Can a Personal Trainer Edit Testimonials With AI? What Is Allowed

Yes, for light edits. Advertising rules don't stop a trainer using AI to fix spelling, tidy punctuation or shorten a client's testimonial, provided the client approves the final wording and nothing meaningful is added. They do stop invented praise, stronger results and merged messages, and a client's name or photo still needs their agreement.

The rules don't ask whether AI touched the text. They ask whether the published words are a genuine, current opinion from a real client, and whether any result inside them can be backed up. That is why nearly all the risk sits in one habit: pasting a testimonial into a chatbot with "make this sound better", which tends to hand back the claim the client never made.

Follow me on Instagram@sagnikteaches

Where the testimonial rules agree

Advertising rules differ by market and they change, so read this as a map rather than legal advice. The points below come from the published guidance of two separate bodies, a consumer-protection regulator and an advertising code, plus Meta's current advertising standard for health and wellness. They agree on more than they differ, and the points below are the ones that decide what you can do with AI. The wider question of AI-written testimonials has its own tutorial; this one stays with a trainer who has real client words and wants to tidy them.

Connect on LinkedInSagnik Bhattacharya

• A testimonial you publish is your claim. One advertising code's guidance says testimonials are held to the same standards as any other advertising claim. A real client's opinion doesn't exempt "these workouts really work" from needing evidence, and a testimonial on its own doesn't count as that evidence. The consumer-protection regulator's guidance on its rule makes the same point from the other side: a business that puts testimonials on its own website is publishing them, not merely hosting them.
• It must be genuine, and you must be able to show it. The same guidance expects you to hold evidence that a testimonial is genuine, plus contact details for whoever gave it, and it splits "genuine" into two parts: the quote comes from a real person, and it reflects what they said. An AI rewrite can pass the first test and fail the second. So keep the client's original message, not only your polished version.
• Editing mustn't mislead. It says testimonials shouldn't be taken out of context or edited in a way that is misleading. Those few words are the whole basis for "light edits only".
• Fakes are banned, AI-generated ones included. A consumer-protection regulator's final rule on fake reviews and testimonials covers ones written by people who don't exist or have no real experience, names AI-generated fakes, and also targets paying for positive reviews and undisclosed insider testimonials. The regulator can seek civil penalties against knowing violators.
• The words must reflect the client's real experience. That regulator's endorsement Q&A says an endorsement "has to represent the accurate experience and opinion of the endorser". Someone who has only tried something once can't suggest they do it regularly. And if an ad features an exceptional result (the example it gives is losing 20 pounds in two months) without proof that it is typical, the ad has to make clear what results people generally get.
• Platforms add their own limits. Meta's health and wellness standard, updated in July 2026, bars ads that promise specific outcomes within a set timeframe without disclaimers or qualifiers, and ads that use sensational language with exaggerated or extreme claims about health or weight.

Nothing there bans AI as a tool. The rules punish the output: a quote that no longer matches what the client said, or a result nobody can back up. A spell-checker passes that test every time, and a chatbot asked to "make it better" often fails it.

Subscribe on YouTube@sagnikteaches

Where tidying ends and rewriting begins

It helps to think in three zones. The first is mechanical: spelling, punctuation, capital letters and cutting filler words. The second is reshaping: shortening, reordering, changing tone, translating. The third is adding: any new word that carries meaning, such as an adjective, a number, a timeframe, a result or a name. Zone one needs the client's yes to the final text. Zone two needs the same yes, plus a side-by-side check that the sense survived. Zone three isn't editing at all. It is writing something the client didn't say and putting their name under it.

Row four is the grey one. Squeezing two sentences into one is how small drifts begin, so treat anything the AI rephrased as a new draft the client has to read, not a tidy-up you can approve on their behalf.

Twelve raw testimonials, sorted before anyone opens a chatbot

Take a solo trainer with about 40 clients who sends a feedback form at the end of each 12-week block. Over one quarter she ends up with twelve usable messages: seven from the form, three from WhatsApp and two Google reviews. The numbers are illustrative, but the sorting step is the one I'd copy. Before any AI is involved, she puts each message in a bucket that decides what the AI may do with it.

Here is how the week played out. The four clean messages were approved within a day. Of the three long ones, two approved the shortened version and one never replied, so that one stays off the website. Of the two with numbers, a deadlift that rose from 40 kg to 70 kg across the block is backed by her training log, so it goes up with the start and end dates and the twelve-week span. The other claimed 9 kg lost in eight weeks but rested on a weigh-in the client did at home, so the trainer asked what they'd like to say that they could stand behind, and got back "I'm down a few kilos and my jeans fit again". The client who mentioned a slipped disc swapped that sentence for "I can sit through a whole film again without my back complaining". One Google reviewer said yes to being quoted on the website; the other stays on Google only.

Ten of the twelve end up published somewhere, and none without a recorded yes. The sums on time are worth seeing too. Editing by hand costs her about ten minutes a message, so 120 minutes. With the strict prompt below and a read-through it takes about three minutes each, so 36. The 84-minute saving is all in the editing. The twelve approval messages (about two minutes each, 24 in total) and the two evidence checks (about ten minutes each) cost the same either way, and they are the parts that protect her.

A prompt that only tidies, and what comes back

Most of the risk lives in the instructions. Here is a raw message as a client might type it on their phone (illustrative, with [trainer] standing in for your name):

i started with [trainer] in march cos my knees were killing me on stairs. 12 weeks later im doing 3 flights no problem and i even jogged for 10 mins last week which i havent done in years!!! he is so patient and explains everything, never makes u feel stupid. best decision ive made this year tbh

And a prompt that treats the AI as a proofreader rather than a copywriter. Paste one testimonial per chat, never a batch:

You are a proofreader, not a copywriter. Below is a testimonial a client wrote about my personal training.

Rules:
1. Fix spelling, punctuation and capital letters.
2. You may delete filler words and one off-topic aside. You may shorten the text to about 60 words, but only by deleting whole sentences or clauses. Never rephrase in order to shorten.
3. Do not add, replace or strengthen any word that carries meaning: no new adjectives, numbers, timeframes, results, names or claims.
4. Keep the client's order, voice and first-person wording.
5. After the edited version, list every change as "original -> edited", one per line. Mark any change that is not a spelling, punctuation, capital-letter or deletion fix.
6. List every sentence that states a health, weight or performance claim under "CLAIMS TO CHECK". Do not alter those sentences.

Testimonial:
"""
[paste the client's words here, without surnames, phone numbers or health details they did not mean to publish]
"""

A typical reply from a current chat model looks like this (illustrative):

EDITED VERSION
I started with [trainer] in March because my knees were killing me on stairs. Twelve weeks later I'm doing three flights no problem, and I even jogged for ten minutes last week, which I haven't done in years! He is so patient and explains everything, never makes you feel stupid. Best decision I've made this year.

CHANGES
i -> I (x3); he -> He; best -> Best; march -> March; im -> I'm; havent -> haven't; ive -> I've; commas added after "problem" and "last week"; cos -> because (slang expanded, marked); u -> you (marked); 12 -> Twelve, 3 -> three, 10 mins -> ten minutes (number style, marked); "!!!" -> "!"; deleted "tbh"

CLAIMS TO CHECK
- "Twelve weeks later I'm doing three flights no problem"
- "I even jogged for ten minutes last week"

Look at what survived: every claim is still the client's, in the client's order, and the AI has pointed at the two sentences that need your attention. The first is a performance claim you can check against a session note. The second is one jog; it stays as a single jog, and it must not drift into "runs regularly", because the regulator's endorsement guidance says someone who has only tried something once can't suggest they do it regularly. Two things I'd still fix by hand: put your own name back where [trainer] stands, and ask the client what "three flights" means before it goes near an ad, since some people count floors and some count flights.

Now the same message with the prompt most people type, "make this sound more professional for my website" (illustrative):

Training with [trainer] has transformed my life. After just 12 weeks my knee pain has vanished and I'm now running regularly. He is knowledgeable, inspiring and endlessly patient. I couldn't recommend him more highly.

Count what was added. "Transformed my life" is a far stronger opinion than "best decision I've made this year". "Just 12 weeks" implies speed. "Knee pain has vanished" is a medical-sounding outcome, where the client said only that stairs used to hurt and now they manage three flights. "Running regularly" turns one ten-minute jog into a habit. "Knowledgeable" and "inspiring" are two adjectives nobody wrote, and "couldn't recommend him more highly" is a recommendation nobody gave. Seven additions in four sentences, and every one flatters the trainer.

How a polished testimonial goes wrong in practice

A realistic slip, described as an illustration rather than a real case: a trainer pastes all twelve testimonials into one chat with "polish these for my website" and posts the results with first names. The chatbot has done what it was asked and made them sound alike. Two similar messages about finally enjoying exercise have been merged into one quote, and one client's "my back feels a bit better" has become "training fixed my back pain".

It doesn't surface through a regulator. It surfaces when a client comments under the post: "I never said that." The trainer pulls the post, restores the original wording, apologises, and moves to the routine in this tutorial: one message per chat, a changes list, a side-by-side read and a yes from the client. One message per chat matters more than it sounds, because it stops a model blending two voices and keeps one client's words out of the next client's prompt.

The regulator's endorsement Q&A has a useful line for this situation: an endorser isn't liable for changes a marketer makes to their review, and they can ask the marketer to stop using an altered one. Responsibility for the edit sits with whoever published it, and that is you.

When a number in the testimonial becomes your claim

Testimonials come in three flavours. Feelings ("I feel stronger") carry the least risk. Measurements ("my 5K dropped from 34 minutes to 29") can be checked. Body and health outcomes ("lost 9 kg", "my back pain is gone") carry the most, because advertising codes and platforms watch them most closely and because they can only ever be true for one person on one programme. The advertising code mentioned above applies its health-claim rules to testimonials too, so a client's "it cured my back" is judged like your own claim.

Here is how "I lost 9 kg in 8 weeks!!" travels from raw message to something publishable (illustrative):

A quick sum shows why the fourth row matters. Suppose eleven clients completed that eight-week plan last year, with weight changes from 0 to 5 kg and a middle value of 2 kg. A 9 kg headline sits well outside what most clients saw. The ad has to say what people generally get, or lead with something else, such as the client's sentence about her energy. The regulator's endorsement guidance is explicit that a line such as "results may vary" doesn't change how a specific result is read.

The timeframe is a second trap. Eight weeks is a timeframe, and a testimonial in an ad that says "lost 9 kg in 8 weeks" can be read as a promise of that outcome in that time, which is the shape Meta's standard targets. Ask for qualifiers or lead with something else. For health outcomes, the safest move is to let the client describe how they feel in their own words with no cause attached. "I can sit through a whole film again without my back complaining" says something real without claiming you treated anything. The general method for backing up every claim an AI writes in your marketing is in the tutorial on checking each claim before it goes out.

Consent that still holds a year later

The yes you need is a yes to the exact text you'll publish, not to the idea of a testimonial. If a client agreed to the raw message and you then ran it through AI, the agreement doesn't automatically cover the new version. Send the final wording, say where it will appear, and say how to withdraw. A message like this does the job (illustrative; fill in the brackets):

Hi [first name], thanks for your feedback on the form. Would you be happy for me to use it as a testimonial on my website and Instagram? I've fixed the spelling and shortened it. This is exactly what I would publish:

"[final text]"

I'd show it with your first name only, and no photo. Reply YES if that's fine, or tell me what to change. You can ask me to take it down at any time and I'll remove it within [number of days] days. Nothing goes out unless you reply.

[your name]

Three details make it hold. It shows the final text, so the client approves what the public will read. It names the places, so a yes for a website quote isn't stretched into a paid ad. And it promises a removal window you can keep, so pick a number of days you can honestly meet. Keep the reply, the original and the final text in one place. A simple log is enough:

Ask whether the client wants a first name, initials or no name at all, and treat photos as a separate yes. A testimonial that mentions an injury or a condition is also personal information about their health, so the handling side belongs in your data-protection routine; there is a checklist for client health data and AI for exactly that.

What goes into the AI tool, and what stays out

Before you paste anything, remove surnames, phone numbers, workplaces and any health detail the client didn't intend to publish. The AI doesn't need them to fix a comma, and a clean text gives a strict prompt less to be tempted to "improve".

Check where the words go next. Consumer ChatGPT and Claude plans both have a model-training switch in their privacy settings, so look at it before you paste a client's words into either. The vendors' business plans (ChatGPT Business, Claude Team, Microsoft 365 Copilot, Gemini in Workspace) don't train on business content by default, according to the vendors' own statements.

Photos and video follow the same rule, applied harder. AI retouching can narrow a waist or add definition without anyone noticing, which turns a transformation photo into a false claim. Lighting, cropping and straightening are fine; anything that changes how the body looks isn't. For a video testimonial, AI captions are fine if you proofread them. Cutting pauses so that a hesitant "I think it helped" becomes a confident "it helped" is editing in a misleading way, and cloning or regenerating a client's voice is off the table altogether.

Free sessions, friends and old testimonials

Three situations look harmless and aren't.

A free session in return for a review. Offering a reward for a positive testimonial is the pattern the regulator's fake-review rule targets, since it covers compensation conditioned on what the review says. Google's own review policy bars incentives for reviews posted there. A reward for an honest review, whatever it says, is a grey area that depends on your market. If you do it, say so beside the testimonial, and keep it away from Google. The tutorial on getting more Google reviews with automated requests covers the asking side.

Friends, family and staff. A testimonial from your sister, a trainer who shares your gym or someone on your payroll comes with a material connection. The same rule targets insider testimonials that don't clearly disclose that connection. Either disclose it or leave them out.

Old testimonials. The advertising code's guidance says a testimonial must relate to the product advertised, and warns against altering a customer's testimonial so that it seems to relate to a newer incarnation of the business. That is an AI-shaped trap. Asking a chatbot to "update" a quote about your old boot camp so that it praises your new online coaching creates a new, false testimonial. If the service has changed, keep the quote with its date or ask the client for a fresh one.

Three trainers, three set-ups

The one-to-one trainer renting space in a gym. Her testimonials arrive as WhatsApp messages, and the temptation is to screenshot them. A screenshot can't be tidied, which is its advantage: it shows exactly what the client wrote. If she wants clean text, she quotes the message as text with the client's yes to that version. She never edits a screenshot, with or without AI, because a changed image of a message is a fabricated message.

The online coach with about 60 clients on a coaching app. Her testimonials come with numbers and transformation photos, which is where the exposure is. She adds two rules to her process: no AI touches photos at all, and no number goes into an ad without a dated entry in the client's check-in history. She also keeps each testimonial tied to the programme it describes, so a quote about her old six-week plan is never attached to the new twelve-week one.

The small-group studio with a busy Google profile. Most feedback arrives as Google reviews, which belong to the reviewers. The studio leaves them untouched on Google, replies there, and asks permission before quoting any on its own site. AI's only job is to cut a long review down to a pull-quote, and the reviewer sees the pull-quote before it appears. Testimonial posts then fit naturally into a month of social posts once the approvals are in.

Four minutes before you post

Run this on every edited testimonial. It takes about four minutes, and roughly a minute of that is the first step.

1. Put the original and the final side by side. For each sentence in the final version, find the sentence in the original it came from.
2. Search the final text for numbers, timeframes and superlatives such as "best", "amazing", "never" and "completely". Each one has to appear in the original.
3. Look for anything medical: pain, a condition, medication, "cured", "healed", "fixed". It must be the client's own description of how they feel, with no cause attached, or it goes.
4. Check that the approval covers this exact text, this platform and the way you show their name.
5. If a result is in there, confirm the evidence is on file and that the timeframe and context appear in the text.
6. Confirm the testimonial still describes something you offer.
7. Save the original, the final text, the approval and the date in your log.

The test underneath all seven fits in a sentence. For every claim in the published version, can you point to the client's own words and their yes? If you can, publish. If you can't, the AI has written something the client didn't, and the edit has gone too far. For paid ads that carry result claims, take advice for your own market first, because the rules above vary and change.

Questions trainers ask before posting an edited testimonial

Can I copy a Google review onto my website as a testimonial?

A Google review stays the reviewer's own words, so ask before you copy it, show them exactly what you plan to publish and keep their reply. Quote it as written, apart from typo fixes they have approved, and never shorten a mixed review into a purely positive one. Leaving it on Google and linking to your profile needs no editing at all.

Do I have to say that AI tidied a testimonial?

None of the guidance covered here makes you label a typo fix as AI-assisted. It judges whether the published words are genuine and not misleading, and the client's approval of the final text is your protection. If AI did more than tidy, for instance a translation, say so, and never present AI-written praise as coming from a real client.

Can AI translate a client's testimonial into another language?

Yes, but translation shifts tone and sometimes meaning, so treat it as an edit that needs approval. Send the client the translated version if they read that language, or ask a fluent speaker to check it, and keep the original beside it. Label the quote as translated and run the same checks for added claims.

What if one client's result was far beyond what most people get?

Give the context. An exceptional result in an ad needs the timeframe, what the client actually did and what most clients can expect, and endorsement guidance says a bare 'results may vary' doesn't do that job. If you can't supply that context, quote what the training felt like instead of the number, or leave the result out.

How long can I keep using a testimonial?

The guidance covered here sets no fixed age limit, but the testimonial must still relate to what you offer now and still reflect the client's view. Review your testimonials once a year, remove any that describe a programme you no longer run, and take one down promptly if the client asks you to.

Further reads

• AI Prompts for Personal Trainers: Session Recaps and Renewal Texts — Prompt wording for client messages that stays inside what actually happened.
• AI Video for Salons: Turn Before-and-After Photos Into Short Reels — The same consent and honesty questions applied to before-and-after media.
• Health Food Store Marketing Ideas: AI Posts Without Health Claims — How to keep health claims out of AI-drafted posts.
• AI Agent for Personal Trainers: Check-Ins and Where It Must Stop — Where AI client check-ins must stop and a human step in.
• How to Use AI to Respond to Google Reviews — Replying to Google reviews without rewriting what the reviewer said.
• Personal Trainer Progress Reports: Monthly Client Summaries With AI — A worked month for one client: measurement table in, honest report out, with a three-step check so the AI never adds results you didn't record.
• Can AI Draft a Personal Trainer Contract? What a Solicitor Must Check — Turn your session rules into a personal trainer contract with AI, then see which clauses on cancellation, packs, injury and liability a solicitor must check.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Meta's Health and Wellness advertising standard (July 2026 change log); published advertising-code guidance on testimonials and on claims in testimonials; a consumer-protection regulator's final rule on fake reviews and testimonials, its questions and answers on that rule, and its endorsement guidance in question-and-answer form; Google's review policy.