Can I run it over the records already in the base?

Yes. The package includes a manually run version that reads every record in a view where Category is empty, tags them in batches, and stops. Run it once, then leave the polling version on for new rows.

What if Claude picks a category that is not on my list?

The prompt tells Claude to choose from the list or return Unclear. Unclear sets Needs Review to yes and posts the record to Slack, so a person decides, and the list grows from real cases rather than guesses.

Does it work with Google Sheets instead of Airtable?

The same flow ships for Google Sheets as a second export: a row-added trigger, the same Claude step, and a write-back to columns on that row. The Slack step is unchanged.

How does it handle errors and rate limits?

Airtable allows five requests a second per base, and a polling trigger with single-record updates stays inside that. If the Claude call fails, the record is left with Category empty and the error is written to the execution log, so it is picked up on the next poll once the cause is fixed.

Does this run on Make or Zapier?

Both. Airtable's new-record trigger, an Anthropic or ChatGPT step, an Airtable update and a filtered Slack message are standard modules on each platform, and both versions are in the package.

AtCSMade to order

AI drafting, summaries & agents

Airtable new record → Claude tags + summary → Slack flag for review

Free-text records in Airtable get a tag, a one-line summary and a flag as they arrive.

Survey answers, interview notes, application essays, feedback forms: the long text ends up in an Airtable base and nobody reads it row by row. Point this at the table. Claude reads each new record's text field and writes back a category from your list, a one-line summary and a sentiment label. Records that match a rule you set, such as a complaint or a safety mention, are posted to Slack for a person to read the same day.

How it works

At
Trigger
New record in the table or view you choose, polled on a created-time field
Airtable
if
Step 1
Skip empty or already-tagged rows
Records continue only when the text field has content and the Category field is blank, so a re-run never re-tags a row a person has already edited.
C
Step 2
Read and classify with Claude
The text plus your category list and review rules go to Claude in one request. Back comes one category from the list, a one-line summary, a sentiment label and a yes or no on each rule.
At
Step 3
Write the results back to the record
Category, Summary, Sentiment and Needs Review are updated on the same record. The original text field is never touched.
if
Step 4
Check the review rules
Continues to Slack only when Needs Review is yes, which happens when a rule fired or when Claude returned Unclear instead of a category from your list.
S
Step 5
Post flagged records for a person
One message per flagged record with the summary, the rule that fired and a link straight to the Airtable record.

What changes once it runs

Each new record carries a category, a one-line summary and a sentiment label within one polling interval.
Categories come from your list, so the views and filters already in Airtable keep working.
Records that match a review rule reach Slack with a direct link; the rest are tagged quietly.
Rows already tagged are skipped, so a re-run or a batch import creates no double work.
The category list and the review rules are plain text in the prompt file, edited without touching a node.

What you get

Three workflow exports: the n8n original plus Make and Zapier versions of the same flow
Airtable base template with the fields named as the workflow expects
The Claude prompt as a file, with the category list and the review rules marked for editing
Written setup guide: the Airtable token, the trigger field and the Slack channel
A recording of one record being tagged and one being flagged
Email support on setup for the first thirty days

Before you start

Airtable base with a long-text field and empty Category, Summary, Sentiment and Needs Review fields; a base template is included
Airtable personal access token with data.records:read and data.records:write scopes on that base
A created-time field in the table for the trigger to poll on
Anthropic API key with a small prepaid balance
Slack app with chat:write, or an incoming webhook URL for the review channel

Built for

Questions before you buy

Can I run it over the records already in the base?: Yes. The package includes a manually run version that reads every record in a view where Category is empty, tags them in batches, and stops. Run it once, then leave the polling version on for new rows.
What if Claude picks a category that is not on my list?: The prompt tells Claude to choose from the list or return Unclear. Unclear sets Needs Review to yes and posts the record to Slack, so a person decides, and the list grows from real cases rather than guesses.
Does it work with Google Sheets instead of Airtable?: The same flow ships for Google Sheets as a second export: a row-added trigger, the same Claude step, and a write-back to columns on that row. The Slack step is unchanged.
How does it handle errors and rate limits?: Airtable allows five requests a second per base, and a polling trigger with single-record updates stays inside that. If the Claude call fails, the record is left with Category empty and the error is written to the execution log, so it is picked up on the next poll once the cause is fixed.
Does this run on Make or Zapier?: Both. Airtable's new-record trigger, an Anthropic or ChatGPT step, an Airtable update and a filtered Slack message are standard modules on each platform, and both versions are in the package.

Nearby on the shelf

Whole catalog