How Verbatim decides a podcast was spoken by a person

The Verbatim library standard, from The Art of Service, admits moments spoken by people doing the work, across 900+ business podcasts. After every daily refresh, each show is scored on six signals that it may be machine-made. Flagged shows are held out of Advisory briefs until a person reviews them, and shows judged machine-made leave the library and go on the free Excluded Shows Register. Gerard Blokdyk, founder of The Art of Service, reviews every exclusion.

Updated

The screen right now

Counted live each time this page loads. The screen runs after every daily refresh.

930shows screened
6flagged, held from briefs, awaiting review
2on watch, not held
0excluded after review
1cleared after review
16%of moments from a confirmed speaker

See the Excluded Shows Register · Report a show

Why this exists

synthetic episodes at a dollar each
“By automating the podcast production process with AI, from market research to idea generation to scripts to voices, they'd reduce the price to make an episode to a dollar or less, and the time to under an hour.”
♫
Said on Shell Game
Why it worked $1 per episode cost, under an hour production time Share this moment

Framework

The Spoken-Word Sieve

It shows three mesh layers that moments settle through: a named speaker, a human cadence, and a person's review. Synthetic items stay on top.

The Spoken-Word SieveIt shows three mesh layers that moments settle through: a named speaker, a human cadence, and a person's review. Synthetic items stay on top. Named speaker: someone is introduced by name early in the episode. Human cadence: real guests, real titles, a believable schedule. Person's review: every flagged show heard and judged by a person. The library: what reaches search, alerts, briefs, API and MCP.machine-made items stop hereNamed speaker: someone is introduced by name early in the episodeNamed speakersomeone is introduced by nameearly in the episodeHuman cadence: real guests, real titles, a believable scheduleHuman cadencereal guests, real titles, abelievable schedulePerson's review: every flagged show heard and judged by a personPerson's reviewevery flagged show heard andjudged by a personThe library: what reaches search, alerts, briefs, API and MCPThe librarywhat reaches search, alerts,briefs, API and MCPThe Spoken-Word SieveIt shows three mesh layers that moments settle through: a named speaker, a human cadence, and a person's review. Synthetic items stay on top. Named speaker: someone is introduced by name early in the episode. Human cadence: real guests, real titles, a believable schedule. Person's review: every flagged show heard and judged by a person. The library: what reaches search, alerts, briefs, API and MCP.Named speaker: someone is introduced by name early in the episodeNamed speakersomeone is introduced by nameearly in the episodeHuman cadence: real guests, real titles, a believable scheduleHuman cadencereal guests, real titles, abelievable schedulePerson's review: every flagged show heard and judged by a personPerson's reviewevery flagged show heard andjudged by a personThe library: what reaches search, alerts, briefs, API and MCPThe librarywhat reaches search, alerts,briefs, API and MCP
Moments settle through each layer; items with no person behind them stay on the top mesh.

What are the screening steps?

  1. Refresh. Each day the library takes in new episodes from 900+ shows.
  2. Score. Every show is scored on six signals: no speaker introduced by name, few named operators, date-only episode titles, a daily cadence with no named guests, one brand running three or more look-alike shows, and formulaic titles.
  3. Hold. A show that scores high is held out of Advisory briefs. Its moments stay visible on the site until a verdict, so a wrong score never silently removes a real show.
  4. Review. A person listens and decides: keep, or machine-made.
  5. Exclude. A machine-made verdict removes the show from search, alerts, briefs, the API and MCP, and lists it on the register.
  6. Appeal. A show can ask for a second look at support@theartofservice.com.

Why does a person make the final call?

Because the signals are only signals. A daily news show can have date-only titles and still be read by real people. A score is cheap to compute and easy to get wrong, and a wrong public label harms a real publisher. So the screen narrows the pile, and a person decides. The register lists only those decisions.

Frequently asked questions

Are AI-generated podcasts in market research data?

They can be, because synthetic shows are cheap to publish at volume. The Verbatim library standard screens every show after each daily refresh, holds flagged shows out of Advisory briefs, and removes shows a person judges machine-made from search, alerts, briefs, the API and MCP.

How does Verbatim screen shows?

Each show is scored on six signals the library already holds, such as whether any speaker is introduced by name and whether episode titles are only a date. The Verbatim library standard never excludes on a score alone: high scores go to a review queue, and a person decides.

Who reviews exclusions?

Gerard Blokdyk, founder of The Art of Service, reviews every show the Verbatim screen flags and every show a visitor reports. Only his verdict excludes a show and lists it on the Excluded Shows Register, with the date of the review.

Can a show appeal?

Yes. A publisher can write to support@theartofservice.com and ask for a second look. If the review finds the show is spoken by people, it is restored to the Verbatim library and removed from the Excluded Shows Register. Shows can also ask to be removed from Verbatim at any time.

Does Team include the check?

Yes. The Human-Voice Check applies to every Verbatim plan, and it matters most on Verbatim Team, USD 299 a month, because Team feeds the library into your own AI through the REST API and MCP. Excluded shows never reach those results.

Is the register free?

Yes. The Verbatim Excluded Shows Register is free to read with no sign-up, and so is this page with its live counts. Anyone can report a show from the register page; reports go to the same review queue as the screen's flags.

Only what people said

See the shows kept out, or report one you suspect.

See the register