How Long Does an AI Podcast Take? Benchmarks From Our Generations

These numbers come from real customer requests to the AutoContent API, not from a lab test. We publish them because people who build podcast workflows ask the same questions: how long to wait, how often a request fails, and how long the episode will be. Every number on this page uses one definition, one window and the privacy rules in the method section.

Key numbers for podcasts

Time to file and success, by format

Time to file is the time from the moment the API receives a request to the moment the finished file is ready to download. The median is the typical request; the last three columns show the share of requests that were ready within a quarter hour, half an hour and one hour.

Not shown because they do not pass the privacy rules in the study window: Infographic, Short video, Quiz, Cinematic video, Data table, Briefing document, Text, FAQ. For current figures on every format, see the status page, which shows the last seven days.

How long is the podcast?

The API takes a duration setting of short, default or long for podcasts. The table shows the length of the finished audio file for each setting. The range column runs from the 10th to the 90th percentile.

The style setting (deep dive, brief, critique or debate) changes the kind of conversation. Styles that do not pass the privacy rules are left out: debate, brief, critique.

Podcasts by language

The language is the language value of the request. A language appears only when enough customers used it to pass the privacy rules. In the study window, 24 other languages had too few customers to show.

Podcast length by language, at the default length setting only, so that the settings do not mix:

What the numbers mean for your pipeline

• Do not block on the request. Start the job, store the request ID, and poll the status endpoint every 30 to 60 seconds, or set a webhook and let the result come to you.
• Do not set a timeout at the median. A timeout at the median would stop about half of your jobs while they are still running. In the study window, 91.5% of podcasts were ready within one hour. Use a webhook, or poll for longer, to get the rest.
• Send batches within your plan's concurrent limit. Requests above the limit wait in line, and that wait counts in the time to file. The pricing page shows the limit of each plan.
• Retry failures. A failed generation uses 0 credits, so a retry costs nothing extra.
• Choose the length setting by the length you need. The default setting gives long episodes; use short when you need a quick summary.

Method

• What we counted: every request that a customer sent to POST /content/Create from August 1, 2026 to October 31, 2026 (UTC). Our own test, review, preview and staff keys are not in the data.
• Time to file: the minutes from the time the API received the request to the time the request succeeded and its file was ready. Scheduled requests are left out. The time includes any wait behind the same account's earlier requests when an account sends more requests than its plan runs at once.
• Success: succeeded requests divided by succeeded plus failed requests. Requests that were still running when we pulled the data do not count.
• Length: the duration of the finished audio or video file.
• Percentiles: the median is the 50th percentile. Percentiles are interpolated between ranks, as SQL Server's PERCENTILE_CONT does.
• Privacy: every row comes from at least 10 customers and at least 200 generations, and no single customer makes up more than a quarter of a row. When one customer is above that share, we keep a fixed random sample of its generations. We read no prompts, sources, URLs or titles.
• Scope: this study reports outcomes only: time to file, success, length and language.
• How the outputs are made: Audio and video overviews, slide decks and infographics are made with Google's Gemini Notebook (formerly NotebookLM), on accounts that AutoContent operates. AutoContent has no agreement with Google. Google can limit or stop this access at any time.

Frequently asked questions

How long does an AI podcast take to generate?

In the study window, the median podcast made with AutoContent API was ready 10.6 minutes after the request, and 91.5% were ready within one hour.

How long is the podcast that the API makes?

In the study window, the median podcast at the default length setting ran 19.4 minutes. The length table above shows each setting.

Does a failed podcast request use credits?

No. A failed generation uses 0 credits.

Where can I see current numbers?

The status page shows success and time per format for the last seven days. This study covers a longer window and adds length and language.

Changelog

• November 8, 2026: first edition, with data from August 1, 2026 to October 31, 2026.

Related reading

• Does Gemini Notebook (NotebookLM) have an API?
• Webhooks, polling, retries and status for content generation APIs
• Developer quickstart
• Plans and credits