Subscription Review Safe Workflow

For the method in Finding Subscription Drift Without Exposing Your Bank History. Never connect an AI tool to your bank account. You export, you redact, you cancel.

Not for

Autonomous cancellation, refund requests, or any agent action on a live financial account. You review the flagged list and cancel yourself, every time.

Before you start

Exported 12 months of transactions (CSV or clean copy-paste), not just the last month
Redacted full account and card numbers
Generalized sensitive merchant lines (clinics, legal services, benefits offices)
Checked the specific tool and plan’s current training, retention, access, and deletion settings; an incognito browser window alone does not change the provider’s data handling
Not uploading the raw export file — pasting redacted text only

Recurring-merchant table

Merchant (as shown)	Amount	Interval	12-month total	Confidence	Still used?
				☐ High ☐ Medium ☐ Low
				☐ High ☐ Medium ☐ Low
				☐ High ☐ Medium ☐ Low
				☐ High ☐ Medium ☐ Low
				☐ High ☐ Medium ☐ Low

Every low-confidence line checked manually against the actual statement: ☐ Yes

Decision questions (one per high-confidence subscription)

Cancellation verification checklist

For each subscription you decide to cancel:

Logged into the service directly (not via a generated link)
Cancelled from account settings, not just paused or downgraded
Confirmation email or on-screen confirmation received
Checked the next billing statement to confirm the charge actually stopped
Noted the cancellation date below

Merchant	Cancelled on	Confirmed stopped on next statement?
		☐ Yes ☐ No
		☐ Yes ☐ No

The two prompts to reuse

1. Group and flag recurring charges

Here are twelve months of transactions, with account numbers removed
and sensitive merchant lines generalized: [paste redacted list].
Group any merchant that appears 2 or more times at a similar amount
and a regular interval (monthly, quarterly, annually). For each
group, show: merchant name as it appears, amount, interval, total
paid over the twelve months, and a confidence flag (high/medium/low)
for whether this looks like a genuine subscription versus a
coincidental repeat purchase. Do not recommend cancelling anything —
just group and flag.

2. Frame the decision question

For each high-confidence subscription in this list [paste], write
one specific question I should ask myself before deciding whether to
keep or cancel it — based on frequency of use, whether a free tier
would cover my needs, or whether I forgot I had it. Do not tell me
which ones to cancel.

Running total

Quarter	Subscriptions cancelled	Estimated annual savings