AI Mastering for Spotify: Complete 10-Step Guide

The right target for AI mastering to Spotify is -14 LUFS integrated, with a true peak ceiling of -1 dBTP, and a loudness range (LRA) between 8 and 12 LU for most genres. That’s it. That’s the whole game. Everything else in this tutorial is about getting there cleanly, without handing your music to an algorithm and hoping for the best.

AI mastering has gotten good enough to be a real option for independent releases. We’ve pushed tracks through LANDR, eMastered, RoEx, and Ozone 11’s AI assistant, and we can tell you exactly where each one earns its fee and where you need to step in. This is not a “set and forget” tutorial. It’s a hybrid workflow. You’ll get speed from the AI and control from you.

Step 1, Export Your Mix at the Right Specs

Before any AI touches your audio, your mix export needs to be clean.

Export as 24-bit WAV minimum. 32-bit float if your DAW supports it. Sample rate matching your session (44.1kHz for most releases, 48kHz if the session was built that way). Do not convert sample rates before exporting. Let the mastering stage handle that if needed.

Your mix’s integrated loudness should land between -18 and -23 LUFS before mastering. We know that sounds quiet. It is quiet. That’s the point. You need headroom for the limiter to work cleanly. If your mix is already sitting at -10 LUFS, the AI is going to struggle to add anything except distortion.

True peak on the mix should be no louder than -3 dBTP. Check this with Youlean Loudness Meter 2 (free) or iZotope Insight 2. Don’t skip this check. It’s frustrating to diagnose a bad master and trace it back to an overcooked mix.

Step 2, Set Up Your Loudness Meter for Spotify Targets

Open Youlean Loudness Meter 2 on a new session or your monitoring chain. Set the target to -14 LUFS integrated. Enable true peak display and set your ceiling warning at -1 dBTP.

Spotify’s desktop and mobile apps normalize playback to -14 LUFS by default on the “Normal” loudness setting. If your master comes in louder than that, Spotify turns it down. If it comes in quieter, some listeners may never hear the full version since they won’t manually switch to “Loud” mode.

Here’s the data that most tutorials skip: the top 25 tracks on Spotify average -8.4 LUFS. That’s nearly 6 dB louder than the platform target. Every one of those tracks gets turned down by roughly 5.6 dB on playback. Chasing that loudness is a waste. Master to -14 and let your dynamics do the work.

Keep Youlean open throughout the entire process. You’ll reference it constantly.

Step 3, Choose Your AI Mastering Platform

Pick based on your workflow, not hype.

LANDR at $12.49/month (Pro tier, unlimited masters) delivers a master in around 30 seconds using genre-matched models. It’s fast and the low-end translation is genuinely solid on pop, hip-hop, and electronic. We’ve used it on EP releases and had distribution-ready masters in under 10 minutes.

eMastered at $9/month is the budget option. It works. We love it for lo-fi and singer-songwriter material where the processing needs to stay transparent. Don’t expect deep control.

RoEx Automix at $9.99/month (unlimited downloads) or $24.99/month for WAV and HD WAV is underrated. It handles mix balance issues better than the others, which matters if your mix isn’t perfect going in.

iZotope Ozone 11 is the ITB option. The new Clarity module specifically addresses masking between low-mids and vocals, which is brilliant for dense arrangements. It costs more upfront but you own it. Check the manufacturer’s site for current pricing.

Step 4, Upload and Set the Genre Tag Correctly

This step is where most people leave quality on the table. Every major AI mastering platform uses genre tagging to route your audio to a trained model. Wrong genre tag means wrong processing decisions.

Don’t round up. If you’re making dark ambient, don’t tag it “electronic” because it feels closer to a genre with more training data. Tag it accurately. LANDR’s models for ambient and jazz process transients and low-end very differently from dance music models.

For metal: be aware that AI mastering platforms genuinely struggle here. The complexity of the low-end, the density of the high-mids, and the listener expectation of controlled aggression don’t always survive automated processing. We’d use AI as a starting point on metal and plan on spending time in Ozone 11 after, tightening the limiter manually and checking the low-end with a spectrum analyzer.

Step 5, Run the AI Master and Don't Touch It Yet

Let it process. Don’t adjust anything on the first pass.

We know the instinct is to grab the EQ slider before the master even finishes loading. Resist it. You need a baseline. The first AI master is your reference point, not your final output.

Download the master, load it into your DAW next to your original mix, and level-match them. Your loudness meter should tell you if the master hits around -14 LUFS. If it’s significantly louder, the AI over-processed. If it’s quieter than -12 LUFS, it under-limited.

Listen on at least three playback systems before touching anything. Headphones. Speakers. Phone speaker. Write down what you hear, not what you feel.

Step 6, Check True Peak and Fix Intersample Distortion

Run the AI master through Youlean and check the true peak reading. It must be at -1 dBTP or lower. This is non-negotiable for Spotify’s Ogg Vorbis encoder, which creates intersample peaks during conversion that weren’t in your original file.

If the true peak reads higher than -1 dBTP, you have two options. Go back into the AI platform and select the “streaming” or “conservative” limiting preset if one exists. Or run the master through a standalone limiter set to -1 dBTP ceiling with true peak limiting enabled. We use Limiter No6 (free) or Ozone’s Maximizer module for this.

Don’t skip this. We’ve seen -0.3 dBTP masters clip audibly after Spotify’s encoder touches them. Ugly. Avoidable.

Step 7, Check Your Loudness Range

Pull up Youlean and look at the LRA reading. Your target for Spotify is 8-12 LU for most genres. This keeps your music dynamic enough to breathe but controlled enough to survive normalization without sounding flat.

Heavily compressed electronic music and modern pop can sit at 6-8 LU. Jazz and classical benefit from 12-16 LU. Lo-fi sits happily around 9-11 LU.

If your LRA is below 6 LU, the AI over-compressed. You’ll hear it as fatigue within 30 seconds of listening. If it’s above 14 LU on a dense production, the master may feel inconsistent across playback systems.

Go back into the AI platform and reduce the “intensity” or “compression” parameter. Most platforms expose this. On LANDR it’s the mastering style slider. On eMastered it’s the intensity knob.

Step 8, Adjust EQ and Stereo Width if Needed

This is where the hybrid approach earns its name.

If the AI boosted the high-end too aggressively (common on LANDR for folk and acoustic material), pull it back. A 1.5-2 dB shelf cut starting at 8kHz usually fixes the harshness without losing air.

Check the low-end. AI platforms tend to add 1-2 dB below 80Hz on tracks that don’t need it, especially on streaming presets. A high-pass filter at 30-40Hz with a gentle slope cleans this up without thinning the bass.

Stereo width: the satisfying and dangerous part. A subtle widener set to add width only above 200Hz can help your master translate on headphones. We’d keep it under 20% added width. Beyond that, mono compatibility suffers and you’ll lose energy when Spotify collapses the signal on smaller speakers.

Step 9, A/B Test Against a Spotify Reference Track

Download a reference track from a similar artist via Spotify’s desktop app. Import it into your DAW. Level-match it to -14 LUFS using your loudness meter.

Now A/B your master against the reference at matched loudness. Use a plugin like Reference 2 by Mastering The Mix (check the manufacturer’s site for current pricing) or do it manually with volume automation.

Listen for frequency balance, not loudness. Your low-end should sit in a similar register to the reference. Your vocal (or lead element) should occupy comparable space in the high-mids. Your transients should hit with similar authority.

If the reference sounds noticeably fuller, your mix needed more work before mastering. No AI fixes a thin mix. Write that down.

Step 10, Export Your Final Master for Distribution

Export the final master as a 24-bit, 44.1kHz WAV. Most distribution platforms (DistroKid, TuneCore, CD Baby) accept this format. Some accept 16-bit but we’d always deliver 24-bit and let them handle the conversion.

Filename convention: “Artist_Title_Master_24bit_44k_v1.wav”. Version control matters more than it sounds. You’ll thank yourself when you’re staring at three files called “final_final_REAL_final.wav” at midnight.

TuneCore’s built-in AI mastering costs $5/song if you want a last-minute option during distribution. It’s basic. We’d use it only if you’ve missed a deadline and have no other option.

Pro Tips

Check your master on Spotify’s own player before release. Upload to a private SoundCloud or YouTube and enable normalization to simulate what Spotify will do. Some platforms show the integrated LUFS in analytics after upload.

For metal and heavily distorted material, use Ozone 11’s Clarity module before the AI stage. Feed a cleaned-up, de-masked signal to the AI and your results will be dramatically better. The Clarity module targets 200-500Hz masking, which is exactly where dense guitar arrangements turn to mud.

Keep a loudness log. We track every master we release: platform, LUFS integrated, LRA, true peak, and listener feedback. After 20 entries you’ll see patterns. You’ll learn which AI settings work for your genre faster than any tutorial can tell you.

Data privacy note: read the terms. LANDR, eMastered, and RoEx all use uploaded audio for model training unless you opt out (where the option exists). If you’re releasing unrefinished material or work under NDA, check the privacy policy before uploading. This is an area the industry has been frustratingly quiet about.

Common Mistakes to Avoid

Uploading a mix that’s already too loud. If your integrated loudness is above -14 LUFS before mastering, the AI has nowhere to go. It’ll over-limit, compress transients into nothing, and deliver a master that sounds worse than what you started with. Mix quiet. Master loud.

Ignoring true peak on the final download. The Ogg Vorbis encoding Spotify uses creates intersample peaks. A master at -0.5 dBTP will clip. Check every time.

Accepting the AI’s EQ decisions on acoustic instruments. AI mastering platforms are trained predominantly on pop and electronic releases. Acoustic guitar, upright bass, and live drums get over-brightened and over-compressed on nearly every platform we’ve tested. That’s annoying, and it’s correctable, but only if you listen critically.

Skipping the reference comparison. We’ve seen producers spend hours adjusting AI settings when a two-minute A/B test against a professional reference would have shown the mix itself was the problem. Always reference before mastering.