AI Fitness Showdowns — ChatGPT vs Claude vs Gemini vs Perplexity

ChatGPT (GPT-4o)

Program: 4-day Upper/Lower split. Upper A (horizontal push/pull focus), Lower A (squat-dominant), Upper B (vertical push/pull focus), Lower B (hinge-dominant). Each session: 2 compounds, 3 accessories. Progressive overload: add 5lb upper, 10lb lower every 2 weeks.

Nutrition: Recommended a recomp approach — 2,200 calories (maintenance), 165g protein (1g/lb), 65g fat, remaining carbs. Weekly meal prep template with chicken thighs, rice, eggs, frozen vegetables, oats, peanut butter. Estimated $55/week.

What stood out: Correctly identified recomp as ideal for skinny-fat beginners. Provided a specific grocery list that genuinely fits a $60 budget. Included a "College Meal Hack" section with dining hall tips.

Score: 9.2/10

Claude (Sonnet)

Program: 4-day full-body split (not Upper/Lower). Each day hits all major patterns with different exercises and rep ranges. Day 1: Strength (3-5 reps). Day 2: Hypertrophy (8-12 reps). Day 3: Power (explosive variations). Day 4: Volume (12-15 reps). Progressive overload via daily undulating periodization.

Nutrition: Also recommended recomp but at a slight deficit (2,000 cal) for the first 8 weeks to reduce body fat faster, then increase to maintenance. Emphasized the protein window is wider than commonly believed (within 3-4 hours post-training, not 30 minutes).

What stood out: The DUP approach is more sophisticated than a standard Upper/Lower — possibly over-engineered for a true beginner, but demonstrates real exercise science knowledge. The slight deficit recommendation for skinny-fat is debatable but defensible. Added a thorough section on "what to expect" timeline: 4 weeks to feel different, 8 weeks to see changes, 12 weeks for others to notice.

Score: 8.8/10

Google Gemini

Program: Offered a basic 4-day Push/Pull/Legs + Full Body split. Exercise selection was solid. Rep ranges appropriate. But no periodization scheme, no progressive overload plan, and no deload protocol.

Nutrition: Recommended 2,400 calories — which is likely a surplus for a 165lb sedentary-except-lifting college student. This would result in fat GAIN, not the recomp he's looking for. Protein recommendation was fine (160g) but the total calories were too high.

What stood out (negatively): The caloric recommendation was the critical error. A skinny-fat beginner eating at surplus will get fatter, not leaner. Also suggested "muscle confusion" language ("vary your exercises to keep muscles guessing") which is a red flag for outdated training philosophy.

Score: 6.5/10

Perplexity

Program: Didn't design a specific program. Instead provided a research summary of "best approaches for skinny-fat body recomposition" with 6 cited studies. Recommended finding a structured program like GZCLP or Greg Nuckols' free programs.

Nutrition: Cited Helms et al. (2014) for protein recommendations and a 2020 meta-analysis on recomposition in untrained individuals. Solid science. But no actual meal plan, no grocery list, no budget consideration.

What stood out: Perplexity doesn't build programs — it researches them. The citations were genuinely useful for validation but this doesn't solve the student's problem of needing an actionable plan NOW.

Score: 5.5/10 (excellent research, poor actionability)

ChatGPT (GPT-4o)

Program: Designed a progressive 12-week plan in 3 phases. Phase 1 (weeks 1-4): Stability and Pain-Free Movement — mostly bodyweight, emphasizing core stabilization (dead bugs, bird dogs, pallof press), hip mobility, and single-leg balance. Phase 2 (weeks 5-8): Controlled Loading — introduces goblet squats, hip hinge patterns with light weight, cable work. Phase 3 (weeks 9-12): Building Strength — barbell movements with specific modifications (front squats instead of back squats for L4-L5, avoiding deep lumbar flexion).

What stood out: Correctly avoided both back squats (axial loading with L4-L5 issue) and deep knee flexion beyond 90° (meniscus). Included daily hip flexor stretching routine. However, didn't flag when to see her physio for reassessment during the progression.

Score: 8.5/10

Claude (Sonnet)

Program: Started with a RED FLAGS section — explicitly listed signs to stop immediately and see her doctor (neurological symptoms, sharp/shooting pain vs. dull ache, new numbness or tingling, pain that wakes you from sleep). Then designed a similar 3-phase program but with MORE conservative initial loading and MORE specific modification rationale.

What stood out: Claude's safety-first approach was noticeably stronger. It recommended a physio check-in at week 4 before advancing to Phase 2. It explained WHY each exercise modification protects her injuries (e.g., "Goblet squats instead of back squats reduce shear force on L4-L5 by positioning the load anteriorly"). The educational value was significantly higher than ChatGPT's.

Also included a "Desk Worker Daily Movement" section: 5—minute movement snacks every 90 minutes of sitting, specific stretches for her desk setup, and a standing desk transition protocol.

Score: 9.4/10

Google Gemini

Program: Offered reasonable exercise modifications but included barbell back squats in Phase 2 — which is questionable with an L4-L5 herniation. Also recommended running (building back to it) without addressing the knee issue's implications for impact loading.

What stood out (negatively): Didn't adequately address the interaction between her two injuries. Someone with both L4-L5 and knee issues needs careful kinetic chain programming — Gemini treated them as independent problems rather than an interconnected movement system.

Score: 6.0/10 — The back squat recommendation with a known herniation is the critical error.

ChatGPT (GPT-4o)

Meal Plan: Built around eggs (5 dozen/week at $8), chicken thighs (5lb at $7.50), rice (10lb bag at $6), dried beans/lentils (3lb at $4), oats (large canister at $4), peanut butter (2 jars at $5), bananas (bunch at $2), frozen vegetables (4 bags at $5), whole milk (2 gallons at $6), bread (2 loaves at $4). Total: $51.50.

Hit 3,180 calories and 178g protein daily. Included Sunday meal prep schedule (3 hours, all lunches and dinners prepped for the week). Each meal was dead simple: variations of eggs + oats + banana for breakfast, chicken + rice + beans + veg for lunch/dinner, PB sandwiches and milk for snacks.

What stood out: Remarkably realistic budget math. Actually priced items at typical US grocery store prices (not organic/premium). The meal prep schedule was practical and time-efficient.

Score: 9.5/10

Claude (Sonnet)

Meal Plan: Similar core foods but added canned tuna, cottage cheese, and sweet potatoes. Total budget estimate: $48. Hit 3,200 calories and 182g protein. Added a "Flavor Rotation" system — 5 different spice/sauce combos for the same base meals to prevent diet fatigue during a 16-week bulk.

What stood out: The flavor rotation system is genuinely brilliant for diet adherence on a boring budget bulk. Also included a micronutrient analysis — flagged that this meal plan was low in omega-3s and vitamin D, and recommended a basic daily vitamin (~$0.07/day) as the one supplement worth adding.

Score: 9.3/10

Google Gemini

Meal Plan: Overestimated the budget capacity significantly — included items like salmon ($12 for 2 portions), Greek yogurt ($7/week), and almond butter ($8) that blew the budget. Total came to approximately $75-80/week despite the $50 constraint.

What stood out (negatively): Failed to take the budget constraint seriously. The meals were nutritionally excellent but financially impossible. Also didn't include any meal prep strategy.

Score: 5.5/10 — Budget math failure.

ChatGPT (GPT-4o)

Program: 4-day split — Chest/Triceps, Back/Biceps, Legs, Shoulders/Arms. Correctly identified the three big challenges: 1) Back training without a pull-up bar is hard — used dumbbell rows, renegade rows, seal rows (lying face-down on bench), and gorilla rows. 2) Heavy leg training with max 52.5lb dumbbells — used Bulgarian split squats, single-leg RDLs, and tempo manipulation (4-second eccentric). 3) Progressive overload when weight is capped — used rep progression, tempo manipulation, pause reps, and 1.5 reps.

What stood out: The tempo and intensity technique programming was excellent. When you can't add weight, you add time under tension — ChatGPT nailed this. Included a weekly progression scheme: Week 1-4 standard tempo, Week 5-8 slow eccentrics, Week 9-12 pause reps at bottom position.

Score: 9.0/10

Claude (Sonnet)

Program: Similar structure but organized as an Upper/Lower split (4 days: Upper Strength, Lower Strength, Upper Hypertrophy, Lower Hypertrophy). The strength days used lower reps with tempo manipulation; hypertrophy days used higher reps and mechanical dropsets.

What stood out: Claude introduced mechanical dropsets — starting with the hardest variation and dropping to easier variations without rest (e.g., Bulgarian split squat → reverse lunge → regular squat, all with the same weight). This is a genuinely advanced technique for maximizing stimulus when weight is limited. Also provided a detailed comparison of "what you'll lose vs. maintain" — honestly stated that maximal strength on squat and deadlift will decrease but muscle mass can be fully maintained or even increased with proper programming.

Score: 9.2/10

Google Gemini

Program: Offered a generic "full body dumbbell workout" that looked like it was designed for a beginner, not a previous 315 squatter. Included standard exercises (goblet squats, dumbbell bench press, rows) but no intensity techniques, no tempo prescription, no periodization, and no progression scheme beyond "increase reps."

What stood out (negatively): Completely ignored the user's training history. This program would provide zero stimulus for someone who was squatting 315lb. 52.5lb goblet squats with no tempo manipulation is a warm-up for this person, not a training stimulus.

Score: 4.5/10

Cross-Showdown Patterns