A podcast intro is thirty seconds that do a lot of work: they establish tone, brand the show, and set listener expectations before the first word of content. Recording one properly means a decent mic, a calm room, and an editing session, which is a lot of overhead for a fixed thirty seconds. AI tools collapse that overhead, but only if you chain them in the right order and accept what each one cannot do. Here is the process that actually works, with the honest limitations of each step.
Step 1: Write a script that sounds like you
Everything downstream inherits the script. A good intro script is short, concrete, and says what the listener gets and when. If you are stuck on wording, a general assistant like ChatGPT will draft options fast, but the editing judgment is yours, because the intro sets the voice of the entire show. Aim for 40 to 70 words, one hook, one promise, one sign-off. Read it aloud; if you run out of breath, it is too long.
Step 2: Generate the voiceover
ElevenLabs generates lifelike speech with control over emotion, pacing, and delivery across 128 languages, and its instant cloning can capture a host's tone from a short sample, which is the difference between a generic read and a branded one. Its limits: the free tier is limited and serious production requires a paid plan, and access requires a VPN in some regions. If you want an off-the-shelf voice instead, Murf AI offers a library of studio-quality voices with granular control over pitch, speed, pause, and emphasis, suitable for professional show intros with customizable tone and pace. Its free plan is capped at 10 minutes of generation, and voice cloning sits only on the Enterprise tier.
Step 3: Compose the music bed
The bed is where most AI intros sound cheap, so choose deliberately. Suno generates complete tracks from prompts in seconds and handles mood requests well, but tracks can sound generic without careful prompt refinement, and free usage is limited with commercial terms to check. Soundraw generates royalty-free instrumental music from mood, genre, and length parameters, and its section editing lets you trim the bed to exactly fifteen or thirty seconds, which is the feature that matters for an intro. Its limits: some regions require a VPN, and it is instrumental only, no generated vocals. Prompt for instrumental explicitly; lyrics in an intro bed almost never work.
Step 4: Mix voice and music
Descript is the fastest assembly tool for this job because it is built around the transcript: paste the script, generate the voiceover, drop the music bed underneath, and edit both by editing text. It removes filler words, clones your voice for overdubs, and keeps recording and editing in one workflow. Its limits are structural: it is not a music DAW, and complex multitrack music editing belongs in dedicated software, so treat it as assembly, not mastering. For the final polish, Auphonic analyzes the uploaded audio and automatically levels, denoises, and restores it, hitting loudness targets for podcast platforms. The free tier caps monthly processing time, and cloud processing means uploading before hearing results.
Production platforms if you want the whole show handled
Podcastle covers remote recording, AI-assisted editing, text-to-speech voiceover, and publishing in one browser workspace, which suits solo creators recording interviews from different locations. Heavy editing sessions can strain older computers, and the free tier caps recording time and advanced AI features. Adobe Podcast AI is a browser-based enhancement service that removes background noise, echo, and room reverb from recordings made on ordinary microphones, and offers speech-to-text transcription for show notes. Its free tier limits upload length, and cloud processing requires uploading audio, which may not suit sensitive recordings. For the mix itself, PodMix AI automatically balances levels between speakers, matches room tone, and normalizes loudness to broadcast standards, though its fixed automation leaves little room for creative mixing choices.
How to choose between them
The decision tree is short. For the voice: ElevenLabs if you want cloning and emotional range, Murf AI if you want a pre-built calm voice with no cloning. For the bed: Suno for full compositions you will curate, Soundraw for instrumental beds you can trim to exact length. For assembly: Descript for transcript-based editing of voice plus bed. For final polish: Auphonic for whole-episode leveling, PodMix AI for speaker balancing across a show. All of them are Freemium, so the whole intro can be prototyped for free.
A workflow that chains them
The full chain: draft the script, generate the voiceover with ElevenLabs or Murf AI, generate and trim the music bed with Soundraw or Suno, assemble voice over bed in Descript with the music ducking under the voice, then run the finished intro through Auphonic for platform-standard loudness. Render once, then reuse the same voice, bed, and loudness target for every episode so the show sounds consistent from episode one.
Common mistakes
First, music louder than the voice, the single most common intro failure. Second, a bed that is too busy; intros need space for the hook to land. Third, skipping the loudness pass, so your intro is quieter than every other show in the feed. Fourth, letting the generated voice drift in pace between episodes, which breaks the brand. And fifth, treating the free tiers as production quotas: voice generation limits are hit fast when you iterate.
FAQ
Can I clone my own voice for the intro? ElevenLabs supports instant cloning from short samples, and cloning your own voice is the standard way to keep a branded intro consistent.
How long should the music bed be? Fifteen to thirty seconds, trimmed to hit your edit points, which is why section editing in tools like Soundraw matters.
Will an AI intro sound professional? With a tight script, a calm voice setting, and a loudness pass, yes. The cheap sound usually comes from the mix, not the synthesis.
Do I need a paid plan? For a one-off intro, the free tiers are enough. For a weekly show with revisions, the voice generation caps push most creators toward paid plans.
Write it short, generate the voice, trim the bed, mix it quiet, and normalize the loudness. Thirty seconds, done once, reused every episode.
