Before AI narrates your thank-you video, check the names.

Imagine making a thank-you video for the volunteers who kept a local club open. The AI voice sounds warm and assured, then gets someone's name wrong. That is the word its owner will be listening for. ElevenLabs' existing pronunciation-dictionary guide documents corrections for names and places. It also warns that phoneme rules work only with certain models; others skip those rules and need spelling aliases instead. Saving a correction is not enough reason to assume the next voice will say it correctly. I'd make a short names-only preview before generating the full narration. Ask each person how they say their name, then let them hear the result in the chosen voice. Keep their real spelling in the visible credits; a pronunciation workaround belongs in the speech setup. These are production suggestions, not a tool test or a new launch. Would you rather record the names yourself, or have the AI voice read a version each person has checked?