monthly with a prompt.

From last month's posts and the CRM contacts matched to them:
- Rank posts by contacts, then by deals.
- For the top 5, what do they share (pillar, format, network, angle, day)?
- Which pillar produced engagement but no contacts?
- Recommend next month's pillar shares and three angles to repeat.

Change the pillar shares in the planning prompt based on the answer. That closes the loop.

A sample week

An example for a fictional B2B invoicing product during a "month-end close" theme.

Running the calendar in Tracee

In Tracee, an agent in ChatGPT, Claude or Cursor connected over MCP, or Tracey (the built-in assistant), can plan, write and schedule into one calendar across 13 networks. Each draft becomes a version per channel held to that network's rules, brand rules flag banned phrases before anything goes out, and agent posts wait in the approvals queue by default. RSS auto-post and repeated posts cover recurring content. From the CLI:

tracee posts next-slot --channel-ids <channel_id>
tracee posts create --body "..." --channel linkedin --channel x --link https://example.com/post --publish 2026-11-18T09:00
tracee analytics engagement --days 7

Links go out with UTM tags per channel, and HubSpot contacts are matched to the post that brought them in, which feeds the monthly review. Plans and limits are on the pricing page (at the time of writing, November 2026).

Next step

Write your three or four pillars and this month's theme, then run the weekly planning prompt once with real facts from this week. Approve the plan, batch the drafts in one sitting, and tag every link. After four weeks, run the monthly review prompt and change one thing.