A buyer lands on your pricing page with one question: what will this actually cost me? Most pages answer it for one kind of buyer and leave the other two guessing. A pricing teardown forces the question three ways, by working out the real monthly cost for a solo user, a team of 5, and a company of 25, then checking whether your page lets each of them arrive at a number without doing arithmetic in their head.
Here is what that looks like run against a made-up tool. Call it Flowmark, with a pricing page that shows three plans: Starter at $9 per user per month, Team at $19 per user per month, and a Business tier marked "Contact us." Billing toggle set to annual by default. I gave the task to my AI assistant in plain words: "Run a pricing teardown of flowmark.example/pricing." It rendered the page, ran the three calculations, screenshotted what it saw, and came back with a ranked review.
The solo buyer: can one person tell what they pay
For the solo persona, the calculation is one seat on the cheapest plan that is actually usable. Flowmark's Starter is $9 per user per month, so the surface answer is $9. The teardown caught two problems under that number.
First, the $9 was the annual rate, billed yearly at $108 up front. The monthly option was $14, a 55 percent gap that the page showed only after you flipped the toggle, which defaulted to annual. A solo buyer skimming the page reads $9 and budgets $9, then meets $14 or a $108 charge at checkout. That is the kind of mismatch that gets a card abandoned.
Second, Starter capped projects at three. The feature that the solo buyer came for, custom exports, sat on Team. So the real solo cost was not $9 or $14. It was $19 per month, because the cheapest plan that does the job is one tier up. The teardown flagged this as the highest-cost finding, because it changes the buyer's number by more than double and the page never says it plainly.
The team of 5: where per-seat math hides the total
Five seats on Team at $19 each is $95 a month, or $1,140 a year billed annually. Simple enough, except the teardown checks whether the page ever shows that total, and Flowmark's did not. It showed "$19/user" and left the multiplication to the reader.
The fine print caught here was a minimum seat count. Team required a floor of 5 seats even if you only needed 3, so a three-person shop pays for five. The page mentioned this in a footnote below the fold, in gray text. For the team persona this is the finding that reframes the decision: a four-person team is paying the five-seat price whether they like it or not. The teardown surfaces the floor, states the real monthly figure, and notes that the page buries it.
If you want to see how this kind of finding reads when it lands in front of a founder, the walkthrough on sending a pricing page teardown a founder will read shows the ranked format on a phone.
The company of 25: the tier that refuses to quote itself
Twenty-five users is where Flowmark's "Contact us" tier kicks in, and this is where the teardown earns its keep. The company persona cannot self-serve a number. The page gives no range, no starting price, no per-seat figure for Business, just a form. For a buyer evaluating three vendors in an afternoon, a tool that will not quote itself is a tool they skip.
The teardown does not pretend to know the hidden price. It reports exactly what the page lets a 25-person buyer learn, which was nothing, and ranks that as a real cost because an unquotable tier loses the deals that never fill out the form. It also checks whether staying on Team at 25 seats is viable: 25 times $19 is $475 a month, and if nothing forces the upgrade, the honest answer is that the buyer should stay on Team. The page did not say that either.
The fine print each pass tends to catch
Across the three personas, the recurring surprises cluster into a short list. The solo pass catches the annual-versus-monthly default and the cheapest-usable-tier jump. The team pass catches per-seat totals that are never summed and minimum seat floors. The company pass catches unquotable tiers and the missing "you can stay where you are" answer. The teardown also checks the fine print and the phone view on every pass, because a footnote that is readable on desktop often vanishes on a 390-pixel screen.
All three of these run from public pages with no browser automation, because Cobalt renders the pages itself. You describe the study in plain words and walk away. To run one against your own pricing, point your assistant at the pricing page teardown study and give it your URL. If you would rather read the mechanics first, running a pricing teardown from Claude chat covers setup, and the comparison with a hand-captured review explains when you want the automated study versus marking up screenshots yourself.
Connect Cobalt to the assistant you already use by pasting the server URL into its connectors, which the connect-your-agent guide spells out. The studies are free to try: unlimited reviews with no password, and a review stays editable for 7 days before it goes read-only unless you keep it on Pro at $5 a month. Run the three-persona pass once and you will know which of the three buyers your page is quietly losing.