I tapped the same chord into a few reverse-lookup tools to see what came back.
One handed me a wall of forty-odd names. One said "no chord found." One led with a name built on a note I wasn't even playing.
They couldn't all be right. So how would I - the person actually asking - know which one to trust?
That's the real question, and it's the one this post is about. Not "which tool is best." How you tell a good chord identifier from a weak one, using nothing but your own ears and a bit of theory. By the end you'll have a yardstick you can hold up to any of them, including mine. Especially mine.
TL;DR A good chord identifier tells you what you're playing. A weak one lists every chord your notes could theoretically belong to - a different question, and not the one you asked. This post hands you a seven-point rubric to judge any reverse chord lookup, lets you verify it in 30 seconds against the tool you already use, and - as supporting evidence, not the headline - shows a blind, reproducible test where two independent AIs scored seven tools against music theory without knowing which one was Fretscape. Both, scoring independently, gave it a perfect 94 out of 94 - and no other tool got there.
The difference that matters
A good chord identifier answers "what am I playing?"
A weak one answers a different question - "which chord definitions contain these notes?" - and hands you the list. For three notes, dozens of chords qualify. For four, some tools will give you forty. A guitarist looking at those notes isn't hearing forty chords. Listing every superset isn't help. It's noise wearing the costume of knowledge.
Identification and set-membership look similar from a distance. They're not the same job. One tells you the chord under your fingers. The other tells you every harmonic context those pitches could be borrowed into. Only the first is what you asked for.
A chord name should be a claim, not a possibility.
So here's the yardstick.
The rubric: seven things a good chord identifier does
None of these need the source code to check. Every one is something you can hold up to a tool and watch it pass or fail.
1. It identifies, it doesn't list. It names what you're playing - not every definition your notes are a subset of. A short, confident answer beats a long, hedged one.
2. Restraint earns trust. A curated, ranked set beats a wall of forty maybes. More names is a negative signal, not a feature. If a tool can't decide what to put first, it hasn't finished the job - it's handed the hard part back to you.
3. It doesn't fake knowledge. When the notes genuinely underdetermine the chord, a good tool names what's defensible and declares what's missing. A bare root-and-fifth is honestly a power chord. Three notes with the fifth left out are honestly a seventh chord with the fifth disclosed. And when there's truly nothing defensible to say, honest silence beats a confident label invented to fill the gap.
4. It doesn't invent notes to justify a name. A label that needs a note you didn't play doesn't belong on your voicing. If a tool reaches for a fifth that isn't sounding just to make its spelling work, it's describing a chord you aren't holding.
5. Right name first, honest alternatives behind. The exotic readings can exist - they just don't get to be the headline. The most likely name leads; everything else is clearly subordinate and clearly labelled.
6. Simpler spelling wins. Given two valid names for the same notes, the one a player would actually write on a chart beats the one that needs three contortions to arrive at. Am7, not G6/9sus4(omit5).
7. It names like a guitarist, not a database. Omitted fifths, slash voicings, rootless shapes - judged the way a player hears them, not the way a pitch-class table sorts them.
Two of those deserve a defence, because a sharp reader will say I wrote the rubric to win. Restraint (2) and simpler-spelling (6) aren't "theory-correct" - both Am7 and G6/9sus4(omit5) describe the same notes, and theory shrugs. So why prefer one?
Because of who's asking. A beginner staring at three notes is genuinely better served by Am7 than by a six-token spelling that's accurate and useless. The simpler name teaches; the contorted one obscures. That's not me bending the rules to come first - it's the whole reason a person taps notes into a tool instead of reading an interval table. The other five points need no defence. Theory referees those directly.
Verify it yourself in 30 seconds
Don't take my word for any of that. Here's a shape you can test right now.
That's 0 7 6 7 8 x - E, G#, D and G. It's E7#9, the Hendrix chord: the most recognisable single voicing in rock guitar, with the high G grinding a semitone above the G# to make the famous #9 crunch. The fifth (B) isn't in there - like most working dominant voicings, it drops the most disposable note.
Now paste it into whatever chord lookup you already use. Does it come back with E7#9? Or does it stall, relabel it as something it isn't, or bury the right answer in a list? Then tap the same shape into Fretscape's chord identifier and see what comes back.
Your tool is the control here. Theory is the referee. And you don't have to take my word that this one trips tools up - when I ran it in the blind test below, three of the seven stumbled on it: one returned "no chord found," one relabelled it as a minor chord, and one buried it in a wall of names. The most famous chord in rock, and nearly half the field fumbled it.
Deep dive - how the "scariest" chord name is also the simplest
Skip this if you just want chords that work.
Written out, E7#9 looks like a lot: a dominant seventh with a sharpened ninth bolted on. But for this exact shape it's the simplest honest label, because the voicing literally is that chord - root (E), major third (G#), flat-seventh (D), and the sharp-ninth (G) sitting a semitone above the third. Anything shorter would be throwing away information that's actually sounding.
There's one note missing - the fifth (B) - and that's fine. The fifth is the most disposable tone in almost any chord: the third decides major or minor, the seventh makes it dominant, the #9 is the whole point, and the fifth just confirms a centre your ear already assumes. So the honest name is "E7#9, fifth omitted": the working name a player writes, plus a note on what's technically absent. What you don't do is refuse to name it because the fifth is gone - or invent a B to make the spelling tidy.
Back to the story.
A few more, all checkable against a fretboard
The same rubric, on shapes you can verify with nothing but theory. I've hard-coded these as static examples - the names below are claims you can check, not live output.
x 0 2 0 1 0 is A, E, G, C and E. Those are the same pitches as a C6 chord - but the lowest note is an open A, and on guitar the bass does most of the work of telling your ear what it's hearing. The honest name is Am7. A tool that leads with C6/A is technically allowed and practically unhelpful. Point 7 of the rubric, in one shape. (I wrote a whole post about this exact decision - the wrong name, ranked first.)
x 3 2 3 3 x - C, E, B♭, D - is a C9 with the fifth left out, one of the most ordinary jazz and blues voicings there is. Guitarists drop the fifth constantly; it's the most disposable note in the chord. A good tool names it C9 and tells you the fifth is absent. A brittle one stalls on "no chord found" because one disposable note is missing - which is exactly what one tool in the test did, on this chord and ten others.
x x 0 2 1 x is D, A and C - root, fifth and flat-seventh, with no third anywhere. The major-or-minor quality genuinely isn't present. The honest name is D7(no3): here's what's defensible, here's what's missing. A weak tool does one of two things - invents a third to dress it up as a full D7, or refuses outright. The honest answer sits in between, and that gap is point 3 of the rubric.
Then I tested it - blind, and not in my favour
The rubric is the point. But a rubric I scored myself proves nothing, and I'm an unknown brand asking you to trust a yardstick I just handed you. So I ran a test designed to take my own opinion out of it entirely.
Here's the whole method, because the method is the receipt:
- I didn't pick the chords. Two AIs each generated 25 test voicings. I combined them and removed three exact duplicates, leaving 47 cases - a deliberately fair, tame set, not a gauntlet hand-built to embarrass anyone.
- I didn't vet the answers myself. Rather than mark my own homework, I had a separate AI check every reference answer before any tool was run. Corrections logged up front.
- Seven tools, identical inputs, each run in its most complete mode.
- Two independent blind judges - one ChatGPT, one Claude, fresh conversations, the tools anonymised as A to G in randomised order, scored against music theory. Neither judge was told any tool was mine. Fretscape's identity was revealed only after each one had delivered its verdict.
I didn't pick the chords, I didn't grade the chords, and I didn't tell the judges which tool was mine. The only thing I controlled was running the same 47 inputs through every tool.
What the judges said
This is the part I can't write in my own voice without it sounding like a sales pitch. So I'll let the judges talk. Anonymised, their words, no adjectives I added.
Both of them, independently, ranked the same anonymous tool - Tool G - the single most useful for a guitarist learning theory. They reached it from different directions and landed in nearly the same order:
| Blind rank | Claude judge | ChatGPT judge |
|---|---|---|
| 1 | Tool G | Tool G |
| 2 | Tool E | Tool E |
| 3 | Tool D | Tool D |
| 4 | Tool F | Tool A |
| 5 | Tool A | Tool F |
| 6 | Tool B | Tool B |
| 7 | Tool C | Tool C |
Top three identical. Bottom two identical. The only daylight between them was whether to put Tool A or Tool F fourth.
One judge noted Tool G was "the only tool with a perfect primary record" - it gave the correct lead name on all 47 cases - and that "it never makes you choose between 'correct' and 'well-organised.'" Both judges, without being shown my rubric, kept circling the same differentiator: honest omitted-tone naming. One quoted Tool G's own output approvingly - "strict: C9(omit5)… normally still heard and named as C9" - the working name and the technical caveat in a single breath.
Then the reveal: Tool G was Fretscape.
It's worth being precise about what that perfect record actually is. No tool won by being clever on one hard chord - on the easy ones everyone's right, and on the hard ones the good tools usually are too. Fretscape never once stood alone at the top of a single case. It won the other way: it was the only tool that never dropped a point anywhere - the only one without an off day across all 47. And both judges, scoring independently, arrived at the same number: a perfect 94 out of 94. Neither of them handed any other tool a perfect score.
A perfect record isn't forty-seven flashes of brilliance. It's the absence of a single slip.
The competitors stayed anonymous, and they will here too - but their failure modes, in the judges' words, map onto the rubric almost point for point:
- One tool refused eleven textbook-nameable voicings - "no chord found" on ordinary omitted-fifth dominants. (Rubric point 3: faked silence where an honest name existed.)
- One buried every answer in a 26-to-62-name unranked wall, often leading with a rootless reading over the obvious chord. (Points 1 and 2: listing instead of identifying.)
- One framed plainly omitted notes as inferred ones - quietly teaching that notes were present when they weren't. (Point 4: inventing notes.)
- Even the strong runners-up occasionally reached for contorted enharmonic spellings on the hard cases. (Point 6: the harder spelling where a cleaner one fits.)
I built the test to be fair to everyone but me - and two referees who'd never heard of Fretscape still picked it. That's the only kind of "best" worth publishing.
The honest concession
Here's the part most comparison posts leave out.
On the easy cases - an open C, a plain G, a basic triad - every tool got it right. The free ones, the brittle ones, all of them. The judges' rankings only pulled apart on the hard tail: the omitted-fifth dominants, the altered extensions, the sparse shells, the symmetric chords with several faces. On a plain major chord, you do not need Fretscape, and I'm not going to pretend you do.
That's a real concession, not a staged one. Theory was the referee and the sweep was honest - the common chords bunched everyone at the top, exactly as they should. Marking the boundary where Fretscape doesn't matter is what earns the right to talk about where it does.
Concise by default, deep when you ask
One thing the blind test slightly misrepresents, and I'd rather say so. In the test, Fretscape ran at full depth - every rootless, slash and deep-theory reading on show - so the judges saw its widest output. That's not what a beginner sees.
By default, the same voicing returns one clean primary name. Turn the depth on and the rootless and altered readings appear - but in their own partition, ranked and tagged, not flooded into the main answer. It's breadth and restraint in the same frame: the deeper readings don't fight the plain ones on a sliding scale, they live in a different room, and you decide which rooms are open. The reverse chord lookup post goes into why that partition matters more than it sounds.
So the wall-of-names tools and Fretscape-at-full-depth can surface similar amounts of information. The difference is what's first, what's labelled, and what's off by default.
Where even good tools break
One more thing, and then I'll let you go try your own shapes.
The test above was deliberately tame - a fair set, no trick questions. That was the point: judge the field on honest ground. But on the genuinely hard tail - the voicings where theory itself starts to argue with itself - the field fractures in ways a 47-case fair test only hints at. So I went looking, on purpose: where even a good chord finder breaks is the hand-picked deep-dive into exactly that, and it's less flattering to everyone, Fretscape included. This post was about the rubric. That one's about the edge.
The whole thing, in one line
A chord name should tell you what you're holding, ranked the way you'd hear it, with nothing invented to make the label look clever. That's the rubric. Hold it up to any tool - including this one.
The best way to test it isn't to believe my results. It's to tap in your own hardest shape - the one your usual tool fumbles - and see whether the name that comes back is a claim you can trust or a possibility you have to sort through yourself.
I'd rather be useful than impressive.
You can try your own in Fretscape's chord identifier - tap in the notes you're holding and see the name a voicing actually earns.
A few questions this raises
Isn't a tool that gives more names just more thorough?
No - it's usually less finished. For any handful of notes, dozens of chord definitions technically contain them. Finding the matches was never the hard part; deciding which one to put first is. A tool that hands you forty names has done the easy half and left you the hard half. Restraint isn't a missing feature. It's the work.
Why should I trust a blind test the tool's own maker ran?
You shouldn't trust it on my say-so - that's why the whole thing is reproducible. I didn't choose the chords, didn't grade them, and didn't tell the judges which tool was mine. The inputs, the prompts, the corrections and the full judge transcripts are linked below. Theory is checkable by anyone with a fretboard, and it predates Fretscape by a few centuries. The test borrows its credibility from things that aren't me.
Doesn't Fretscape also show lots of names?
It can - if you ask it to. By default it gives one clean primary name. The rootless, slash and altered readings are off until you turn them on, and when you do they appear in their own labelled partition rather than flooding the answer. The difference between that and a wall of names is what's first, what's labelled, and what's hidden until you want it.
Was the test rigged in Fretscape's favour?
It was rigged against me, if anything. The chords were AI-generated, not hand-picked. A separate AI checked the answers so I couldn't tilt them. Two judges scored blind, anonymised, against theory, neither told which tool was mine. The honest caveat: because the inputs are published, a determined reader could re-run them and work out which anonymous tool was which. Keeping the competitors anonymous is my editorial choice, not a guarantee - so I'd rather say it plainly than pretend otherwise.
What if my voicing has no clean name at all?
Then a good tool says so. A reverse lookup confident enough to come back with "not enough here to name cleanly" is more useful than one that confabulates a label to fill the silence. Honest uncertainty beats a tidy-looking guess - on chord names exactly as much as it does on keys.
Appendix: run it yourself
The whole test is reproducible. These are the actual transcripts - generation, the independent check, and both blind judging sessions - shared exactly as they happened.
The dataset, generated by two AIs:
- Claude - https://claude.ai/share/d5b4fb4b-f73c-41f5-95a8-7e6f6872ba29
- ChatGPT - https://chatgpt.com/share/6a4276d4-fe20-83ed-bcae-8d0f478e7bbe
The independent answer-check (so I couldn't grade my own homework):
The two blind judges:
- Claude - https://claude.ai/share/05d29842-1716-476b-a78a-a3339ff71c74
- ChatGPT - https://chatgpt.com/share/6a438d0b-3984-83eb-b9ae-8bef0af712f3
Seven tools, 47 cases, two independent judges scoring against theory. The competitors stay anonymous A to G; only Fretscape's letter was ever revealed. If you want to check any single call, every case is theory - a fretboard and a bit of patience is the only equipment you need.
