Build or Break: What Search Looked Like in Week One

Build or Break is a weekly interview podcast that launched on September 3, 2026 with three episodes. We were hired in late August to make it findable. The site was already built in Webflow, the design was finished, and the launch date was fixed. So the job was everything search needs that a finished design does not give you.

The starting position

The show's name did not appear as text anywhere on its own page. It existed in the logo and in images, all with empty alt text. The page title was “Podcast Episodes”. There was no sitemap, no canonical tags, no share image, and no structured data anywhere on the site. Dozens of article URLs left over from an old WordPress version returned errors.

A search for the show mostly returned other podcasts the host had guested on. That is the predictable outcome rather than bad luck. A page that never states its own subject in text is a page a search engine has nothing to match.

What we did

First, without touching the design: the show's name and the host into real text, title tags and descriptions written per page, canonicals, and a sitemap. The structured data and the redirect map for the dead URLs are written and validated but not live yet, because publishing to the site runs through the client's team.

Second, a page per episode. The guest's name in the title, a summary, the full transcript, chapter titles with timestamps, the trailer, and links out to every platform it streams on. Three went live with the launch.

From episode four onward we build each page from the transcript the team drops into a shared tracker, fetch the platform links ourselves, and stage it through Webflow's API. The week that pipeline went live it caught a dead video button on a published page about an hour after it went up, because the tracker held a stale video ID.

What happened in search

Search Console started collecting five days before launch, so there is a real before and after rather than a launch-day snapshot. Every figure below comes from seventeen days of it, twelve of them after launch, and the run rates are taken from the settled period rather than the launch spike.

The hub page we rebuilt sits at an average Google position of 3.8, and 11.0% of the people who see it click through, against 1.7% across the rest of the site. It earns clicks at about six and a half times the site's own rate, and it takes 49% of every click the site receives, 54% counting the three episode pages. On the show's own name it is the first result, position 1.1, and more than a third of the people who see it click it.

Before launch that same page was live and indexed and earned nothing at all from its appearances. So the launch is what created the demand, and we will not pretend otherwise. What the search work decides is who captures it: the show's own page, or a platform and a press release.

The part that interests me most is the AI answers. In those seventeen days the podcast pages were cited in roughly a quarter of all the times Google's AI answers drew on the site, from four pages out of thirty-two. Site-wide, citations inside AI answers rose 118% after launch.

What those numbers do not say

Launch week brought national television and trade press. That coverage is why the site was shown in Google as often as it was. We did not create that demand and will not claim it.

What search work controls is position, and whether people pick you when they see you. A week earlier the same attention would have gone to the platforms and the press release, because the site had no title, no description, and never said the show's name in text.

Two weeks of data is a short read however good the numbers look. The redirect import for the dead article URLs is also still outstanding, and it shows: several of those old addresses are being cited inside Google's AI answers right now, which is attention landing on a page that no longer exists.

The part that argues for transcripts

We asked ChatGPT two questions. Who hosts the show and where can I listen: it answered correctly and cited the platforms and the press release. The show's own site was not among its sources.

Then we asked what a specific guest had said about a specific idea. It quoted the interview in detail and cited our episode page by the exact chapter title we had written, naming the timestamp. That chapter title exists in one place on the internet, which is the page we built.

Platforms win “where do I listen”. A transcript page wins “what did he say”. Audio cannot be quoted, and the answer engines are increasingly in the quoting business.

Search works the same way for a podcast as it does for a property manager, and so does the work.