Free Resource · Checklist

The 10 AI Music Tells

How to spot AI-generated audio artifacts like fake metallic sound, hollowness, and flat dynamics. Free printable checklist for Suno & Udio creators.

· Works on Suno & Udio · By SebasKhz
← Back to Resources

You've generated a track. It sounds good almost. But something feels off. You can't always name it, but your gut says: this was made by AI.

That instinct is real. Over thousands of generations, Suno and Udio users have identified ten consistent artifacts; the "AI tells" ; that separate synthetic from human. This checklist gives you the vocabulary to name what you're hearing, so you can fix it before you release.

Print it. Keep it by your DAW. Check every track before you upload.

Download the Printable PDF

10 tells. One page. Checklist format with tick boxes.

Download Checklist PDF →

The 10 AI Tells

Check each box when you've verified the tell is NOT present in your track.

Fake Metallic Sound

A metallic, synthetic gloss on sustained notes; especially keys, pads, and synths. It sounds "too perfect," like a glossy photo filter on audio. Real instruments have micro-variations in timbre; AI smooths them out.

Frequency: 2-6 kHz persistent resonance
Solo the track. Sweep a narrow EQ bell at 3-5 kHz. If a harsh ring appears on multiple sustained notes identically → fake AI metallic sound.

Hollow Bass

Low end that lacks physical weight. No string friction, no cabinet resonance, no room interaction. It's a sine wave with envelope; not a bass guitar or synth played by human hands.

Sub-80Hz: flat spectrum, no harmonic richness
High-pass at 100Hz. If the track loses all "body" instantly → hollow bass. Real bass has harmonics up to 500Hz+.

Flat Dynamics

Compression that's baked into the generation. No push/pull, no breath, no human variability in note velocity. Every hit lands at the same intensity. It feels "stuck" at one volume.

RMS variance < 2dB across full track
Check the waveform. If verses and choruses have identical thickness → flat dynamics. Real music breathes.

Robotic Sibilance

Vocal 's', 't', 'sh', 'ch' sounds that repeat identically every time. Human sibilance varies by vowel context, breath pressure, and microphone angle. AI sibilance is a copy-paste artifact.

4-8 kHz: identical spectral shape on every 's'
Find three 's' sounds in the vocal. Zoom spectral view. If they overlay perfectly → robotic sibilance.

Quantized Groove

Drums and bass locked perfectly to grid. No push, no drag, no human micro-timing. The "pocket" is missing. It sounds like a MIDI file played back, not a performance.

Kick/snare deviation < 2ms from grid
Import into DAW. Enable "snap to grid." If everything aligns perfectly without nudging → quantized groove.

Unnatural Stereo Width

Elements hard-panned that would never be in a real room. Synth pads stretching 180°. Drums wider than a kit physically allows. Phase issues on mono summing.

Correlation meter drops below 0 on mono sum
Mono the master. If elements disappear or thin out drastically → phase/width issues. Check with a correlation meter.

Missing Transient Detail

The initial "crack" of a snare, the pick attack on guitar, the hammer strike on piano; softened or absent. AI generations often blur transients because the model predicts averages, not peaks.

Snare initial transient > 3dB quieter than real
Compare your snare transient to a reference track. Zoom waveform: real snare has sharp vertical spike. AI = rounded hill.

Vocal Formant Drift

Vowel sounds that shift unnaturally between words; the "AI accent." Formants (resonant frequencies that define vowels) should stay consistent for a given singer. AI models interpolate between training examples, creating hybrid vowels.

F1/F2 formant trajectories don't match human vowel space
Listen to "ah-oh-ee" sequences. If the vocal character morphs between words like different singers → formant drift.

Harmonic Sterility

Sustained notes with only perfect harmonics (2x, 3x, 4x fundamental). Real instruments have inharmonicity; stretched partials, noise floor, sympathetic resonances. AI harmonics are mathematically clean.

Spectral analyzer: only integer multiples of fundamental
Spectral view on a held piano/guitar note. Look for energy BETWEEN harmonic peaks. Silence between = harmonic sterility.

Structural Predictability

Every 8 bars: fill. Every chorus: same energy. Bridge appears exactly where expected. AI follows learned song templates. Human writers break rules; extend a phrase, cut a bar, surprise the listener.

Phrase lengths: all 4, 8, or 16 bars. Zero variation.
Map your song structure on paper. If every section is a multiple of 4 bars with zero deviations → structural predictability.

Download the Printable Checklist

One page. 10 tells. Tick boxes. Keep it in your studio.

Download Checklist PDF →

How to Audit Your Track in 15 Minutes

Use this workflow every time before you send a demo for finishing or upload to DistroKid.

  1. Listen on 4 systems: Headphones → Studio monitors → Phone speaker → Car/Bluetooth. AI tells reveal differently on each.
  2. Check low end first: Solo sub-80Hz. If it disappears or sounds like a pure sine → hollow bass (Tell #2).
  3. Analyze vocal sibilance: Find 3 's' sounds. Overlay spectrally. Identical = robotic sibilance (Tell #4).
  4. Inspect transients: Zoom waveform on drums. Rounded peaks = missing transient detail (Tell #7).
  5. Test mono compatibility: Mono sum. Elements vanishing = width/phase issues (Tell #6).

If you catch 3 or more tells, your track needs finishing. Free diagnostic tells you exactly what's fixable.