AI Product Marketing
Teardown

How to do a messaging teardown

To do a messaging teardown, write down why you think the page fails, then quote its copy word for word. Score the copy on eight reads, from who it is for to whether a rival could run it. Name one primary failure. Finish with replacement lines for what the diagnosis condemned, set beside the originals.

Last updated by Omair Dawood.

On this page

The nine steps

These steps, and the Eight Reads lens in step 4, are our own method, which the Teardown tool on this site runs (read 2026-10-13). The method treats steps 4 to 6 as the ones that produce the value.

  1. Set two rules for your own words. Use no em dashes in your prose, and write only claims that could be proven wrong. They bind the rewrite too.
  2. Write your opinion first, and date it. Write one or two sentences on why the page fails. The method will not start without them: it treats your judgment as the one part of a teardown a rival cannot fake.
  3. Quote the page, then check the markup. Record the URL and the date you fetched it. Copy the hero, every button label in order, the section headers and any proof, typos included. Then read the raw HTML. The method's reason: a summarising fetch can hand you an announcement banner as the hero. If the two disagree, the markup wins.
  4. Score Reads 1 to 7. Mark each Holds, Soft or Fails, with a quote under it. The seven are the Reader, the Category, the Claim, the Ask, the Clock, the Vocabulary and the Proof. The sample table below gives the question each one asks.
  5. Fetch two competitors, then score Read 8. Read 8 is the swap test: put a competitor's logo on the page and mark every line it could publish unchanged. It waits until both pages are fetched and dated, because the method calls a swap test from memory a guess.
  6. Write the diagnosis from your opinion. Name exactly one primary failure. Your opinion picks it, not the severity of the marks, so a Read that failed harder becomes a secondary. Then write the cost sentence: a specific visitor, where they came from, the moment they give up and what they do instead. The method fails "hurts conversion" on sight.
  7. Write the after as literal copy. Rewrite the hero headline, the subhead, the primary button and the one section that fixes the primary failure. Set each new line beside the one it replaces. Every new claim must be supportable from the company's own public site.
  8. Steelman the original. Write one paragraph on the best reason a competent team shipped this page. Check that the diagnosis still stands, and cut any sentence that mocks the company or its people.
  9. Assemble it, and get a review. Keep every URL and fetch date, and have someone else read it before it goes out.

Worked sample: Slack's homepage

"All your people and AI agents working together"

"Slack connects your team. Slackbot multiplies what they can do."

"Get started" · "Find your plan"

Read on . Source: Slack homepage

That is the hero. Further down, a section headed "Meet Slackbot: AI for the whole team" says:

"Slackbot knows your whole business and coordinates work across every connected app and agent. Now it works in channels too, so one person’s answer becomes the whole team’s."

Read on . Source: Slack homepage

Step 2, our opinion. Before scoring, we wrote: "The headline could sit on any team chat page. It says people and agents work together, but not why they should do it in Slack." Our view, given 2026-10-13.

Step 3, the markup check. The browser's text view ran the buttons together in capitals: "GET STARTEDFIND YOUR PLAN". The raw HTML holds two labels, "Get started" and "Find your plan", so we quote those.

What the copy does well. The Ask holds: "Get started" asks for little, and "Find your plan" links to the pricing page. The Proof holds too, because the numbers lower on the page carry footnotes naming their source.

The Eight Reads, in short.

Read: the questionMarkEvidence
1. Reader: who does the copy assume is reading?Fails"your team" and "your people" name no role
2. Category: what noun does it claim?SoftNo category noun in the hero; only the browser title says "AI Work Platform"
3. Claim: could its best line be proven wrong?FailsNobody would publish the negation of "multiplies what they can do"
4. Ask: has the page earned each button?Holds"Get started", then "Find your plan"
5. Clock: why act this quarter?FailsNothing in the hero gives a reason
6. Vocabulary: do internal names outrank plain words?Soft"Slackbot" appears before the hero says what it is
7. Proof: whose words carry the evidence?HoldsFootnoted numbers and a named customer, lower on the page
8. Swap: could a rival run these lines?FailsSee below

Read 8, the swap test. We picked two other team chat pages. The Teams H1 reads "Microsoft Teams: The AI-Powered Platform for Work" (Microsoft Teams, read 2026-10-13). Google Chat's hero reads "AI-powered chat for teams of all sizes" and "Agentic collaboration to help teams stay connected and get more done." (Google Chat, read 2026-10-13). Slack's H1 could run unchanged on either page. The Slackbot line names a behavior, an answer in a channel that reaches the whole team, and neither hero makes that claim.

The diagnosis. The primary failure is Read 8, because our opinion points there. Read 1 failed as hard, so it is the secondary. In our view, the cost looks like this: an IT lead who searched for team chat tools has the Teams and Google Chat pages open in other tabs. Nothing in Slack's first screen sets it apart, so they leave to compare plans on price. That reader is our sketch, not a claim about who visits Slack.

The fair reading. In our view, the best reason for this hero is reach: a broad line can speak to founders, team leads and IT buyers at once. We think the diagnosis still stands, because Teams and Google Chat could run that line too.

The after, hero only, built from the page's own claims. The new headline uses the one behavior neither rival hero claims.

  • Headline. Before: "All your people and AI agents working together". After: "Ask Slackbot in a channel, and your whole team gets the answer."
  • Subhead. Before: "Slack connects your team. Slackbot multiplies what they can do." After: "Slackbot coordinates work across every app and agent you connect. One person's answer becomes the whole team's."
  • Primary button. Keep "Get started". The Ask held, and the method rewrites only what the diagnosis condemned.

Common mistakes

  • Starting without an opinion. Write your one or two sentences before you fetch anything, or the method refuses to run.
  • Tidying the quotes. The method treats a paraphrased before section as evidence bent your way. Copy the page exactly, typos included.
  • Letting the worst mark choose. The method makes the hardest-failing Read a secondary when your opinion points elsewhere.
  • Writing advice as the after. "Lead with the outcome" is a note, and the method rejects it. Write the replacement line itself.

Questions people ask

How do you do a teardown of a website?

Take one public page you can open without a login and work in three parts: the page as it reads, a diagnosis and replacement copy. Quote it exactly, score it, name one failure and rewrite only what that failure condemns. This method reads only the words.

How do I know if my homepage messaging is working?

Run the swap test on your hero: fetch two competitors' homepages today and put their names on your headline. If it still reads true for them, in our view the hero is not yet saying anything only you can say. Then check Read 1: could a stranger name the role your page is written for?

Sources

Run your own with the Teardown tool

The tool asks for three things: one public page URL, your opinion on why it fails in a sentence or two (20 to 600 characters), and two competitors. Leave those blank and it finds two. You get the page's copy quoted as it reads today and the Eight Reads, each marked Holds, Soft or Fails. You also get one primary failure, what it costs, and replacement copy line by line. It takes about 2 minutes of typing.

Tear down a page and get replacement copy, line by line