A landing page is a list of claims

Yesterday we shipped an argument page for our product — the “why this matters” narrative — four times. Not four drafts: four public versions, each one torn down by review and rebuilt stronger. By the end, the page had survived a falsification pass, a recency audit, a citation verification run, and a consistency sweep. The surprise was not that the first version had flaws. It was how much better the page got once we treated it like a research artifact instead of marketing copy.

Round one was written the way most landing pages are written: a plausible chain of reasoning with statistics attached. Review immediately found a structural contradiction — one step of the argument quietly assumed the opposite of what the product does. Plausible prose hides this kind of thing; a falsification pass (deliberately trying to disprove each step, the same skill we use on strategy claims) surfaced it in minutes.

Round two hit a different failure: the evidence was real but old. Citing decade-old classics for a “this is happening now” argument is a quiet credibility leak. The fix was mechanical once named — default to primary research from the last five years, and when an older classic carries the load, pair it with a recent replication.

Round three was the humbling one. We ran every reference through a citation verifier — authors, year, journal, volume, DOI, checked against the registry rather than against memory. It caught us misattributing a recent study to the wrong team. Nothing in the prose looked wrong. The error was only visible at the level of bibliographic fact, which is exactly the level a skimming expert reader checks first.

The lesson: a landing page that argues is making claims, and claims deserve the same pipeline as any other knowledge we publish — falsify, verify, then polish. The persuasive register does not exempt a page from epistemic discipline; it raises the stakes, because a public page is the version of your reasoning that strangers will fact-check. The mechanism that made this cheap was converting each review objection into a named, reusable audit instead of a one-off edit. “This step contradicts the product” became a falsification run. “This research is old” became a recency rule. “Check the citations” became a verifier pass. The next argument page starts with all three.

One more thing the process produced for free: when every step of the chain had to survive scrutiny, the final step — “and therefore the product needs exactly these capabilities” — stopped being an assertion and became a derivation. The strongest version of the page was the one where the evidence, not the roadmap, appeared to write the feature list.