TL;DR: An AI subscription audit is a method, not a number. List every charge including annual ones, note each renewal date, write down what each tool is uniquely for, find the overlaps, check the last 30 days of real use, test whether a free tier covers you, then cancel in order.
What is an AI subscription audit?
An AI subscription audit is a pass over everything you pay for that touches AI, ending in a decision per line item rather than a feeling about the total. Eight steps: inventory, renewal dates, the unique job of each tool, overlaps, real usage in the last thirty days, free-tier coverage, what breaks on cancellation, and the order to cancel in.
It takes about an hour the first time and twenty minutes after that. The output is a table you can re-read in ninety days, which matters more than the hour, because the reason people end up with six AI subscriptions is not extravagance. It is that each one arrived alone, on a different date, for a good reason at the time, and nothing ever put them on one page together.
Why is there no dollar figure in the headline?
Because we do not have one, and the ones in circulation are mostly invented.
The genre convention here is a headline number: most people waste this much, the average stack costs that much. We considered publishing one and could not source it. We have no survey of AI spending, we have not run one, and repeating someone else's would mean citing a study we had not read. So there is no number, and you should be mildly suspicious of any version of this article that has one without a linked methodology.
The number was never the useful part anyway. Your stack is not the average stack, and an average would not tell you which of your six tools to cut. A method does.
Is it safe to paste a bank statement into a chat tool?
No, not as it comes off your bank's export. Strip it first.
A raw statement line carries far more than a price. Depending on your bank it can include a partial or full card number, an account reference, a billing address, a merchant location, and a transaction ID that is unique to you. None of that helps the categorisation you want, and all of it is worth keeping out of a chat log.
Run this before anything else touches a chat window.
I am about to paste financial lines into a chat tool. Tell me what to remove
first.
Here is one sample line, with everything I intended to keep:
"""
[ONE LINE ONLY]
"""
List every element in that line that identifies me, my account, or my
location, and show me the same line with each of them removed and the
merchant, date and amount kept. Then give me the find-and-replace rules I
should apply to the rest of the file before pasting it anywhere.
Step 1: What are you actually paying for?
Pull twelve months of charges, not one. A single month misses every annual plan you are on, and annual plans are where the surprises live.
The first prompt is deliberately constrained. It is told to work only from the lines you supply, never to add a subscription it assumes you have, and to put anything ambiguous in an "Unclear" pile with a question attached rather than guessing at it. That constraint is the whole difference between a useful inventory and a plausible fiction.
You are auditing a list of card charges. Work ONLY from the lines I paste
below. Do not add any subscription you think I probably have. Do not infer a
price from your own knowledge of a vendor's pricing. If a line is ambiguous,
put it under "Unclear" and say exactly what you would need to identify it.
Lines (already redacted; no account numbers, no addresses):
"""
[PASTE THE CHARGE LINES: date, merchant string, amount, currency]
"""
Return four sections:
1. AI and software subscriptions. One row per line: date, merchant string as
written, amount, and your best reading of the product. Mark each reading
"confident" or "uncertain".
2. Probably not a subscription. One-off purchases, transfers, refunds.
3. Unclear. Every line you cannot categorise, with the question you would ask
me to resolve it.
4. Duplicates. Any merchant appearing more than once in the period, with the
dates, because that is either two plans or a double charge.
Do not total anything. I will do the arithmetic in a spreadsheet.
Here are twelve months of charge lines, oldest first. I am looking for the
ones that happen only once a year, because they never show up when I look at
last month.
Work only from these lines:
"""
[PASTE 12 MONTHS OF LINES]
"""
List every merchant appearing once or twice in the whole period with an
amount at least five times a typical monthly charge from the same merchant.
For each: merchant string, date, amount, and the month I should expect it
again. Flag anything you are unsure about instead of dropping it.
Card statements are not the whole picture. Four common hiding places, and none of them appears on the card line you are reading.
| Where it hides | Why the statement misses it | What to open |
|---|---|---|
| Mobile app store subscriptions | Billed as one lump under the store's name | The subscriptions list in your phone's account settings |
| Wallet payments | Merchant shows as the wallet, not the product | The recurring payments page in the wallet |
| Business or company card | Never touches your personal statement | Accounting export, filtered by vendor |
| Team plan owned by a colleague | You use it, someone else pays | Ask who the billing owner is |
I want to find subscriptions that never appear on my main card statement.
Give me a checklist of the places a recurring AI or software charge can hide,
covering: mobile app store subscriptions, web checkouts on a different card,
wallet payments, annual invoices paid from a business account, team plans
where a colleague is the billed owner, and anything bundled into a larger
service I already pay for.
For each, tell me the screen or page where the list of active subscriptions
lives, and what to do if that list looks empty but a charge exists. Where you
are not certain a screen is still named the way you remember, say so rather
than sending me somewhere that no longer exists.
Find the charges that look like a trial that converted.
Lines, oldest first:
"""
[PASTE]
"""
A converted trial usually looks like a small or zero charge, then a full
charge from the same merchant 7, 14 or 30 days later, then nothing else of
note. List every merchant matching that shape, with both dates. Then list
any merchant whose first full charge is also its only charge so far, because
those are the ones I have probably not decided about yet.
I may be paying for the same product twice through different channels.
Charges, exactly as they appear:
"""
[PASTE LINES FROM CARD, APP STORE AND WALLET, EACH LABELLED WITH ITS SOURCE]
"""
Group them by the product they most likely belong to, keeping the original
merchant strings visible. Flag any product appearing under two sources. For
each flag, tell me what to check inside the account to confirm it is two live
plans rather than one plan and one refund. Do not merge anything you are less
than confident about. Put those in a "possible, unconfirmed" group.
Step 2: When does each subscription renew?
A renewal date turns a vague intention into a deadline. It also decides the order you cancel in, because cancelling something three days after it renewed costs you a full period of nothing.
Turn this list into a renewal calendar. Work only from what I give you.
Subscriptions:
"""
[NAME | PRICE | BILLING PERIOD | LAST CHARGE DATE | CARD OR ACCOUNT]
"""
For each, return the next renewal date, how many days that is from today
([TODAY'S DATE]), and the last sensible day to cancel if I want to avoid the
next charge. Sort soonest first. If a last charge date is missing, put the row
in a "cannot date" list rather than estimating one.
Annual plans deserve their own line in your notes. They are cheaper per month and much harder to reverse, which makes them the right choice for a tool you have already proven and the wrong choice for one you are still testing.
Step 3: What is each tool uniquely for?
Write one sentence per tool, in a fixed form, and notice which sentences you cannot finish honestly.
For each tool below, write one sentence in this exact form:
"I pay for [TOOL] because it is the only thing I have that [SPECIFIC JOB]."
Tools:
"""
[LIST]
"""
Then, for any sentence where the job could plausibly be done by another tool
on the same list, rewrite it as: "I pay for [TOOL] because I prefer it for
[JOB], which [OTHER TOOL] can also do."
Do not soften this. The second form is the honest one for most tools and I
need to see which ones land there.
Preference is a legitimate reason to pay. A tool you like using gets used, and a tool you tolerate does not. But preference and necessity should be labelled differently, because when you get to the cancel list they behave differently under pressure.
Step 4: Which tools overlap?
Overlap is where the money is, and it is invisible until you draw it. Two prompts: one maps it, one tests it.
Map the overlap in this stack. Use only the capabilities I list. Do not add
features from your own knowledge of these products, because plans change and
mine may not include what you are thinking of.
Tools and what I use each for:
"""
[TOOL, then the two or three things I actually do with it]
"""
Return:
1. A table of capability against tool, marking each cell "primary", "also does
this", or "not used by me".
2. Every capability I pay for more than once, with a note on which tool I would
keep and why.
3. Every capability I pay for exactly once. These are the load-bearing ones.
4. Any capability I listed that none of these tools covers well.
Act as a substitution analyst. For each tool, answer one question: if it
vanished tonight, what would I do tomorrow morning?
Tools and jobs:
"""
[LIST]
"""
For each, give exactly three lines.
Replacement: the specific thing on my remaining list that would take over.
Friction: what gets slower or worse, in concrete terms, not adjectives.
Verdict: "clean substitute", "workable downgrade", or "no substitute".
Do not recommend buying anything new. The point is what I already pay for.
Some of what I pay for separately may already be included in something else I
pay for. Using only the plan inclusions I paste below, find the overlaps.
What I pay for and what each plan says it includes:
"""
[TOOL, PLAN NAME, INCLUSIONS AS WRITTEN ON THE VENDOR'S PAGE TODAY]
"""
List every case where one plan's inclusions appear to cover another
subscription entirely, and every case where it covers it partly. For each,
state which line of the pasted text you are relying on. If nothing in the
pasted text supports an overlap, say there is none rather than reasoning from
what these products usually include.
Step 5: What did you actually use in the last 30 days?
Not "do I use it". That question is answered by identity, and everyone passes. Ask for a specific task on a specific date.
Interview me about the last 30 days, one tool at a time. For each tool I name,
ask these four questions and wait for my answer before moving on:
1. What is the last specific thing you did with it? Name the task, not the
category.
2. Roughly what date?
3. Would you have noticed that day if it had been unavailable?
4. Was that a typical use or an unusual one?
Do not offer suggestions or reassurance. When I have been through every tool,
summarise each as "used, load-bearing", "used, replaceable", or "not used in
30 days", and quote the answer that decided it.
I cannot remember whether I used these tools last month. Give me a list of
places to look for evidence of my own usage inside each product: history
pages, export files, activity logs, billing usage meters, saved items, and
email receipts or notifications.
Tools:
"""
[LIST]
"""
For each, tell me what to search for, and what a genuine "no usage" result
looks like, so I do not mistake an empty search for an empty month. Where you
are not certain a feature exists in the current version, say so instead of
guessing.
If a tool has a usage meter, read the meter rather than your memory. Memory is generous about tools we chose.
Step 6: Would a free tier cover you?
Free tiers in this category are more capable than they were and they change without notice, which cuts both ways. The tier that was too thin last year may be enough now, and the one you signed up under may have tightened.
Do not let the model recall a vendor's limits. Paste them from the vendor's current page. Every price and limit in this category ages fast, which is itself the argument for re-auditing.
For each tool, I will paste the free plan limits from the vendor's own current
pricing or FAQ page. Do not use remembered pricing. If I have not pasted it,
say "not supplied" and stop.
Tool: [NAME]
Free plan limits as published today:
"""
[PASTE]
"""
My typical month:
"""
[WHAT I ACTUALLY DO, AND ROUGHLY HOW OFTEN]
"""
Answer three things:
1. Does the free plan cover my typical month? Yes, no, or "only if I change
how I work", and name the behaviour that would have to change.
2. Which single paid feature is the one I would actually miss?
3. What is the cheapest test that would settle it, and how long would it take?
For each tool, I want the option between "keep paying this" and "cancel".
Tool: [NAME]
Plans as published on the vendor's page today:
"""
[PASTE THE PLAN NAMES, PRICES AND LIMITS]
"""
What I actually use:
"""
[LIST]
"""
Is there a cheaper plan that still covers what I use? If yes, name it and say
exactly which of my uses it would cut. If no, say so plainly. Use only the
pasted plan text. If the page does not mention something I use, say "not
addressed on the page" rather than assuming it is included.
A shortlist of free options is a reasonable place to start for tools in our own category; we keep one at best free prompt managers, and the same logic applies to every other line on your list.
Step 7: What breaks if you cancel?
This is the step people skip, and the one that produces the regret. Cancellation is not only a saving. It is an exit, and exits have costs that do not appear on the invoice.
Before I cancel anything, I want to know what I lose. Answer from what I
paste, and write "not stated" wherever the vendor does not say.
Tool: [NAME]
What I have stored in it:
"""
[SAVED WORK, HISTORY, TEMPLATES, INTEGRATIONS, SHARED LINKS]
"""
What the vendor's help page says about export and cancellation:
"""
[PASTE, OR WRITE "not checked yet"]
"""
Return: what leaves with me, what stays behind, what breaks immediately for
other people, and the exact order of operations to get my data out before the
plan ends. If I wrote "not checked yet", your first line should be the
specific page I need to read.
I am on an old price for some of these, and cancelling means losing it.
For each tool, tell me what to look for in my own account and receipts to
work out whether my current price is still offered to new customers today:
the plan name on my invoice, the price on the invoice, and the equivalent
plan name and price on the vendor's current page.
Tools: [LIST]
Then give me a rule for deciding at what size of gap between my price and
today's price cancelling stops being cheaply reversible. Ask me what that
threshold should be rather than picking one for me.
If I cancel and regret it, what does coming back cost?
For each tool, tell me what to check before cancelling: whether my account and
data are retained after cancellation and for how long, whether my current
price would apply if I returned, and whether anything is deleted permanently
rather than archived.
Tools: [LIST]
Give me the question to ask support for each, in one sentence, so I have the
answer in writing before I cancel rather than after.
This is a team plan with seats I pay for every month.
Seats and last known activity:
"""
[NAME OR ROLE | SEAT TYPE | LAST TIME THEY USED IT, AS FAR AS I KNOW]
"""
Sort into: active, dormant but likely to return, and dormant with no expected
return. For the second group, write the one message I should send to find out,
under 60 words, that does not read as an accusation. For the third, list what
I need to check before removing a seat: shared content ownership, links that
would break, and anything only that seat can access.
Step 8: What order should you cancel in?
Score first, then sequence. The matrix below is the decision; the prompt after it does the scoring.
| Question | Yes | No | Unknown |
|---|---|---|---|
| Used in the last 30 days for a task you can name | Keep looking | Cancel candidate | Go and check |
| The only tool you have that does that job | Keep | Overlap candidate | Run step 4 again |
| Free plan would cover a typical month | Downgrade candidate | Paid is justified | Read the vendor page |
| Something breaks for other people | Cancel last, after telling them | Cancel freely | Ask before acting |
| Your price is better than today's | Keep or pause, not cancel | Cancel is reversible | Check the invoice |
Score each tool against these five questions. One row per tool, answer each
with yes, no or unknown, and never leave a cell blank.
1. Did I use it in the last 30 days for a task I can name?
2. Is it the only thing I have that does that task?
3. Would the free plan cover my typical month?
4. Does anything break for other people if I cancel?
5. Is my price better than what is offered today?
Tools and my notes:
"""
[PASTE MY ANSWERS FROM THE EARLIER STEPS]
"""
Then sort into Keep (yes to 1 and 2), Downgrade (yes to 1, yes to 3), Cancel
(no to 1 and no to 4), and Decide later (any row with two or more unknowns).
For every "Decide later", write the single question that would resolve it.
Argue against me. I am about to cancel the tools below and I want the
strongest case for keeping each one before I do.
Tools and my reason for cancelling:
"""
[TOOL, then REASON]
"""
For each, give the best argument for keeping it in under 80 words, based only
on the reason I gave. Then state what evidence would settle the argument. Do
not be encouraging and do not hedge every sentence. If my reason for
cancelling is sound and you cannot find a real counterargument, say "no case"
and move on.
Put my cancellations in an order that minimises regret.
Cancel list with renewal dates and monthly cost:
"""
[PASTE]
"""
Order them by these rules and show your reasoning per item:
- Anything renewing in the next 14 days goes first, because waiting costs a
full period.
- Anything with data to export goes after the export, never before.
- Anything another person depends on goes last, after I have told them.
- Where two items tie, the one I used least recently goes first.
Return a numbered list with a date I should act by for each.
Write the message I send to cancel or downgrade a plan that other people
are on.
Situation:
"""
[TOOL, PLAN, WHO ELSE USES IT, WHEN IT ENDS, WHAT REPLACES IT]
"""
Under 120 words. State the change, the date it takes effect, what those people
need to do before then, and where their existing work goes. No apology and no
justification of the budget. Give me a plain-text version and nothing else.
What does this audit say about Prompt Architects?
Applying it to ourselves, checked against our own pages on 29 August 2026.
Step 6, free-tier coverage. Our FAQ page states the Free plan includes "5 prompt enhancements per day, forever. No credit card required." If five a day covers you, the audit says stay on it. That is roughly 150 enhancements a month, which is more than many people use. One caveat worth knowing: /pricing currently renders no Free card at all, so the free plan's terms live on /faq rather than on the pricing page. We should fix that, and it is why the number above is attributed to the FAQ.
What the free plan does not cover, per the same FAQ: saved prompts, the personal context library, and image prompt generation. Free is enhancement only. If your reason for wanting the tool is a library you build up over months, the free plan will not give it to you, and you should know that before you sign up rather than after.
Step 2, renewals. Pro is $4.99 a month or $49.99 a year; Advanced is $9.99 a month or $99.99 a year; Team is $10 a month plus $3.50 per member for 2 to 20 members, with the owner seat free. Those are launch-promotion prices on our pricing page as of 29 August 2026, and like every price in this article they should be re-checked at the source rather than trusted from a blog post. There is a 14-day refund window on paid plans. The full plan-by-plan breakdown is in our pricing explainer.
Step 3, the unique job. Honestly stated: we are not the only thing that can restructure a prompt. You can do it by hand, and if you already write structured prompts by habit, the sentence you would end up writing is the second form from step 3, not the first. What we are actually for is not having to, and keeping the results somewhere you can find them again.
Step 7, what breaks. Cancelling drops you to the Free plan at the end of the billing period rather than deleting the account, per our FAQ. Your saved prompts are the thing to think about, because Free does not include saved prompts. Export before you downgrade, not after.
That is the audit run against us with the same questions as everything else. If it comes out "cancel", cancel. A tool that survives this test is worth more to us than one that survives inertia, and the same is true of the other five lines on your list. The same reasoning applies to per-prompt purchases, which we looked at separately in are prompt marketplaces worth it.
When should you run the audit again?
Every ninety days, plus once before any annual renewal.
The reason is not that your habits change quarterly. It is that this category's pricing does not sit still: free tiers get more generous, paid limits get restructured, plans get renamed, and features move between tiers. An answer that was correct in March can be wrong in June without you doing anything at all. That instability is exactly why this article carries no competitor prices, and why every prompt above tells the model to read the vendor's current page instead of its own memory.
Write me a re-audit brief to open in 90 days.
Today's decisions:
"""
[WHAT I KEPT, WHAT I DOWNGRADED, WHAT I CANCELLED, AND WHY, ONE LINE EACH]
"""
Produce a short document in three parts:
1. The decisions, restated in one line each, so future me does not have to
reconstruct the reasoning.
2. For each kept tool, the specific thing to check next time: usage, a price
change, or a limit I was close to.
3. Three questions to answer before renewing anything on an annual plan.
Keep it under 300 words. I will paste it back into a chat in 90 days.
Save the set you actually used. An audit you have to rebuild from scratch is one you will not repeat, which is the whole argument for keeping working prompts in a personal prompt library rather than in a chat history you will never find again.
The reusable AI subscription audit checklist
Copy this into your notes and run it top to bottom.
AI SUBSCRIPTION AUDIT — [DATE]
PREP
[ ] Export 12 months of charges, not 1
[ ] Redact: card digits, account refs, address, transaction IDs
[ ] Open a spreadsheet for all arithmetic
1. INVENTORY
[ ] Card statement lines categorised, "unclear" pile kept separate
[ ] Annual-only charges identified
[ ] Mobile app store subscriptions listed
[ ] Wallet and business-account charges listed
[ ] Team plans where someone else is billed
[ ] Converted trials identified
[ ] Duplicate payers flagged
2. RENEWALS
[ ] Next renewal date per tool
[ ] Last cancel-by date per tool
[ ] Annual plans marked separately
3. UNIQUE JOB
[ ] One sentence per tool, in the fixed form
[ ] Sentences that had to be rewritten as "prefer" marked
4. OVERLAP
[ ] Capability-by-tool grid filled in
[ ] Capabilities paid for twice listed
[ ] Substitution verdict per tool
[ ] Bundle inclusions checked against the vendor's own text
5. REAL USAGE
[ ] Specific task and date named per tool
[ ] Usage meters and history pages checked, not memory
[ ] "Not used in 30 days" list written down
6. FREE TIER
[ ] Current free limits pasted from the vendor's page today
[ ] Coverage verdict per tool
[ ] Cheaper paid plan identified where one exists
7. WHAT BREAKS
[ ] Export path confirmed per tool
[ ] Grandfathered price checked against today's price
[ ] Return cost and data-retention answer in writing
[ ] Other people's dependencies listed
[ ] Team seats reviewed
8. ORDER
[ ] Five-question matrix scored, no blank cells
[ ] Keep / downgrade / cancel / decide-later sorted
[ ] Best case for keeping heard before cancelling
[ ] Cancellations sequenced by renewal date and export need
[ ] Messages sent to anyone affected
CLOSE
[ ] Re-audit brief written
[ ] 90-day reminder set
[ ] Prompts saved for next time
The honest bottom line
The reason a subscription stack drifts is not weak willpower. It is that nothing ever forces the six decisions onto one page at the same time, and each one alone is defensible. An audit is just the thing that forces the comparison.
Run the eight steps, let a model do the sorting and the awkward questioning, keep the arithmetic in a spreadsheet, and check every price at the vendor's own page rather than in an article, this one included. Then apply the same five questions to us. If our free plan is enough for your month, do not upgrade. That is the same standard this page has spent two thousand words asking you to hold everyone else to.
Stop rewriting prompts. Start shipping.
Works with ChatGPT, Claude, Gemini, Grok, Midjourney, Ideogram, Veo3 & Kling. 5.0★ on the Chrome Web Store.
Create An Account