ChatGPT and Claude: Music Theory, Harmony Analysis, Composition Tips
Vendor pages do not establish which assistant analyzes music better, so this afternoon plan tests harmony, cadences, composition, voice leading, score reading, and iterative feedback against your own references.
Sources checked 2 Oct 2026
For music theory, harmony analysis, and composition tips, the vendor pages do not establish that ChatGPT or Claude is better at the task. OpenAI and Anthropic instead warn that their assistants can produce incorrect or misleading answers, and neither page supplies a music-theory comparison. Chat Picker has not tested either assistant for this use, so this page gives you a repeatable afternoon test rather than a declared winner.
Know these limits before you start
OpenAI says ChatGPT may sound confident even when it is wrong and recommends treating it as a first draft while verifying important information. Anthropic says Claude can occasionally be incorrect or misleading, should not be your only source of truth, and should be checked against original sources when using web results.
As read on October 1, 2026, OpenAI’s ChatGPT pricing page gives broad usage descriptions but no exact message count, and Anthropic’s Claude pricing page likewise lists no message count. Check the current limit labels during the run instead of estimating capacity from a subscription tier.
As read on the same date, OpenAI says a training opt-out is available on Free, Go, Plus, and Pro; Anthropic says it is available on Free, Pro, and Max, while Team is not trained on by default. Confirm the setting in the account you actually use.
If your use is an elevated-risk or covered professional workflow, Anthropic’s Usage Policy calls for relevant human expertise and qualified-professional review of covered advice before finalization or dissemination; the user or organization remains responsible for accuracy and appropriateness. When model output is presented directly to consumers, Anthropic says AI involvement must be disclosed at the beginning of each session.
The test plan
Use the same score material, fresh chats, and one task per chat. Decide the reference answer before running either assistant, and keep drafts and corrections in one simple log.
-
Harmonic analysis. Input: Prepare a short four-part chorale in a major key with a seventh chord and a chromatic passing tone. Prompt: “Analyze the attached chorale chord by chord. State the key; give each Roman numeral and inversion; identify non-chord tones; explain the voice leading; cite measure numbers; and mark any unclear notation instead of guessing.” Look for: matching labels, correct inversions, and uncertainty tied to visible evidence. Log: discrepancies against the score, omitted measures, and unsupported claims.
-
Deceptive cadences. Input: Prepare a text-only progression in C major: I, ii in first inversion, V, vi. Make one version with a clear final barline and another without phrase marking. Prompt: “Assess this progression in C major: I, ii in first inversion, V, vi. Is the final V–vi a deceptive cadence? Separate what the symbols establish from what depends on meter, phrasing, and bass, and explain the expected dominant resolution.” Look for: a distinction between chord identity and cadence classification. Log: mislabeled harmony, ignored context, and unsupported certainty.
-
Composition alternatives. Input: Use this common-time C-major melody: C quarter, D half, E quarter | F half, G quarter, A quarter | B half, C half | C whole. Prompt: “Give distinct four-part harmonizations of this melody. Show chord symbols and voice assignments, explain non-chord tones, and identify each intended cadence.” Look for: notes that fit the stated chords, workable part writing, and explanations that match the notation. Log: rule violations, unnecessary leaps, unclear labels, and unstated assumptions.
-
Voice-leading rules. Input: Paste one candidate realization into a fresh chat without its previous explanation. Prompt: “Audit this four-part harmonization measure by measure for parallel fifths and octaves, direct fifths and octaves into outer voices, unresolved tendency tones, voice crossing or overlap, spacing, range, and doubled tendency tones. Quote each measure and propose the smallest repair; label rules that are stylistic rather than absolute.” Look for: actual violations, false alarms, and repairs that preserve the melody. Log: each confirmed error, missed error, and proposed repair.
-
Score reading. Input: Prepare a clear score image containing an enharmonic spelling, a tie across a barline, a tuplet, and a transposition. Prompt: “Read the attached score without relying on prior chat. State clef, key, meter, and any transposition; transcribe the opening measures beat by beat; preserve note spellings; and mark every uncertain symbol with the visual evidence behind it.” If image input is unavailable, mark the task as not run. Look for: visual grounding rather than conventional rewriting. Log: each symbol difference against the score and every invented detail.
-
Real-time feedback. Input: Reveal one measure of the same chorale per turn. Prompt: “I will send one measure at a time. After each, identify only the clearest voice-leading concern, explain its evidence, and ask one clarifying question. Do not give a full analysis until I send ‘summarize.’” Look for: use of earlier context, acceptance of corrections, and obedience to the stop instruction. Log: repeated advice, ignored corrections, lost context, and observed response delay.
How to read your results
Create one row per run with the assistant, plan, prompt, attachment status, reference checked, factual errors, theory errors, unsupported certainty, rule violations, observed delay, and ignored instructions.
Set pass conditions before viewing the responses, and keep factual mistakes separate from stylistic preferences. If both assistants meet your must-have tasks, use the documented price, data setting, context, and current usage limits as tie-breakers. If neither does, verify the passage against the score or a qualified teacher. Chat Picker does not score or rank your test.
Where the plans differ
On the Claude vs ChatGPT vs Gemini matrix, inspect the rows Free plan, Main paid plan, Heavy-use plans, Context window in the app, How usage limits are described, and Training on your chats. The context and usage rows matter if your test includes a full score or repeated runs; neither pricing page supplies a universal message allowance.
As read on October 1, 2026, OpenAI’s What is ChatGPT Plus? Help Center lists Plus at $20 per month billed monthly, while Anthropic’s Claude pricing page lists Pro at $20 per month on monthly billing or $17 per month on annual billing, with $200 paid up front. Use the training row to confirm the setting for your account rather than making a general assumption about either service.
Chat Picker’s method leaves an unconfirmed figure blank and marks it “Not verified.” Confirm current prices and limits with the vendor before subscribing.