Instagram caption writer for an Airtable editorial plan

Instagram caption writer for an Airtable editorial plan

An n8n workflow that watches an Airtable editorial plan for new rows. A minute after a row appears, an OpenAI agent
reads its briefing, looks up your target group and wording notes in a second table, writes an Instagram caption and
saves it in the same row.

What is inside

• workflow.json: the workflow as n8n exports it, 9 nodes and one sticky note. It holds no credential.
• README.md: these notes.

What we changed from the original copy

• Every Airtable base, table and record address is now a placeholder: YOUR_BASE_ID, YOUR_TABLE_ID,
  YOUR_BACKGROUND_TABLE_ID and YOUR_RECORD_ID. The original copy's Get Airtable Record Data pointed at a different
  table than the trigger and the save did. All three nodes now share YOUR_TABLE_ID, so they must be one table.
• The field list inside Post Caption into Airtable Record shrank to the four fields the node writes. The original copy
  held the full field list of its author's own table.
• The sticky note is rewritten and holds no personal links.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• An Airtable personal access token that can read and write your base.
• An OpenAI API key that can use gpt-4o.
• An Airtable base with two tables, described below.

The Airtable base

The editorial table is where you plan posts. It needs these fields:

| Field | Type | Used for |
|---|---|---|
| created_at | Created time | The trigger watches it for new rows |
| Briefing | Long text | What the post is about, read by the agent |
| SoMe_Text_KI | Long text | Where the caption is saved |
| Posten | Checkbox | Set to unticked by every run |
| Werbeanzeige | Checkbox | Set to unticked by every run |

The background table holds one record that describes your audience and wording: who the readers are, what tone to use,
which words you like or avoid. Its fields are free. The agent reads whatever the record holds.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Add an Airtable credential with your access token. Select it in Airtable Trigger: New Record,
   Get Airtable Record Data, Background Info and Post Caption into Airtable Record.
3. Add an OpenAI credential and select it in OpenAI Chat Model.
4. In those four nodes, replace YOUR_BASE_ID with your base's ID. It is the part of the base's web address that starts
   with app.
5. Replace YOUR_TABLE_ID with the editorial table's ID, the part that starts with tbl, in the trigger,
   Get Airtable Record Data and Post Caption into Airtable Record.
6. In Background Info, replace YOUR_BACKGROUND_TABLE_ID with the background table's ID, and YOUR_RECORD_ID with the
   ID of the record that holds your notes. A record's ID starts with rec.
7. Save, then press Publish. The trigger only polls while the workflow is switched on.
8. Add a row with a short briefing and wait a couple of minutes.

How a run goes

1. Airtable Trigger: New Record checks the editorial table every minute and fires on a row whose created_at is new.
2. Wait 1 Minute pauses for one minute, so whoever made the row can finish writing the briefing.
3. Get Airtable Record Data reads the row again by its ID.
4. AI Agent gets the row's Briefing as its message. Its system message asks for an Instagram caption with a clear
   call to action and nothing else around it.
5. The agent can call Background Info, which reads one record from the background table. OpenAI Chat Model runs
   gpt-4o, and Window Buffer Memory keeps the last five exchanges under the row's ID.
6. Format Fields puts the agent's answer in a field named SoMe Text.
7. Post Caption into Airtable Record updates the row by its ID. It writes the caption into SoMe_Text_KI and sets
   Posten and Werbeanzeige to unticked.

Change the look

• The caption's rules: the system message of AI Agent. It is written in capitals and names Instagram, so change that
  line for another network.
• What the agent knows about your readers: the record Background Info reads, or its table.
• The model: gpt-4o in OpenAI Chat Model.
• The wait: the amount in Wait 1 Minute. A longer wait gives a slower writer more time.
• The fields: Briefing in AI Agent, and SoMe_Text_KI, Posten and Werbeanzeige in
  Post Caption into Airtable Record.

Good to know

• OpenAI's deprecations page, read on 12 November 2026, lists the dated gpt-4o versions for shutdown on 14 January
  2027, with gpt-6-luna as the replacement. It does not list the bare name gpt-4o, which this copy keeps.
• The agent's prompt reads Briefing in two places and Briefing/Notizen in a third. Unless your field has that second
  name, the third comes out empty.
• Every run writes unticked to Posten and Werbeanzeige, so a box you ticked on that row is cleared. Your table needs
  both fields. Otherwise Airtable refuses the update, and you can remove them from the node.
• The trigger polls once a minute and the workflow waits another minute, so a briefing typed later than that is missed.
• Background Info reads one fixed record. The agent never picks a different one.
• OpenAI bills a model call for every step the agent takes, at least one a run, and the briefing and the background
  record both go to OpenAI.
• Anyone who can add a row to the table starts a run that spends your OpenAI key.
• The caption is only saved in the row. Nothing is posted to Instagram.
• Window Buffer Memory keeps its history in the n8n process. n8n drops a history after an hour without use, and a
  restart clears it.