What Happens to a Podcast Interview After Twelve Months?

What we already know bounds the answer

Two measurements published earlier in this series make the twelve-month question sharper than it first appears.

On attention: research examining 606 Al Jazeera English articles found that traffic to roughly 80% of them decreased monotonically during the first twelve hours, with separate work putting article half-life at a power-law distribution averaging 36 hours. Whatever a podcast appearance does for readership, it does almost all of it immediately.

On persistence: Pew Research Center found 38% of webpages that existed in 2013 unreachable a decade later, and that 9% of all losses occurred because an entire root domain stopped functioning — taking every page with it.

Between those two facts sits the actual question. The listening is over in days. The page is gone in years. What matters commercially is what happens across the twelve months in between, and nobody has measured it for podcast episodes specifically.

The gap between those two measurements is where almost all the commercial argument sits. Placement is sold on the premise that coverage is a durable asset, and the evidence available describes an attention spike measured in hours and a survival curve measured in years. Nobody has published the middle, which is precisely the period a buyer is paying for.

The four ways an appearance disappears

They are distinct, they have different causes, and a study that measures only one will miss most of the loss.

Delisting: the episode leaves a directory while remaining on the show's site. Truncated feeds are the usual cause — we found Security Now and TWiT exposing only ten episodes each, so older episodes simply fall out of anything reading the feed.

Page removal: the episode page stops resolving. This includes show closures, platform migrations and content management changes, which as documented elsewhere in this series removed thousands of articles from two newspapers that never closed at all.

Transcript loss: the page survives but the transcript does not, or the transcript file host changes. Our measurement found transcripts served from six different hosts across ten feeds, several of them platform CDNs rather than the show's own domain.

Access change: nothing moves, but the robots.txt does. A show that adds a blanket disallow makes its entire archive unreadable overnight, as one publisher in our sample has already done.

Worth noting that these failures are silent. None generates a notification, none appears in any dashboard, and a guest who does not re-check will simply never learn that the appearance stopped being findable. That is why the design measures quarterly rather than asking anyone afterwards.

Each needs its own check, because no single test catches all four.

The design

Take a cohort of episodes at publication, record everything measurable, and re-measure the identical set every quarter for at least twelve months.

At baseline: the episode URL, its server-rendered word count, whether a transcript exists and at which host, the presence of PodcastEpisode markup, the show's robots.txt rules, the feed's episode count, and the episode's presence in each directory. An archived copy is captured on day one, because a study of disappearance that does not archive its own subjects will lose its evidence.

At each quarterly window: re-fetch every URL, re-parse every robots.txt, re-count the feed, and record which of the four failure modes has occurred. Search Console impressions are recorded where the subject controls a property, to separate findability from mere existence.

The cohort deliberately mixes show types — shows that publish transcripts against shows that do not, self-hosted against platform-hosted — because the hypothesis worth testing is that durability tracks publishing infrastructure rather than show popularity.

Cohort size matters less here than duration, which is unusual. Because each episode is measured against itself over time rather than compared across subjects, even a few dozen episodes tracked properly for a year produces a usable survival curve — and a large cohort abandoned at six months produces nothing.

The archived baseline copy serves a second purpose beyond evidence. When a page later changes rather than disappears, comparing it against the capture shows what was removed, which distinguishes a quiet edit from an outright loss.

What the design predicts

Three predictions, stated in advance so they can fail.

First, that directory delisting will be the most common failure and the least noticed, because a guest who checks the show's website will find the episode intact while anything reading the feed no longer sees it.

Second, that transcript loss will exceed page loss, because transcripts frequently live on platform infrastructure the show does not control — and a hosting change that preserves episode pages can silently break every transcript link.

Third, that access changes will be rare but catastrophic when they occur, affecting an entire archive at once rather than individual episodes. One publisher in our sample already blocks all crawlers across a domain carrying 1,129 declared transcripts.

If those hold, the practical advice is not about choosing popular shows. It is about choosing shows whose infrastructure is unlikely to move, and archiving independently regardless.

Each prediction is falsifiable in a specific way, which matters because durability studies are easy to fudge. If transcript loss turns out to be rarer than page loss, the second prediction fails outright; if access changes affect individual episodes rather than whole archives, the third does. Stating them this way means the study can be wrong rather than merely descriptive.

Why this matters more than the citation question

Because durability is measurable now, and citation is not.

Every other protocol in this series depends at some point on systems whose operators publish no selection criteria. This one does not. Whether a URL resolves, whether a transcript downloads, whether a robots.txt permits fetching — these are facts with definite answers, checkable by anyone, repeatable quarterly.

It is also the question with the clearest commercial consequence. As Google states, a page must be indexed and snippet-eligible to participate in its AI features at all. An episode page that stops resolving, or a domain that starts refusing crawlers, has removed itself from every downstream system simultaneously — and as the robots.txt documentation makes clear, that is a fetching decision the publisher makes unilaterally.

So the twelve-month question is upstream of everything else. There is no point asking whether an AI cites an appearance that no longer exists in a readable form.

There is also an ordering argument for buyers. Durability is the cheapest thing to verify and the most expensive to discover late, so it belongs at the front of an evaluation rather than after a campaign has run.

What a guest should do regardless of the result

Archive on publication day. A public web archive capture plus a saved PDF costs minutes and is the only copy that does not depend on decisions by the show, its host or its platform.

Record the metadata separately: show name, episode title, date, URL and transcript URL. As documented elsewhere in this series, immigration adjudicators require title, date and author for published material, and those are precisely the details that vanish with a page.

Re-check annually. An appearance that mattered enough to book is worth ten minutes a year to confirm it still resolves, and a failure found early is sometimes fixable by asking the host.

Media Morpheus books podcast appearances, and the durable part of what a booking produces is a page on somebody else's infrastructure. We would rather say that plainly, and recommend archiving, than let a client discover in year two that the only record they had was a link.

The annual re-check costs less than the booking did.

Methodology and limitations

This report contains no new findings. It is a study protocol published in advance of execution, built on two previously published measurements: Castillo, El-Haddad, Stempeck and Pfeffer's analysis of 606 Al Jazeera English articles, and the Pew Research Center's 2024 link rot study. The intended design records a cohort of podcast episodes at publication and re-measures the identical set quarterly for at least twelve months across four failure modes, with an archived copy captured at baseline. Five limits apply. First, no results from this design are reported and none should be inferred. Second, the prior research cited measures news articles and general webpages, not podcast episode pages; it bounds the question rather than answering it. Third, quarterly sampling will miss transient failures that resolve between windows. Fourth, robots.txt records stated permission rather than crawler behaviour. Fifth, attributing any disappearance to a cause requires the publisher's cooperation, which the design cannot assume.

Conclusion

The twelve-month question sits between two findings already established: podcast attention is spent within days, and web pages disappear within years. What happens in between has not been measured for podcast episodes, and it is the most tractable question in this entire body of work, because every variable is checkable. Does the URL resolve, does the transcript download, does the feed still list the episode, does the robots.txt still permit fetching. The design tracks four distinct failure modes quarterly, predicts that directory delisting will be commonest and access changes rarest but most severe, and commits those predictions in advance. Whatever the result, the advice it produces is already available: archive on publication day, because it is the only copy that outlives someone else's decisions.

Frequently Asked Questions

Sources

1. peer-reviewed Castillo, C., El-Haddad, M., Stempeck, M. and Pfeffer, J., 2013. Characterizing the Life Cycle of Online News Stories Using Social Media Reactions, for the attention decay figures.
2. peer-reviewed Pew Research Center, May 2024. When Online Content Disappears, for webpage and domain-level loss rates.
3. official Google Search Central. Introduction to robots.txt, for what an access change does and does not control.