Guard your sender reputation on cold outreach campaigns

Build an agent workflow that gates a cold outreach campaign so we never burn sender reputation on a junk list. The agent framing is right because the per-row decision combines multiple signals and the drafting step is open-ended.

Trigger: cron, every weekday (Monday through Friday) at 9am in the user's local timezone.

Step 1. Use Google Sheets Get Values to read pending rows from the 'outreach_queue' tab of the user's spreadsheet. The columns are: name, company, email, talking_point, status. Process every row where the status column is empty. Skip rows where status is already filled in (they've been handled in a previous run).

Step 2. For each empty-status row, call ZeroBounce Validate Email on the email address, then call ZeroBounce AI Email Scoring on the same address. Capture the deliverability verdict (status / sub-status) and the AI score (0 to 10).

Step 3. Decide what to do with the row based on the combined signals:

- If the ZeroBounce verdict is invalid, abuse, spamtrap, or disposable, do NOT send. Mark the row's status as 'skipped: bad_email'.

- Else if the AI score is below 8, do NOT send. Mark the row's status as 'skipped: low_score'. (Note to the user: 9 is a conservative cutoff, 8 is balanced. Pick one and stick with it.)

- Otherwise, draft a short, personalized cold email using the talking_point column for context. Keep it under 120 words, plain text, no marketing fluff, and end with a single soft call to action. Send it from the user's Gmail account using Gmail Send a Message.

Step 4. After each row is processed, use Google Sheets Update Values to write back to that row: the new status ('sent', 'skipped: bad_email', or 'skipped: low_score'), the ZeroBounce verdict, and the AI score. Leaving these reasons in the sheet matters because the operator may want to manually re-check borderline cases (e.g. a 7.9 score for a known good prospect).

Process rows sequentially so a failure on one row never blocks the others, and so the agent can log a clear per-row outcome. At the end of the run, log a one-line summary: how many rows processed, how many sent, how many skipped for bad email, and how many skipped for low score.