How an AI audio edition actually works (the article, not the plugin)

The plugin sold you a player

Type audio edition, newsletter audio, or listen to this article and the pages that come back are mostly the same product: a widget that reads whatever is currently in the page. Those tools are not wrong about their own job. They are incomplete if the next thing you actually need is a narrator you can keep when the CMS changes, and a file you can still rewrite when the lede does.

An audio edition is the article, spoken, in one voice, ready when the writing is. Collapsing that into a plugin is convenient until the related-posts rail gets narrated, a proper noun lands wrong, or you want the same masthead voice on the newsletter, the site, and a private feed.

The product page for this job is Articles to Audio. This post is the craft version: the three jobs hiding under one search, what we return as a file, and what Substack and Beehiiv actually take. How to write the spoken column already lives in how to write a voiceover script. A daily three-item digest is a different clock; that lives in News Briefings. If you meant the book, not the article, how to make an AI audiobook already prints the ACX front door.

Three jobs, not one generate button

People type the same phrase for three products. Mixing them up is how a newsletter gets a fake two-host recap, or a help-center article gets treated as a screen reader.

A newsletter edition is still a script

The Articles to Audio worked example is this morning's lede: three spoken lines, one narrator, ready with the issue. House math on that page is about 9,000 characters in a 1,500-word article, about nine minutes of audio at about 1,000 characters a minute. Treat it as an estimate. Speed varies by voice.

Paste the finished article, not the live page. Strip captions, pull-quotes, bylines, and the related-posts rail. Those strings read as junk out loud. If a sentence is hard to say, cut it before you generate. We already wrote that fight in how to write a voiceover script.

Lock one voice and keep it. Readers learn a narrator the way they learn a masthead. Switching voices every Tuesday is how an edition starts feeling like a different publication.

A generation in Text to Speech has a character cap (30,000, about half an hour by house math). A typical newsletter fits in one pass. A long-read that does not should be split at a heading, not truncated and blamed on the voice.

A listen button is still a file

If the writing lives on a site you control, the honest product is still the file. Export it. Host it next to the article. Put a player on the page. When paragraph four changes, regenerate that section and swap the file. Stale audio is worse than no audio: it hands the listener yesterday's instructions.

We do not sit inside WordPress, Ghost, or your theme. We do not scrape the live DOM. We do not skip the related-posts rail for you. That is the whole difference between a file you own and a widget that reads whatever the template currently prints.

An audio edition is also not a screen reader, and it does not make a page meet WCAG. Conformance lives in the markup. We already said that on Accessibility and we will not overclaim it here. Offer the file as one more way to take the article. Do not advertise it as the accessibility strategy.

Same article, two engines

The usual failure is directing a draft, or drafting a keeper.

Write the words on Kokoro until they are actually the words. It is the unlimited draft engine: a clean, plain read, fast enough that rewriting a name is cheaper than arguing about the take. Daily publishing is bought on that trait. The Articles to Audio recommendation is Kokoro for a reason.

For a weekend feature that wants a more characterful narrator, audition Grok Voice: five English personas, a confident take that still has to survive being the same voice next Sunday. Zonos holds American and British voices when the edition's accent is the brand, not the emotion. A lede that needs a cue in brackets can move onto Gemini Flash, the engine on this roster that acts them. Most article jobs do not need that. The short definition of a cue is what is a TTS cue.

If you wanted two hosts chatting about the piece, that is a different product. We do not invent banter from a URL. The honest version of that argument is in how an AI podcast actually works.

Substack and Beehiiv's pages, not the rumor

Two different platform features get mashed into one internet sentence. A voiceover is an audio file you attach to a post. An auto-narrate toggle is the platform reading the post with its own voice. They are not the same form.

As of November 2026, Substack's voiceover help still describes an upload or an in-browser recording, with a toggle for "This is my narration of the post." One voiceover per article, pinned at the top. Their audio embed help is the headphones icon in the editor: upload a file, drop it where you want, including below a paywall. We are not Substack. Read their page. Verify before you rely.

As of November 2026, Beehiiv's audio block help still says type /audio, upload an MP3 or WAV (they also list ogg, flac, aac, and webm), and style the block. In email, the block shows a Play Online button that sends the reader to the web post. That is an uploaded file, not a widget scraping your archive. Same caveat: their live page wins.

Neither page is a ban on generated narration. Neither page is a promise that a plugin will keep the audio in sync when you edit paragraph four. If the words change, regenerate the file and replace it. Acceptable use still forbids impersonating a real person. A roster narrator on copy you wrote is a read, not a fake byline.

Where it still fails

A fluent take can still be the wrong file. These are the misses we see when people treat the search phrase as a spec.

• Pasting the live page, related posts and all, then blaming the voice for reading the chrome.
• One forty-minute generate for a feature you will correct in paragraph four.
• Calling the MP3 a screen reader, or a WCAG fix.
• Asking the studio to install a WordPress plugin, scrape the CMS, or highlight words as they play.
• Leaving last week's audio on this week's correction.
• A two-host recap of the article (we do not offer that, and the podcast post already said so).

Honest limits

We generate files. We do not sit inside WordPress, Ghost, Substack, or Beehiiv. The output here is audio you own: commercial use, no watermark, same as the rest of the studio (you own what you make here).

No CMS plugin. No auto-sync. No word-highlighting player. No caption file. The console API tab is still labeled soon; we will not pretend a public render endpoint exists.

Cue-following is Gemini Flash only. The other engines on the roster read plainly. We strip brackets so they do not speak the direction. We do not pretend they performed it.

A generation has a character cap. Split a long feature at a heading. Do not paste a 12,000-word investigation into one box and call the truncation a voice problem.

Premium generation sits inside a flat monthly allowance (published free-tier premium budget: 10,000 characters, about 10 minutes by house math). Kokoro drafting stays unlimited so you can rehearse a name without burning keeper budget on every micro-edit. Paid plans raise the flat allowance; they do not put you on a per-character anxiety meter.

Voice cloning is not a live path today. When it ships it will be English-first with a consent attestation. Until then, roster voices and personas are the honest path.

Cut one this week

The shortest proof is this morning's issue, not a roundup of widgets.

• Pick one finished newsletter or article. Count it.
• Strip captions, bylines, and related links. If a sentence is hard to say, cut it.
• Draft the words on Kokoro until they are the words.
• Listen to the lede without looking at the page. If a name is wrong, respell it and regenerate that section.
• Export the MP3.
• Upload it as a Substack voiceover, a Beehiiv audio block, or a player next to the article you already host.
• When the text changes, replace the file the same day. Do not leave yesterday's read on today's correction.

Open the studio, not the widget

If you came here from an audio edition or newsletter audio search, the honest product is a spoken file you can hear this afternoon. Open Articles to Audio, or start in Text to Speech if you already have the issue.

If you came for a help-center read-aloud, the honest scope is on Accessibility. If you came for a 7 a.m. digest, that clock lives on News Briefings. When you want the full studio, create a free account. Written to be heard includes the edition next to the writing, as long as we do not lie about what that file is for.