Not if the words are invented. A testimonial written by AI and presented as a customer's counts as a fake review under consumer-protection rules in many markets, and it breaks Google's review policies. You can use AI to tidy a real customer's own words if they approve the final version, and to summarise genuine reviews accurately.
The test is easy to remember. A testimonial is a claim that a real person had a real experience and said this about it. Anything that breaks one of those three links (the person, the experience or the words) is on the wrong side of the line, however it's labelled. Calling an invented quote an "illustrative testimonial" doesn't rescue it; the whole point of a testimonial is that someone real said it. This is general guidance, not legal advice, and the rules vary by market and profession.
Where the line falls: allowed, grey and never
| What you do with AI | Where it sits | Why |
|---|---|---|
| Fix spelling and punctuation in a real customer's words, with their approval | Allowed | Same person, same experience, same meaning |
| Shorten a long testimonial, keeping the meaning, with approval | Allowed | As long as cutting doesn't change what they meant |
| Summarise your genuine reviews ("most reviews mention short waits") | Allowed if accurate | A fair summary of real content, not a new quote |
| Turn a customer interview into a case study they approve | Allowed | Their experience, their sign-off |
| Suggest topics a customer might mention in a review | Grey | Fine as prompts; not if you script their words |
| Polish wording so it sounds more enthusiastic | Grey to never | Adding warmth they didn't express changes the meaning |
| Cut the criticism out of a mixed review and quote the rest | Never, if it misleads | Selective editing can misrepresent their view |
| Write a testimonial and ask a customer to "approve" it | Never | The words aren't theirs |
| Generate testimonials from invented or "composite" customers | Never | No person, no experience |
| Post AI-written reviews on Google or other platforms | Never | Fake engagement, with platform penalties |
What the rules actually prohibit
You don't need to read the legislation to stay safe, but knowing what it covers explains why the table looks the way it does:
- The EU Unfair Commercial Practices Directive, amended in 2019 with the fake-review provisions applying from 28 May 2022, blacklists submitting or commissioning false consumer reviews or endorsements, and misrepresenting consumer reviews to promote a product. It also bans claiming reviews come from real customers without taking reasonable steps to check. Blacklisted practices are unfair in all circumstances, whatever the intent. If you sell to customers in the EU, this applies to you.
- A dedicated fake-reviews rule in another major market, in force since October 2024, covers reviews and testimonials attributed to people who don't exist, which its authors explicitly said includes AI-generated fakes, and to people who had no real experience of the business. It also covers undisclosed reviews by insiders, suppressing negative reviews, and buying reviews that express a particular sentiment, with fines per violation.
- Google's user-generated content policy prohibits fake engagement: content not based on a real experience, reviews paid for directly or in kind, and content posted from multiple accounts at one person's request. Businesses may not offer incentives for reviews. When Google finds a business has broken these rules, it can remove the reviews, stop the profile receiving new reviews for a period, unpublish existing ones, and show a warning to customers that fake reviews were removed.
The same thinking applies to social media endorsements. If you give someone a free treatment, product or discount and they post about it, the connection has to be disclosed in the post, and an AI-drafted caption you hand them to publish is your words in their mouth. Let them write it, and ask them to label it as a gifted or paid post.
Other review platforms have their own versions of the same rules. The common thread is that nobody regulates whether you used AI. They regulate whether the testimonial is true, from a real person, and fairly presented. AI simply makes breaking those rules much easier.
Three mistakes, and how each one came to light
These illustrative cases show that fabricated or over-edited testimonials rarely stay hidden, and that the discovery usually comes from an ordinary customer, not an investigator.
The recycled quote. A dental practice's marketing agency supplied five testimonials for its new website, "based on typical patient feedback". A patient searching for a second opinion found the same quote, word for word, on another practice's site built by the same agency. She mentioned it at her next appointment. The practice took all five down that day, but the patient's trust in everything else on the site had gone with them.
The upgraded claim. A pharmacy asked AI to "make this testimonial sound more professional". The customer had written, "The pharmacist sorted out my repeat prescription mix-up in ten minutes." The polished version read, "The pharmacist expertly managed my medication and resolved my prescription issues." The customer, seeing it on the pharmacy's page, pointed out that nobody had managed his medication; they'd fixed a paperwork error. The new wording also made a clinical-sounding claim the pharmacy hadn't intended. The original sentence was better on every count.
The helpful receptionist. At a podiatry clinic, a receptionist offered to post Google reviews for patients who said they were happy "but didn't have time", typing their comments into the clinic's tablet under new accounts. Google's systems noticed a cluster of reviews from one device and location, removed them and put a warning on the profile. Every review was sincere in spirit, and every one broke the rule that content must be posted by the person who had the experience.
Asking for testimonials that are worth using
Much of the temptation to have AI "improve" testimonials comes from asking for them badly. "Would you leave us a review?" gets "Great service, thanks!" Specific questions get specific, usable answers, and AI is good at writing the request. Ask for three short things: what the customer was worried about before they came, what actually happened, and what they would tell a friend thinking of coming. An illustrative request from a veterinary practice, sent a week after a routine operation: "We're glad Bella's recovering well. If you have a moment, we'd love to hear, in a sentence or two: what were you worried about before her operation, and how did it go? We'll only share it with your permission." The replies that came back talked about being phoned after surgery and nurses remembering the dog's name, which is far more persuasive than anything a model could invent, and needed nothing more than a spelling check.
The approved-edit workflow for real testimonials
This is the process I'd put in place for any business that wants to use AI with testimonials:
- Collect in the customer's own words. A short form, an email reply, or a voice note they send. Rough is fine; honest is the point.
- Let AI tidy, under strict instructions (the prompt is below). Its job is spelling, punctuation and length, nothing else.
- Compare the original and the tidied version side by side. Check every sentence still says what the customer said, with the same strength of feeling.
- Send the tidied version back for approval, saying where it will appear and whether you'll use their name.
- Keep a record of the original, the approved text, the approval date and where it's used.
- Review once a year. If the service has changed, or the customer asks, retire it.
The tidying prompt:
Below is a customer's testimonial in their own words. Correct spelling
and punctuation, and shorten it to under 60 words if needed. Rules:
do not add any claim, adjective or feeling that isn't in the original;
do not make it more enthusiastic; keep their phrasing wherever
possible; if cutting would change the meaning, keep it longer. Then
list every change you made, one per line.
[paste the original]Asking for a list of changes is the safeguard. It makes the model show its work, so you can spot additions quickly instead of rereading both versions word by word.
Before and after: a real testimonial, tidied properly
An illustrative voice-note transcript from a hearing-aid shop customer:
so yeah I was a bit nervous coming in because my last place was a bit pushy but the lady was really patient she explained everything twice and I didn't feel pressured at all and the hearing aids are good, the TV's much better, still getting used to them in restaurants though
An illustrative first AI tidy, before the strict prompt was used: "I was nervous at first, but the audiologist was incredibly patient and explained everything clearly. My new hearing aids have been life-changing. I can finally enjoy TV and dining out again!" Three problems: "life-changing" and "incredibly" are the AI's words, not his; "dining out again" reverses what he said about restaurants; and the exclamation mark adds a tone he didn't use.
The version produced with the strict prompt, and approved by the customer: "I was a bit nervous coming in, because my last place was pushy. She was really patient, explained everything twice, and I didn't feel pressured at all. The hearing aids are good. TV's much better; I'm still getting used to them in restaurants." It's less glossy and much more persuasive, because it sounds like a person. Keeping the restaurant caveat also makes it more believable to the next nervous customer.
A consent email you can adapt
Subject: Your kind words - OK to share?
Hi [first name],
Thank you for your feedback about [service]. We'd love to share it
on our website and social media. I've tidied the spelling only;
here's the version we'd use:
"[approved text]"
We'd show it as: [first name and initial / anonymously].
Could you reply "yes" if that's all right, or tell me anything you'd
like changed? You can ask us to remove it at any time.
Thanks,
[name], [business]Their reply is your approval record. File it with the original words.
Health businesses: stricter rules than most
If you run a dental practice, osteopathy or podiatry clinic, pharmacy or veterinary practice, the general consumer rules are only the floor. Professional advertising codes often go further. In some places, health professionals can't use patient testimonials about clinical care in advertising at all. In others they can, but must not imply that one patient's result is typical, and must have the patient's explicit consent to be identified.
So, before using any patient or client testimonial, AI-assisted or not, check your professional body's advertising guidance. Testimonials about the non-clinical experience ("the reception team were kind", "easy parking", "they explained the costs clearly") are usually lower-risk than ones about outcomes ("my back pain is gone"). An illustrative osteopath's testimonial page, rebuilt after checking the guidance, kept quotes about being listened to and the clear explanation of fees, and removed every quote about symptoms improving. Confidentiality matters too: even a first name plus a condition can identify someone in a small community. Writing case studies from customer interviews with AI covers the consent side for longer stories.
AI avatars, voice clones and "dramatised" testimonials
Some video tools offer to turn written testimonials into videos read by an AI presenter. The rules above still apply, plus two more:
- The presenter must never be presented as the customer. If an avatar reads a real customer's approved words, say so on screen: "A real customer review, read by an AI presenter."
- Never clone a customer's voice or face for a testimonial, even with enthusiastic permission. The result is indistinguishable from the customer saying words at your direction, which is the exact problem testimonial rules exist to prevent.
The broader decision about AI presenters is in whether to use an AI avatar as your business spokesperson.
Summarising reviews with AI without misleading anyone
Summaries are the most useful legitimate job for AI here: "What our patients say" sections, the opening of a services page, a line in an email. The risk is overstatement. A prompt that keeps it honest:
These are all our reviews from the last 12 months (star rating and
text). Summarise the themes that appear in at least 5 reviews, with
the count for each, including recurring criticisms. Then write a
two-sentence summary for our website that only states what the
counts support. Give the total number of reviews and the average
rating exactly.An illustrative result for a veterinary practice: 86 reviews, average 4.7; 31 mention nurses by name, 22 mention being seen quickly for emergencies, 9 mention prices being higher than expected. The website summary it drafted: "Across 86 reviews this year, clients most often mention our nurses and being seen quickly in an emergency." Accurate and fair. An earlier draft had said "hundreds of five-star reviews" and "loved by pet owners across the area". Neither was supported. Whenever a summary uses a number, check it against the platform. For turning genuine reviews into longer marketing copy, turning customer reviews into marketing copy with AI goes further.
The testimonial record: what to keep for each one
| Field | Example entry |
|---|---|
| Customer | Customer ref 2231 (first name and initial approved) |
| Original words | Voice note transcript, filed 14 Aug |
| Approved text | "I was a bit nervous coming in..." (48 words) |
| Changes made | Spelling, punctuation, shortened; list of changes attached |
| Approval | Email reply "yes, that's fine", 19 Aug |
| Where used | Website home page, hearing-test page, one social post |
| Review date | August next year, or sooner if the service changes |
It takes two minutes per testimonial, and it's exactly what you'd want to hand over if a platform, a customer or a regulator ever asked where a quote came from.
Red flags in your own marketing process
- Anyone on your team writing reviews or testimonials "to get things started".
- An agency or tool offering "review generation" or "testimonial writing".
- Testimonials with no name, date or record behind them.
- Only showing reviews above four stars while implying they're all your reviews.
- Asking only happy customers for reviews, which some platforms treat as review gating; what review gating is and how it gets businesses penalised explains the line.
- Quotes that sound more polished than anything your customers actually write.
Ask a solicitor or your professional body before you use paid endorsements or incentives, publish health outcome testimonials, compare yourself with named competitors, or if you find AI-written testimonials already live in your marketing. For everyday testimonials, the approved-edit workflow above keeps you on the right side of the line without needing advice each time.
Testimonials and AI: the questions that follow
Can AI write a review that a happy customer then posts under their name?
Avoid it. Even if the customer agrees, a review is supposed to be their own account, and platforms treat content not written from genuine experience as fake engagement. The safer version is to ask for their words, even a rough sentence or two, and let them post it themselves. You can suggest what to mention, such as the service they had, but not write it.
Is it all right to offer a discount in exchange for a review?
Google's policy bars incentives for reviews on its platform, and some consumer rules ban rewards tied to positive sentiment or undisclosed incentives. Asking every customer for an honest review, with no reward, is the safe approach. If you run a genuine incentive elsewhere, it must not depend on the review being positive and must be disclosed wherever the review appears.
Can staff or family post reviews or testimonials?
Not as if they were ordinary customers. Reviews by employees, relatives or anyone with a connection to the business are treated as misleading unless the connection is clearly disclosed, and most review platforms don't allow them at all. Staff can share genuine customer reviews on social media, attributed properly, but shouldn't write their own.
We've already published AI-written testimonials. What now?
Take them down from your website, social profiles and any ads now, and don't replace them until you have genuine, approved ones. Keep a note of what was removed and when. If they were used in paid advertising, in a regulated profession, or you've had a complaint, speak to a solicitor or your professional body about whether anything more is needed.
Further reads
- Can You Legally Use AI-Generated Images in Your Marketing? — The parallel legal question for AI-generated images in marketing.
- How to Handle a Fake or Unfair Review With AI's Help — What to do when someone posts a fake review about you.
- How to Back Up Every Claim AI Writes in Your Marketing — Evidence every claim AI adds to your marketing copy.
- How to Write an AI Content Policy for Your Marketing — Put your testimonial rules into a written marketing policy.
- Patient Reviews and AI Replies: Staying Within Confidentiality — Health businesses: replying to patient reviews confidentially.
- AI Marketing for Small Law Firms: Content, Reviews and the Rules — How a regulated profession handles reviews and AI content.
- How to Write an AI Disclosure Statement for Your Website — An AI-use inventory, the five parts of a good statement, a fill-in template, a picture framer's example and wording to avoid.
- AI Ethics for Small Businesses: A Practical Checklist — Twenty-five questions in six groups, each with why it matters and how to check, plus red lines and a nursery example run end to end.
- Should You Tell Customers You Use AI? A Disclosure Guide — When to tell customers you use AI and when you needn't, with a decision table for 11 everyday uses and wording you can copy for chats, quotes and emails.
- AI Mistakes That Damage Customer Trust, and How to Avoid Them — Nine AI mistakes customers notice, why each one stings, how to prevent it, and a 20-minute monthly check that catches problems before customers do.
- Online Shop AI Mistakes That Hurt Trust and Conversions — Eight AI mistakes that quietly raise returns and lower sales in small online shops, each with a real-looking example and the fix.
- AI Mistakes That Put Couples Off Your Wedding Business — Eight ways AI makes a wedding supplier look careless or fake to couples, each with an example from real-looking enquiries and the fix.
- Red Flags When Buying AI Marketing Software or Services — Fifteen warning signs in AI marketing quotes and contracts, each with what it looks like, how to check it and a marked-up deli quote.
- How Much Does an AI Marketing Video Cost Compared With a Shoot? — Itemised costs for AI-made and filmed marketing videos, a worked 30-second launch video, and the hidden costs on each side.
- How to Fact-Check AI Marketing Copy Before It Goes Live — A claim-by-claim checklist for AI-written ads, emails and product pages, with a fact bank template, red-flag words and a sign-off log.
- How to Write Your Website Copy With AI: Home, About, and Services — Page-by-page prompts for your Home, About and Services copy, built from a fact sheet, with sample outputs, a claims check and a 20-minute customer test.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: the EU Unfair Commercial Practices Directive as amended by Directive (EU) 2019/2161 (Annex I, fake reviews); published summaries of a 2024 trade rule on consumer reviews and testimonials; Google Maps user-generated content policy (fake engagement) and Google Business Profile Help on restrictions for policy violations.