Claude vs Gemini: Which Beginner Fitness Plan Survived a Source Check?

Two AI assistants received the same request: create a realistic four-week walking and strength plan for a generally healthy 35-year-old adult who had been mostly inactive for a year. Both answers looked organized, cautious and credible. Both mentioned the 150-minute public-health target. Both supplied sources.

Only one schedule actually added up to 150 minutes.

That small arithmetic check changed the result of our first head-to-head AI Lab test. It also illustrates why a fluent wellness answer should be treated as a draft to inspect—not as evidence in itself.

Test record

• Models tested: Claude and Gemini
• Exact model versions: Not disclosed in the material captured from either interface
• Test date: August 24, 2026
• Mode: Default mode, separate new chats
• Follow-up prompts: None
• Web or research mode deliberately enabled: No
• Editing before evaluation: None; the submitted answers were assessed as received

Because the exact model version was not recorded, this comparison applies only to these two outputs. It must not be read as a permanent ranking of either product. Model behavior can change, and a repeated run can produce a different answer.

The exact shared prompt

I am a generally healthy adult aged 35 who has been mostly inactive for the past year. I can walk independently and I have no known injury or medical restriction.

Create a realistic four-week beginner activity plan that combines walking and basic strength training.

Requirements:

• Show the plan week by week.
• State the number, duration and intensity of the walking sessions.
• Include two simple full-body strength sessions per week using little or no equipment.
• Explain how and when to progress.
• Identify warning signs that mean I should stop exercising and seek medical advice.
• Clearly separate general public-health guidelines from your own practical suggestions.
• Do not claim that this plan is personalized medical advice.
• Cite 3–5 reliable sources using direct links.
• Mention important limitations or uncertainties in your answer.

Keep the answer between 700 and 1,000 words. Do not ask follow-up questions.

What the official benchmark says

The CDC states that adults need at least 150 minutes of moderate-intensity aerobic activity each week, or an equivalent amount of vigorous activity, plus muscle-strengthening activity on at least two days. The strengthening work should involve all major muscle groups. The same CDC page also makes an important qualification: some activity is better than none.

WHO guidance similarly uses 150–300 minutes of moderate aerobic activity per week as the adult range for substantial health benefits and includes muscle strengthening. These are population-level targets. They do not tell us that an inactive individual must reach 150 minutes by the fourth week, nor do they validate a specific AI-generated schedule.

We therefore evaluated two separate questions: did each model report the guideline accurately, and did its own proposed plan match what it claimed?

The headline difference: Claude's minutes added up; Gemini's did not

Claude prescribed five 30-minute walks in week four. That equals 150 minutes, matching its statement that the plan had reached the lower aerobic guideline target.

Gemini prescribed four walks lasting 25–30 minutes. That equals 100–120 minutes, not 150. Yet the section was titled “Reaching Baseline,” and the answer said it was hitting the lower end of the national guidance.

This is not a debate about exercise science. It is a multiplication error paired with an incorrect conclusion. The proposed week may still be a reasonable amount of activity for a beginner, but it cannot honestly be described as 150 minutes.

The safer editorial correction would be either:

• describe week four as a gradual on-ramp below the full target; or
• change the schedule to five 30-minute walks if that progression is appropriate.

Reaching the guideline by a specific deadline was not required by the prompt. Gemini created the problem by claiming a milestone its own numbers did not reach.

Strength coverage: one answer included a pull; the other did not

Claude proposed chair or bodyweight squats, a push-up variation, a supported row, glute bridges and a plank variation. That gives the plan lower-body work, pushing, pulling and trunk work with little equipment.

However, Claude later referred to “the same six exercises” even though only five were listed. This is a smaller internal-consistency error, but it matters when a reader is expected to follow the plan without guessing.

Gemini proposed squats, wall push-ups, reverse lunges, glute bridges and a modified plank. The list contains useful movement patterns, but it lacks a pulling exercise for the upper back. That weakens its claim to provide a full-body routine covering the major muscle groups described by the CDC.

Reverse lunges can also be more demanding for balance and technique than a beginner expects. Gemini did recommend holding a wall or sturdy chair, which reduces—but does not remove—the need for individual modification.

Progression: both had useful ideas, but neither plan is personalized

Both answers recommended increasing gradually and delaying progression when fatigue or soreness lingers. Claude proposed changing one variable at a time: duration, then frequency, then intensity or resistance. Gemini offered a similar order and explicitly said a four-week schedule may take six or eight weeks in real life.

Those are useful practical principles. They are still editorial programming choices, not individualized conclusions derived from an assessment. Neither model knows the person's current comfortable walking duration, movement technique, recovery pattern, environment or preferences.

Claude's jump from one set in week one to two sets in week two, followed by two to three sets in week four, may be manageable for some readers and unnecessary for others. Gemini also progressed to three sets in week four. A stronger answer would make the decision conditional on technique, symptoms and recovery instead of treating the calendar as the main trigger.

Safety language: helpful boundaries mixed with overreach

Both models clearly stated that the output was not personalized medical advice. Both identified chest symptoms, severe or unusual breathlessness, dizziness or faintness, and sharp or worsening pain as reasons to stop and seek appropriate advice.

Gemini opened by saying that physician consultation before any new exercise regimen was “highly recommended” solely because the person had been inactive. This is more cautious than the prompt required and can imply that every otherwise healthy inactive adult needs medical clearance before beginning light walking. Public-health guidance encourages inactive adults to start with small amounts; individual symptoms, known conditions and risk factors determine when professional assessment is especially important.

Claude distinguished common post-exercise muscle soreness from sharp or worsening pain, but its warning section blended urgent symptoms and non-urgent professional review under one heading. A reader benefits from clearer triage: emergency symptoms require emergency help, while a recurring movement-related discomfort may warrant stopping the exercise and arranging an appropriate assessment.

Neither answer should be used as a medical screening tool.

Source audit: mostly recognizable sources, one dead CDC link

Claude linked to the current CDC adult-activity overview. That page supports the 150-minute aerobic target, two strength days and the requirement to work major muscle groups. Its WHO material was linked through PubMed and NCBI rather than directly to the WHO guideline page; the underlying guideline record is relevant, but a direct WHO link would be easier for readers to audit.

Claude's raw output also contained malformed citation artifacts such as “CDCsvg” and “MyMLC + 2svg.” The final source list was more readable, but the mismatch shows that polished citation rendering cannot be assumed.

Gemini's WHO, Mayo Clinic and American Heart Association links resolved to relevant resources when checked. Its CDC URL—cdc.gov/physicalactivity/basics/adults/index.htm—returned a 404 during our August 24 check. The CDC content still exists at a newer address, but the supplied direct link was not usable as delivered.

A citation receives credit only when it opens and supports the attached claim. A reputable domain name is not enough.

Scorecard for this single test

Claude performed better in this run because its walking total matched its claim, it included a pulling movement, and its main CDC link was current. It lost points for the five-versus-six exercise inconsistency, citation-rendering artifacts and safety language that could separate urgent from non-urgent situations more clearly.

Gemini produced a readable, encouraging plan and handled gradual progression well. Its score fell primarily because the week-four arithmetic contradicted its headline conclusion, the strength circuit omitted a pull, and the CDC source link was dead.

What a careful reader could keep—and what must be corrected

The useful shared core is modest: begin below the long-term target, walk several times per week, include two non-consecutive strength sessions, use conversational effort to judge walking intensity, and progress according to recovery rather than pride.

Before either answer could become a responsible published plan, we would:

1. establish the reader's comfortable baseline instead of assuming 15 minutes;
2. remove the requirement to reach 150 minutes by week four;
3. retain push, pull, lower-body and trunk movement categories;
4. make set increases conditional rather than automatic;
5. separate emergency warning signs from routine exercise modification;
6. replace dead or indirect links with current primary guidance;
7. correct every numerical and list inconsistency.

The result

Claude gave the stronger answer in this documented test, but “stronger” does not mean ready to follow without judgment. Its own internal exercise-count error is exactly the kind of small inconsistency that a confident format can hide.

Gemini's central mistake was more consequential: the answer promised that its schedule reached a public-health threshold when its own arithmetic showed otherwise. A reader who never multiplies four by 25–30 would have little reason to notice.

This is the purpose of the FitFat AI Lab. We do not count how many headings an answer contains or choose the friendliest tone. We record the prompt, check the numbers, open the links and compare each conclusion with the evidence it claims to use.

Independent verification sources

• CDC: Adult Activity — An Overview
• WHO: Guidelines on Physical Activity and Sedentary Behaviour
• WHO: Physical Activity Fact Sheet
• U.S. Physical Activity Guidelines for Americans

Review note: Outputs and links checked August 24, 2026. Scores apply only to the captured responses to the exact prompt shown above. FitFat AI received no payment from either provider and has no affiliation with Anthropic or Google.