Member Follow-Ups and Workout Plans for Your Gym, Cut in Half with Claude Code

Friday, 10 p.m. I’d just walked the last member to the door and dimmed half the lights at the front desk, and I was still sitting at the computer. Twelve follow-up messages still left to send for next week. One member’s visits have dropped off since last month. Another signed up two weeks ago and hasn’t booked a single session. One by one I open each file, check the last workout we did, and type some bland “How have you been lately?” message into the app. Do this every week and you’ll have plenty of days where the messaging takes longer than the actual coaching.

And here’s the worst part: while I’m wordsmithing a follow-up, my focus drains away, and I never quite work out the thing that actually matters — why did this person stop coming? So I just nudge a template a little and hit send. Weeks later I notice, “Right, that one ended up canceling.” If you run a gym, you’ve probably felt that exact regret at least once.

I handed most of this “follow-up swamp” over to Claude Code and generative AI. It is not perfect automation. But once I had the AI do the prep work and left the human to make the calls, my weekly admin load dropped by more than half by feel. Today I’ll lay out exactly how I do it, copy-paste templates included.

Key takeaways

• Member follow-up is less about “time spent writing” and more about “time spent deciding who to say what to.” Let the AI do the prep; the human decides whether and when to send.
• Three jobs suit the AI well: pulling the list of at-risk members, drafting the follow-up messages, and producing a first draft of the workout plan.
• I’ve included a prompt template you can use as-is, plus a check script that keeps member data safe before it ever reaches the AI.
• Design the workflow so real names and contact details never go to the AI. Run everything on anonymous IDs only.
• Even for a small gym, cutting 3 to 5 hours of admin a week is a realistic target.

Who this is for, and how the work flows today

This article helps people like this: the owner of a small or personal gym with somewhere between 50 and 300 members, or a trainer on the floor. You have no dedicated admin staff, or maybe one person. You’re bringing in new members fine, but the ones who join keep quietly drifting away within three months. You want to fix that, but daily coaching eats all your time.

Break the current follow-up work down and it usually runs in this order:

1. Open the visit history or booking system and vaguely hunt for members who haven’t shown up lately
2. Recall that person’s file, their last workout, their goals
3. Think up a follow-up message and send it through the app
4. When a reply comes, steer it toward the next booking
5. Build next week’s workout plan, one member at a time

The problem is that steps 1 through 3 are “manual, from scratch, every single time.” The two biggest sources of rework here are “missing the people who stopped coming” and “rebuilding a plan after you finally hear about a member’s condition or changed goal.” The first one feeds straight into revenue, yet it’s the one that keeps getting pushed to later.

Use case 1: Pull the list of at-risk members every week

The first move is figuring out “who needs a follow-up.” This is mechanically lining up decision factors, so it suits the AI well.

If your booking system can export a CSV, hand over the visit date, join date, and whether there’s an upcoming booking — all keyed by anonymous ID — and ask the AI to sort by risk. You do not hand over real names. A member number or a hashed ID is plenty.

Have it pick out the members who meet this checklist as “needs follow-up”:

• No visit in the last 14 days (for a member who normally comes twice a week)
• Joined within the last 30 days and has visited two times or fewer
• No reply and no visit since the previous follow-up
• Less than one month left before their goal deadline

The AI’s job ends there. “What to say to this person” and “whether to reach out at all” are human calls. For example, on-the-floor context like “I heard last week they got injured, so let’s not bring up training right now” doesn’t show up in the data. That’s the part a human picks up.

Use case 2: Draft a follow-up message per member

Once you have the list, the next step is the wording. This is where the AI earns its keep. But you do not auto-send the messages as written. Have it produce drafts, and let the human fix them in three seconds. That turned out to be exactly the right balance.

Asking it to lay out drafts in different tones by member type makes scanning much faster.

Member type	Situation	Direction of the message
Just joined, low frequency	No visit in two weeks	Lower their anxiety. Offer a small next goal
Plateau slump	Visits dropped to twice a month	Review their results. Explain the plateau
Near their goal	One push to go	Sketch the next goal together, post-achievement
Out of touch	A gap of over a month	Don’t scold. Lower the bar for coming back

Put this type breakdown into the prompt and the AI writes all the variations in one pass.

Use case 3: Build a first draft of the workout plan

The third job is plan creation. Feed it the member’s goal, current state, available equipment, and last session’s load, and the AI hands back a weekly first draft.

What I want to stress here is that what the AI produces is a first draft. Range-of-motion limits, existing conditions, how the member feels on the day, quirks in their form — anything that touches safety directly, the trainer always overwrites. What you delegate to the AI stops at the exercise combinations, a rough rep-and-set count, and a starting proposal for progression from last time. Deciding “is it OK to put this weight on this person” is, in the end, a human call.

Draw a clear line around what the human always decides:

• Whether an exercise is allowed given prior injuries or painful areas
• Whether to clear heavy loads or high intensity
• Swapping the plan based on how they look on the day
• Diet and supplement advice tailored to the individual (be especially careful with anything near the medical line)

A prompt template you can copy and paste

This is the template I actually use. Member data is anonymous IDs only — no real names, no contact details.

You are a follow-up assistant for a small fitness gym.
Read the member data below (anonymous IDs, visit history, and goals only) and:
1) Sort the needs-follow-up members from highest risk to lowest.
2) For each member, write one LINE/app message draft that does not scold and
   includes a small next goal.
   - 120 characters or fewer, at most one emoji, polite tone
   - State the type: just-joined / plateau slump / out-of-touch
3) Add a one-line note on what the trainer should double-check.

Output as a table. Do not output real names, phone numbers, or addresses.
Data:
{paste the anonymized CSV here}

For the workout plan, use this one:

You are an assistant to a personal trainer.
Read the member's goal, current state, available equipment, and last plan,
then build a first draft of this week's three sessions.
- Exercise name, rough set count, a progression proposal from last time
- For each day, write one line on the safety points the trainer must check
- Always mark exercises that touch prior injuries as "needs review"
Output as a table. Write it on the assumption the trainer makes the final call.
Member data:
{paste the anonymized data here}

A check script that protects personal data

Before any member data goes to the AI, run a machine check for real names, phone numbers, or emails mixed in. Human eyeballing always misses something eventually, so put a gatekeeper here. It’s a small validation script that runs on Node.js.

import { readFile } from "node:fs/promises";

// Gatekeeper: machine-check the CSV for personal data before sending it.
const text = await readFile(process.argv[2] || "members.csv", "utf8");

const rules = [
  { name: "phone number", re: /(\+?\d{1,3}[ -]?)?\(?\d{3}\)?[ -]?\d{3}[ -]?\d{4}/g },
  { name: "email", re: /[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Za-z]{2,}/g },
  { name: "name column", re: /(full[ _]?name|first[ _]?name|last[ _]?name|name)\s*[:,]/gi },
];

let hit = 0;
for (const rule of rules) {
  const found = text.match(rule.re);
  if (found) {
    hit += found.length;
    console.log(`NG: found ${found.length} value(s) that look like a ${rule.name}`);
  }
}

if (hit === 0) {
  console.log("OK: no personal data found. Safe to send with anonymous IDs only.");
  process.exit(0);
} else {
  console.log("--- Remove the matching columns before sending to the AI ---");
  process.exit(1);
}

Running it is just this:

node check-pii.mjs members.csv

If NG shows up, delete that column before sending. Make it a rule never to send to the AI until you get OK, and the accident of “I just pasted a real name by mistake” disappears. Ask Claude Code to “run this script over every CSV in the folder” and it’ll apply it automatically on every export. For the basics of getting going, see the Claude Code beginner guide, and for teaching it your project-specific rules, see the CLAUDE.md best practices.

What changed, before and after

Numbers land better, so here’s my rough tally. This is the time I used to spend on weekly follow-ups at a gym of about 120 members.

Task	Before	After
Pulling the needs-follow-up list	~60 min	~10 min
Writing follow-up messages (12 people)	~90 min	~25 min
Drafting workout plans (10 people)	~120 min	~40 min
Total (per week)	~4.5 hours	~1.25 hours

That’s a little over three hours saved a week. By hourly rate it’s maybe 100 to 200 dollars a month, but the bigger thing is getting back “time to actually think about coaching.” For tightening up your prompts, advanced prompt engineering is a good reference. If you’re new to AI, starting from the Claude Code intro for non-engineers is the easier way in, and productivity tips help once you’ve got the basics.

As a rough ROI gauge: if a tool runs you around 30 dollars a month, preventing a single member cancellation through stronger follow-up more than pays for it. For churn benchmarks, lean on industry data like the IHRSA fitness association reports while measuring against your own gym’s real numbers.

FAQ

Q. Can I just send the message the AI wrote, as is? Better not to. Use it as a draft and always read it before sending. Adding one line — the member’s name or something recent — kills the machine feel. Not automating the send itself is the trick to not breaking trust.

Q. Can I put member names or contact details into the AI? Don’t. You can run everything on anonymous IDs plus visit history and goals. Make it a rule to run the check script above before sending, and accidents are nearly eliminated.

Q. Can I use this even if I’m not good with computers? Claude Code takes instructions in plain conversation. If you can export a CSV and paste in a prompt, that’s enough. The steps for that first small move are in the beginner guide.

Q. I’m worried about the safety of the plans. Treat the AI’s output strictly as a first draft. Run it on the firm assumption that weight settings and exercises touching prior injuries are always overwritten by the trainer. Decide the line up front, and the AI works safely as your prep assistant.

Q. Which task should I start with? I’d start with pulling the needs-follow-up list. If it goes wrong, nobody’s inconvenienced, and you see the payoff right away. Once you’re comfortable, expand to the messages, then the plans last — that keeps it painless.

What I actually found when I tried it

What I tested was which holds up on the floor: “hand everything to the AI” or “hand over only the prep.” The one that stuck was the latter.

At first I got greedy and wired up auto-send too. But I sent one member “How have you been feeling this past week?” — and they’d come in the day before and reported they felt great. The data import was a day out of sync. I went pale after hitting send. Ever since, I put the send back to a human pressing the button.

The check script and the list-pulling, on the other hand, were a big win. Once I slotted in the real-name check, the “about to paste the whole member roster” near-misses that had spooked me a few times went to zero. The needs-follow-up list also caught more than my gut-feel hunting used to, so first follow-ups for members within their first month now land in time. If your company wants to put this into real operation, training and consulting can help work out a setup that fits your floor. If you’d rather try it solo first, the learning materials and a free PDF are over here. The AI isn’t magic, but just keeping that division — hand over the prep, let the human focus on the calls — got me home from those 10 p.m. nights a whole lot earlier.