You redesigned the homepage. New headline, new hero, a proof section you did not have before. The question that matters is not whether it looks better to you. It is whether the three problems the last review flagged are actually gone, and whether the redesign introduced new ones. Rerunning the same study answers both, and the second review is built to open with the fixes rather than bury them.
Here is the end-to-end flow, from the first teardown to the follow-up review that a teammate can skim in one scroll.
Step 1: Run the first teardown and read the ranked findings
Add Cobalt to the assistant you already use, then ask in plain words: "Run a website teardown of yoursite.com." The study checks what the homepage and key pages put in front of a cold visitor: who it is for, what it claims, whether it proves the claim, and how to start. It judges desktop and the phone view, and it works on public pages only.
The output is a review with real screenshots of each page it looked at, findings ranked by what they cost you, and the verdict up top. A typical first pass flags things like a hero that names a feature instead of a buyer, no proof near the main claim, and a call to action that does not say what happens next. That ranked list is your redesign brief. If you want to see what a first pass catches before you run your own, the worked example of a founder's homepage shows the shape of it.
Step 2: Fix against the ranked list, not your gut
Work top down. The findings are ordered by cost, so the top item is the one losing you the most. Rewrite the hero to name the buyer. Pull a real metric or a named customer up next to the main claim. Change "Learn more" to something that tells the visitor what the click does. Resist the urge to redesign everything at once, because a smaller change set makes the next review easier to read.
If you are handing the redesign to a coding agent, the review already carries clean markdown an agent can read, screenshots included. The findings become tasks without you retyping them. But this works just as well if a designer and a developer are doing the work by hand. The point is that the second review will measure against the same questions, so fix the things those questions ask about.
Step 3: Rerun the exact same study on the new build
Once the redesign is live on a public URL, run the same command: "Run a website teardown of yoursite.com." Same study, same questions, same page set. That consistency is what makes the two reviews comparable. A different study, or a different reviewer with different priorities, gives you noise instead of a before and after.
Because this is a public-page study, it runs start to finish from chat with nothing for you to do. Cobalt renders the pages itself, so you do not need an agent that drives a browser for this one. Walk away and come back to the review.
Step 4: Read the review that leads with what got fixed
This is the payoff. A rerun does not hand you a fresh wall of findings. It opens with what changed: the issues from the first pass that are now resolved, up top, before anything else. You see at a glance that the hero now names the buyer, that proof sits next to the claim, that the call to action reads straight. Then it lists what still needs work, and any new problems the redesign created.
That ordering matters for how you share it. When you forward the rerun to a founder or a stakeholder, the first thing they read is the list of problems you closed. The redesign gets credit in the first paragraph, and the open items are framed as the next round, not a failing grade. A positioning audit run twice becomes a progress report, which is a much easier thing to send up the chain than a static critique.
What the before-and-after actually shows you
Three things the rerun catches that a single review never can. It confirms a fix is real on the live page, not just in your head or in the design file. It catches regressions, where fixing the hero pushed the proof section below the fold on a phone, say, and now the mobile view has a new gap. And it gives you a clean record of the change, with screenshots of both states, that you can keep for the next redesign.
If you want the measurement angle in more depth, there is a companion piece on rerunning a teardown to see what improved. And if your redesign was mostly about the words, the notes format matters: a positioning audit a copywriter can rewrite from turns the findings into something a writer acts on directly.
What it costs to keep rerunning
You can run and reread teardowns with a free account, no password and no credit card. Cobalt asks for an email when a review is ready so it can send you the link. The one thing to know: after 7 days a review goes read-only for you, so editing and exports stop, though the shared link keeps working for anyone you sent it to. If you are running a redesign cycle over weeks and want every review to stay editable and comparable with no cutoff, Pro is $5 a month or $50 a year, reopens read-only reviews instantly, and cancels any time. The details are on the pricing page.
Ship the redesign, then run the same study again. The second review will tell you, in its first line, whether the work landed.