Skip to content

Guide

How to evaluate AI workout software with a 30-minute test on your hardest client

A feature list rewards the vendor with the longest page. One hard client, three changed inputs, and an eight-row score sheet give any tool a comparable score in half an hour.

By RaiNGE product teamUpdated 8 min read

Written for coaches and gym owners comparing AI workout tools before paying for one.

Short answer

Evaluate AI workout software with a 30-minute test on your hardest client. Write a one-paragraph brief, run the tool, change one input per run three times, time your own review, and score the output on eight criteria out of 16. A reasonable pass line is 12 of 16, with no zero on flagged limitations or approval.

  • Test with your hardest client: an injury flag, limited equipment, or a gappy schedule.
  • Change one input per run and watch what moves. A tool that ignores a change is not reading your inputs.
  • Time your own review and score eight criteria from 0 to 2. A high score with a slow review is still a poor trade.

Try a baseline run on one invented client: the free sample workout is one workout PDF after a short intake, so you can see what comes back. Get the free sample workout

On this pageRun the test in five steps

Run the test in five steps

  1. 1

    Write the brief (5 minutes)

    Pick your hardest client: an injury flag, limited equipment, or a schedule full of gaps. Write one paragraph covering age, goal, days and session length, equipment with load ranges, injury and pain notes, and attendance over the last two weeks. Remove the name and any other identifier.

  2. 2

    Run the tool once (3 minutes)

    Enter the brief exactly as written and keep the output. It is your baseline, and it shows how much of the brief the tool used.

  3. 3

    Change one input per run, three times (7 minutes)

    Make three more runs, changing one input each time. The table below says what should move and what should stay put. For the shoulder flag, run it twice: a rule removes the same exercises both times, and a model following a prompt may not.

  4. 4

    Time your own review (10 minutes)

    Start a timer when you open the baseline draft and stop when you would approve it for the client. If the timer reaches 10 minutes and you are not ready to approve, stop: that is the result. Compare your time with how long you take to write that week by hand.

  5. 5

    Score the outputs (5 minutes)

    Score the baseline and the three changed runs on the eight criteria below, 0, 1, or 2 each, using your review time for row 6. Take the worst score across the runs for each criterion, because your roster will include the hard cases. Rows 7 and 8 come from the vendor's written answers and an export you try, so ask for those before you start the clock and score either row 0 until you have it.

These three changes suit most clients. If your hardest client already has a shoulder issue, swap in a different flag, such as a knee, and look for exercises that load that joint to leave the draft or carry a visible caution.

Three input changes, what each should do to the draft, and what should stay put.
ChangeWhat should moveWhat should stay put
Full gym to dumbbells onlyEvery exercise becomes possible with the dumbbells you listed, at loads inside their rangeDays, session length, and every injury flag
Add a shoulder flagOverhead pressing, bench and chest press variations, push-ups, dips, and hanging work leave or carry a visible caution on both runs, unless your brief says you cleared themEverything that does not load the shoulder
Cut four days to threeWeekly sets per muscle group are redistributed across the three daysEquipment, flags, and each main pattern, unless a reason is shown for dropping one
Three input changes, what each should do to the draft, and what should stay put.

The score sheet: eight criteria, 0 to 2 each

Score every row after each run. The last column defines each score, so you are grading against a written rule.

Eight criteria for AI workout software: what good looks like, the red flag, and how to score each.
CriterionWhat good looks likeRed flagScore 0 / 1 / 2
1. Uses your inputsDays, session length, equipment, load ranges, goal, and injury notes all show up in the draftA generic week that fits anyone with the same goal0: ignores two or more inputs. 1: ignores one. 2: uses all
2. Limitations handled by ruleFlagged exercises are absent on every run, and the tool shows what it removedThe only protection is a sentence in the prompt, and results change between runs0: a flagged exercise appears with no caution. 1: removed on some runs, or flagged only with a warning. 2: removed on every run, with a visible reason
3. Real exercise libraryEvery name matches an entry you can look up, and the same movement keeps one nameInvented names, or one movement under three names0: three or more names you cannot match. 1: one or two. 2: none
4. Shows its reasoningYou can see which inputs shaped the draft and any caution on an exercise it keptA polished plan with no trace of why0: no reasons shown. 1: a general explanation only. 2: reasons or cautions you can check per exercise
5. Holds for approvalNothing reaches the client until a coach approves, and every draft shows its statusA plan that can be shared or sent the moment it is generated0: can reach the client unapproved. 1: approval exists but can be skipped. 2: approval is required
6. Edit speedSwapping an exercise or changing a dose takes a few clicks and does not regenerate the weekEvery edit means regenerating, or copying the plan into another tool0: review takes as long as writing it yourself. 1: less, but more than half as long. 2: half as long or less
7. Data answersWritten answers on where client data is stored, who can read it, whether it trains models, and how to delete itNo written answer, or only a line about taking privacy seriously0: no written answer. 1: some of the four. 2: all four in writing
8. Price and lock-inA price that scales with something you can predict, and an export of programs and history that you have testedExport is available 'on request', or nobody will put the post-trial price in writing0: no export. 1: partial export, such as PDFs only. 2: a full export you tested
Eight criteria for AI workout software: what good looks like, the red flag, and how to score each.

The pass line: 12 of 16 is a reasonable starting point, and you should move it to suit your clients. Treat rows 2 and 5 as gates: a 0 on either fails the tool whatever the total, because they decide whether an unreviewed mistake can reach a client.

Worked example: two tools that tie on 12

Tool A and Tool B are made up for this example, and so are their scores and times. Both ran the same brief and the same three changes. Writing that week by hand takes the coach 18 minutes.

Scores for two hypothetical tools, with what the coach saw.
CriterionTool ATool BWhat the coach saw
1. Uses your inputs22Both drafts used every input in the brief
2. Limitations handled by rule02Tool A kept a dumbbell RDL for the low-back client and gave no caution. Tool B removed it on both runs and showed why
3. Real exercise library22Every name matched a library entry
4. Shows its reasoning21Tool A explained each pick. Tool B gave one general summary of the plan
5. Holds for approval02Tool A's plan could be shared by link as soon as it was generated. Tool B held it until a coach approved
6. Edit speed21Review took 8 minutes in Tool A and 10 in Tool B, against 18 by hand
7. Data answers21Tool A answered all four questions in writing. Tool B answered two
8. Price and lock-in21Tool A's full CSV export opened cleanly. Tool B exports PDFs only
Total out of 161212The totals tie
DecisionFail: 0 on rows 2 and 5Pass: no zero on a gateTool B goes to a trial
Scores for two hypothetical tools, with what the coach saw.

The totals tie and the decision does not. Tool A is faster to edit, easier to leave, and explains itself better, and it is still the riskier tool for client work. If you only wanted workout ideas for yourself, Tool A could be the sensible pick.

Questions to put to any vendor, live

Put these to the vendor in the demo and write the answers down. Vague answers score low on the matching row.

Ask these in the demo

  • Can you run my brief with these three changes while I watch?
  • Which parts of this draft came from a rule, and which came from the model?
  • If I flag a limitation, which exercises are removed, and where do I see that list?
  • Can anything reach my client before I approve it? Can you show me a draft's status?
  • Where do exercise names come from, and what happens when the model asks for one that is not in the library?
  • Where is client health information stored, who at your company can read it, is it used to train models, and how do I delete it?
  • What does the price scale with, and what does it cost after the trial?
  • What can I export if I leave, and in what format? Can you show me an export?
  • After a client has a program, how do I change it, and can I undo the change?

The same protocol applies to RaiNGE, which is in private beta: run your hardest client through it and score what comes back.

Common questions

How many clients should I test an AI workout tool on?

Start with one, your hardest. A healthy client with a full gym is the easy case, so it tells you little. If the tool passes, run three more that differ from the first: one with limited equipment, one with a flag, and one with a gappy schedule. Commit only after all four pass the gates.

What should I do if two tools tie on score?

Check that both pass the gates on rows 2 and 5. Then break the tie on the costs you pay every week: edit speed (row 6) first, lock-in (row 8) second. If they still tie, run your second-hardest client through both and see which review you finish faster.

Can I trust a vendor demo as the test?

Treat a demo as a tour. It runs on a client the vendor chose because it works. Ask to run your own brief live, with your three changes. If the vendor declines, score every row you could not see as 0 and ask for a trial account instead.

How do I test a tool without pasting a real client's health details?

Build a composite. Keep the facts that change programming: age band, injury area and pain score, equipment, days, and attendance. Drop or change the name, birth date, employer, and location. Read the tool's data terms before any real client goes in, and use row 7 to get the answers in writing.

Next step

Try a baseline run on one invented client: the free sample workout is one workout PDF after a short intake, so you can see what comes back.