Tweet drafts on a random hashtag saved to Airtable by AI

Tweet drafts on a random hashtag saved to Airtable by AI

An n8n workflow that writes one tweet draft each time you run it. A short code step picks one of two hashtags at random. OpenAI writes a tweet of under 100 characters about it, and the workflow adds a row with the hashtag and the tweet to an Airtable table. Nothing is posted to X.

What is inside

• workflow.json: the workflow as n8n exports it, 5 nodes. All five run, and there are no sticky notes. It holds no credential.
• README.md: these notes.

What we changed from the original copy

• The text step was an HTTP Request node that posted to OpenAI's old v1/engines/text-davinci-001/completions address with a header credential. OpenAI's deprecation page, read on 12 November 2026, lists text-davinci-001 as shut down on 4 January 2024, and the replacement it names, gpt-3.5-turbo-instruct, is itself listed as shut down. The step is now the OpenAI node's message action with gpt-6-luna, the small model that page names for gpt-4o-mini. The node reads the tweet from message.content, where the file read choices[0].text.
• The file set a temperature of 0.7 and a limit of 64 tokens. This copy sets neither: OpenAI's reasoning guide says hidden reasoning tokens count toward the output limit, so a small cap can leave no visible text, and the model's page lists no temperature setting.
• The prompt is reworded for a chat model, and it asks for the tweet text only.
• FunctionItem is replaced by a Code node named Pick Hashtag. n8n's source marks the old node as deprecated and hidden, and points to the Code node.
• Airtable was the first version of the node, which signs in with an API key. Airtable's page, read on 12 November 2026, says API keys stopped working on 1 February 2024. The node is now the current version, with a personal access token and a resource-locator for the base and the table. The file's base ID was its author's and is gone. You choose your base and table, and the file's table was named main.
• Set is rebuilt as the current Set node with the same two fields, and the credential links are gone.

Mindket read this copy against n8n's source. It was not run on a live n8n.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• An OpenAI API key.
• An Airtable personal access token, and a table with two text columns named Hashtag and Content.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Add an OpenAI credential and select it in Write Tweet.
3. Add an Airtable credential that uses a personal access token and select it in Airtable.
4. In Airtable, choose your base and your table.
5. Press Execute Workflow and check that a new row appears.

What happens in a run

1. On clicking 'execute' starts the run when you press Execute Workflow.
2. Pick Hashtag chooses #techtwitter or #n8n at random.
3. Write Tweet asks the model for a short tweet on that hashtag, with the hashtag included.
4. Set keeps two fields: Hashtag, from Pick Hashtag, and Content, the model's reply.
5. Airtable adds them as a new row, mapping the fields to columns of the same name.

What to change

• The hashtags: the list in Pick Hashtag.
• The tweet rules: the message in Write Tweet.
• The model: Write Tweet.
• The column names: the two names in Set, to match your table.
• Where the drafts go: replace Airtable with another node.

Good to know

• The workflow only drafts. Nothing is posted to X, so read each draft before you use it.
• The model does not see earlier drafts, so the same tweet can come back.
• Rows are added on every run, and the workflow has no schedule, so it runs only when you press the button.
• Data: the hashtag goes to OpenAI, and the hashtag and the tweet go to Airtable.
• Cost: one model call per run, billed by OpenAI.