Lately it sounds like the voice/TTS isn’t accurately spelling things out that usually points to a change in how the voice model is handling phonetics, not your prompt logic.
Since both Vapi and ElevenLabs voices were fine before and now aren’t, it’s very likely one of their recent voice model updates changed how they render spelled‑out text. Sometimes certain voices handle sequences (like letter‑by‑letter) poorly after an update.
A quick way to confirm is to try a different voice again if it spells correctly, then the issue is specific to that voice model. If all voices are now doing it, there’s likely been a broader update on the TTS side that changed phonetic handling.
If you tell me which voices you’re currently using, I can suggest ones that reliably spell out letters correctly.
@lotalop