You want a read on your pricing page: can a buyer tell what they would pay and which plan is theirs? You have two ways to get it from Cobalt Capture. One, you hand a URL to your AI and it runs a ready-made pricing teardown. Two, you open the page yourself, capture it, drop pins, and type or dictate your notes. Both end in a shareable review. They are good at different things.
What the agent-run teardown actually does
You add Cobalt to the AI you already use and ask in plain words, something like "Run a pricing teardown of yoursite.com/pricing." Then you walk away. The AI renders the pricing page itself (no browser driving needed for the public-page studies), works out the cost for a solo user, a team of 5, and a company of 25, checks the fine print, and looks at the phone view. It comes back with a review: real screenshots of each page it judged, findings ranked by what they cost you, and the verdict at the top.
The point here is coverage you did not have to produce. You did not calculate the three team sizes. You did not remember to open the mobile layout. The study runs start to finish with nothing for you to do, and it applies the same checklist every time. If you have no researcher on staff, that checklist is the value.
Where it stops: it judges public pages only, and it judges them against a fixed set of questions. It does not know your internal reason for a weird tier, or the sales call where a prospect got confused by one specific line. It gives you a thorough outside read, not your read.
What hand-capturing the page gives you instead
The hand-captured pricing teardown is the opposite trade. You control every item. You click Capture screen, the browser grabs the current frame, and you crop to the tier table, drop a numbered pin on the ambiguous "contact us" button, and say out loud why it loses you the buyer. Each screenshot becomes an item with your comment attached.
This is the move when you already know what is wrong and want to say it precisely. The agent ranks findings by cost; you rank them by what you know matters. You can mark the one line a prospect misread last Tuesday, pin it, and write the exact replacement you want. Nothing in an automated checklist does that, because it is not a general finding, it is your finding.
It also handles the case the agent cannot reach. Behind a login, mid-checkout, a draft tier not yet public: the agent-run study runs on public pages, but you can capture whatever is on your own screen. Dictation works in Chrome and Edge if you would rather talk than type; on Firefox you type instead.
Speed, effort, and who reads the result
The two differ most on where the effort sits.
| Dimension | Agent-run teardown | Hand-captured review |
|---|---|---|
| Your effort | One sentence, then nothing | You capture and comment each item |
| Coverage | Fixed checklist, every time | Exactly what you choose |
| Reaches private pages | Public pages only | Anything on your screen |
| Point of view | Outside read, ranked by cost | Your judgment, your ranking |
| Best when | You have no researcher and want thorough | You know what is wrong already |
Both produce the same shareable output, so the receiver side is identical. On Publish you get a short /r/<slug> link anyone can read, a clean /s/<slug> reading page, and exports as HTML, PDF, and Word. The same review is plain-text markdown at /r/<slug>/markdown, which is what a coding agent reads if you hand the work off to one. People with the link can comment on any item, and you can mark each comment resolved. A copywriter rewriting your tier labels, a cofounder skimming on a phone, an agent editing the page: all three read the same review.
How to pick between them
Ask what you are short on. If you are short on a trained eye and time, run the agent study: it applies a researcher's checklist you do not have and does the three-team-size math for you. If you are short on nothing but a clean way to say what you already see, capture the page yourself and mark it up.
Most teams end up doing both, in order. Run the agent teardown first for a broad, ranked outside read. Then, if a finding needs your context or sits on a page the study cannot reach, open that screen and do a hand-captured pass on top. The automated version is also worth re-running after you ship fixes, because a re-run leads with what got fixed. The hand-captured version is better for one-off, this-exact-line notes that will never be a repeatable check.
If full control is what you want, start with the capture flow. You can browse the full set of ways to capture and compare feedback to see where each flow fits, or go straight to start a new review and mark up your pricing page now. It is free to try with no password and no card; a review stays editable for 7 days and then goes read-only unless you are on Pro, which is $5 a month or $50 a year and reopens read-only reviews instantly.
Not sure a teardown is the right study at all? A usability pass can catch things a teardown misses, and a positioning audit covers the homepage rather than the pricing page. Pick by the question you actually need answered.