If the audio leaves out part of the post, reads something it shouldn’t, or pronounces words wrongly, the cause is almost always the content the plugin received or how OpenAI interpreted it. Check the fixes below, then click Generate New to create a new file. Audio never updates on its own after you edit a post.

Recent edits aren’t in the audio

Generation uses the last saved version of the post. Save or update the post, then generate again.

Whole sections are missing

• Page builder or theme content: only the main post content field is read. Text from builder modules stored elsewhere, custom fields, reusable theme sections and dynamic blocks (such as “latest posts” or a table of contents generated on page load) isn’t included. PRO users can use Replace Content to paste the text in.
• Text inside [tts ignore]: check for an unclosed [tts ignore], which can hide more than intended.
• Content that only appears through a shortcode: shortcodes are removed, not run, so text a shortcode would output (for example from a tabs or accordion plugin) isn’t spoken.

The audio reads things it shouldn’t

• Link addresses, code or captions: web addresses written as text are read (minus “https://www.”). Wrap these in [tts ignore]...[/tts] with PRO, or remove them from the text.
• The title, author or date: turn these off in AI Text-to-Speech › Settings › Content Customisations.
• Tweet addresses: if a Twitter/X embed couldn’t be fetched, its link can be read. Regenerate, or place [tts ignore] before the embed and [/tts] after it.

Words are mispronounced

Add a rule in AI Text-to-Speech › Settings › Custom Pronunciations or use [tts say=""] in the post (both PRO). If a pronunciation rule changes words it shouldn’t, remember rules also match inside longer words, so switch to the shortcode for that term.

Shortcodes like [voice] or [tts say] are ignored

• Replace curly quotes (“ ”) with straight quotes (“).
• Make sure every [tts ...] has a closing [/tts].
• Check the PRO licence is active on AI Text-to-Speech › Account.

Odd pauses, jumps or volume changes

Long posts are generated in several parts that are joined together, and each [voice] section is a separate part. OpenAI can deliver each part with slightly different pacing, so a small change at the joins is normal. Splitting at the end of paragraphs helps: very long paragraphs are more likely to be cut mid-sentence. If a take sounds wrong, generating again often gives a better result.

The wrong language or accent

OpenAI detects the language from the text. Mixed-language posts can confuse it. With PRO and the GPT-4o mini TTS model, add a line to Custom Voice Style Instructions such as “Speak in Spanish with a neutral Latin American accent.” Currency amounts are always converted to English words, so write them out in your language if that sounds odd.

Related

• What is included in the audio
• Fix how words are pronounced