The audio is built from the post’s saved content, with an optional spoken intro made from the title, author and publish date. Before sending the text to OpenAI, the plugin strips out markup and anything that doesn’t make sense read aloud. You control the intro in AI Text-to-Speech › Settings › Content Customisations.

The spoken intro

With all three options on, audio for a post starts like this:

“How to Brew Better Coffee. By Jane Smith. Published on March 3rd, 2026.”

• Title: the post title.
• Author Name: the author’s display name. This is only added for the standard Posts post type, not pages or custom post types.
• Post Date: the publish date in “Month Day, Year” format.

The intro is skipped entirely when a PRO user chooses Replace Content for a post. Global custom text before the content (PRO) is spoken after the intro.

What is removed

• All HTML tags and block editor comments. The text inside them is kept.
• Images.
• All shortcodes, meaning anything in square brackets. The text between an opening and closing shortcode is kept and read, unless it’s inside [tts ignore] (PRO).
• The http://, https:// and www. parts of links written in the text. The rest of the address is still read.
• Straight single and double quote marks, which means apostrophes typed as ' are removed too (for example “don’t” becomes “dont”). The voice usually still pronounces these naturally.

What is rewritten

• Headings, lists and paragraphs are separated with extra spacing so there’s a natural break between sections.
• Currency amounts are written out in English words, for example $4.99 becomes “four dollars and ninety-nine cents” and £2M becomes “two million pounds”. Supported symbols and codes: $, US$, A$, AU$, C$, CA$, NZ$, £, €, ¥, ₹, USD, AUD, CAD, NZD, GBP, EUR, JPY and INR.
• Embedded tweets added with the block editor’s embed block (twitter.com links) are replaced with the tweet’s text, fetched from publish.twitter.com. If the lookup fails, the tweet’s link address may be read out instead.
• Custom pronunciations (PRO) are swapped in across the content, title and author name.

Plain numbers, dates and times aren’t converted. OpenAI reads them itself, which normally works well.

Content that isn’t in the post body

Only the main post content field is used. Text added by your theme, page builder templates, custom fields, widgets or blocks that load content dynamically on the front end isn’t included. If a page builder stores its layout outside the normal content field, the audio may be empty or incomplete. In that case, PRO users can paste the text in with Replace Content.

Related

• Shortcodes reference
• Customise the audio content for a single post
• Audio skips content or reads things wrongly