Automate Instagram posting with your agents

The Instagram Graph API only works on Business or Creator accounts. Personal accounts have no API access. The Business account also needs to be linked to a Facebook page (a quirk of Meta's auth model, even if you never post to FB). Set this up first; without it, none of the steps below work.

Authenticating to the Instagram Graph API takes 4 steps: register an app on Meta for Developers, configure the Instagram Graph API product, run the OAuth flow to get a short-lived token, then exchange for a 60-day long-lived token. The long-lived token then needs refreshing every 60 days unless you wire renewal automation.

Before drafting new content, audit what already worked. Pull the last 90 days of posts and rank by saves (the strongest signal of value to a viewer), reach (the algorithm's signal of share-worthiness), and replies (the relationship signal). Patterns will jump out: which hooks land, which formats, which days. Train the agent on these patterns, not on Instagram-marketing-blog generic advice.

Audit the last 90 days of our Instagram posts and surface the patterns.

Pull post-level data via the Graph API endpoint /{ig-user-id}/media for the past 90 days. For each post, fetch:
- Caption
- Media type (carousel / reel / image)
- Posting day-of-week + hour
- Saves, reach, likes, comments, profile visits

Output:
1. The top 10% of posts by saves-per-impression. What do they have in common? (Hook structure, format, time, length, hashtag count)
2. The bottom 10%. What patterns to avoid?
3. The strongest day-of-week + hour combinations
4. The strongest hashtag SETS (not individual tags) by avg-saves-per-impression
5. Recommended posting cadence: how many of each format per week

Output as a markdown doc; this becomes the agent playbook in the Brief surface.

Without explicit pillars, an agent will default to generic Instagram-marketing voice (emojis, exclamation marks, '🚀 Excited to share'). Define 3-5 content pillars, your brand voice rules (tone, banned words, sentence shape), and your visual identity rules. Hand these to the agent as the system prompt for every draft.

A weekly cadence is too short to plan around campaigns; a monthly view exposes the rhythm. Build a 30-day grid with: date, day-of-week, time, content pillar, format (carousel/reel/single/Story), hook, caption draft, hashtag set, asset link, status. The agent populates rows, the human approves.

The agent reads the Brief doc + the Content calendar row + the audit playbook, and outputs: a hook (first 125 chars), a full caption (up to 2200 chars), a hashtag set (8-15 hashtags), and alt-text (100-150 chars). The human reviews in the Content calendar table and either approves, edits, or rewrites.

Draft the caption + hashtag set + alt-text for an Instagram post.

Inputs from the calendar row:
- date
- pillar (e.g., founder POV, product tip)
- format (carousel / reel / single image)
- asset URL or description

Inputs from the Brief surface:
- brand voice rules
- audience profile
- content pillar definitions
- audit playbook (what worked in last 90 days)

Output:
1. Hook (first 125 chars): the line that hits before the 'more' fold. No emoji-only opens, no '🚀 Excited to'.
2. Body caption (additional 200-1000 chars): supporting context, story, or tip.
3. CTA line (last 1-2 lines): what should the viewer do? (Save, share, follow, click bio link.)
4. Hashtag set (8-15 hashtags): mix of niche (10k-100k posts) + broad (100k-1M). Never reuse the same set within 24 hours.
5. Alt-text (100-150 chars): descriptive for accessibility AND search.

Constraints:
- Adhere to brand voice rules in Brief
- Match the format-specific best practices from the audit
- No exclamation marks unless brand voice explicitly allows
- Do NOT invent product features or stats — flag uncertainty instead

The cleanest pipeline: agent drafts in the Content calendar table, human approves, scheduler queues the post, scheduler publishes via Graph API. Buffer + Later both expose APIs your agent can drive. The free Meta Business Suite scheduler doesn't have an API but works for low-volume manual flows.

Every post has a 30-60 minute engagement window where the algorithm decides who else sees it. Replying to comments in that window dramatically extends reach. Configure your agent to surface incoming comments in a Reply queue, draft response candidates, and ping you for the human-voice reply.

Don't ship the full pipeline to a 12-post-a-week cadence on day one. Run a 30-day pilot at half-cadence (3 posts/week) so you can iterate the agent prompt, the calendar template, and the approval flow before scaling. The metric to track: agent draft → approval rate. Above 80% means the prompt is dialled; below 50% means re-tune the Brief.

After 30 days of pilot, scale up. The pipeline now runs continuously: agent drafts on Sunday for the next week, human approves Monday morning, scheduler queues for the week. Monthly: re-run the audit, update the Brief, surface the top 10% / bottom 10% patterns, and feed back to the agent.

Run the monthly Instagram analytics loop.

For the past 30 days:
1. Fetch all posts via /{ig-user-id}/media
2. Compute: saves-per-impression, reach, reply count, follower delta per post
3. Compare to the previous 30 days (month-over-month)

Output to the Brief doc as a markdown section:
- Top 10% performers + the patterns they shared
- Bottom 10% + what to avoid next month
- Best-performing day-of-week + hour
- Best-performing hashtag sets
- Recommended changes to the content pillar mix
- One concrete experiment to try next month

Update the agent playbook in the Brief based on findings; the next month's drafts should reflect the updated playbook.