Tomasz trains dogs in Gdańsk. Good at it, too: the reactive spaniel that lunged at every bicycle now walks past them like a retired diplomat.
His website has a testimonials section. It holds twelve reviews, all five stars, all lovely. "Best trainer ever!" "Highly recommend!!" "Thank you Tomasz!" He picked them himself from the messages he gets, and he left out the one from the man who thought the course was too expensive.
A friend who runs a café looks at it and frowns. "It's very nice," she says. "It also looks like your mum wrote it."
She has a point, and so does the research. The way social proof is presented can make it more convincing or less, and some of the most common habits, including the one Tomasz has just practiced, are now illegal in several countries.
Tomasz and his spaniels are imagined. The studies, figures, laws and enforcement cases below are real.
Why we follow others
In his 1984 book Influence, psychologist Robert Cialdini named social proof as one of six principles of persuasion. His point was simple: one way we decide what is correct is to look at what other people are doing.
Cialdini argued that social proof works most strongly under two conditions: when we are uncertain about what to do, and when the people we observe seem similar to us. The book draws on many experiments and field observations, his own and others'. It is a synthesis written for a general audience, not a single study, so individual claims are best checked against the research it cites.
Both conditions apply almost perfectly to a website visitor. They are uncertain, because they have never used you. And reviews are written by people who were once in their exact position.
The classic demonstration of how strong the pull can be is older still. In Solomon Asch's conformity experiments, published in full in 1956, participants judged which of three lines matched a reference line, after hearing several other "participants" (actually confederates) give a unanimously wrong answer. Across the critical trials, the 123 real participants went along with the wrong majority 36.8% of the time.
The finding has held up better than many classics. A 2023 replication by Andreas Franzen and Sebastian Mader with 202 valid participants found an error rate of about 33% in the standard version. When the researchers paid people for correct answers, it fell to about 25%.
Two cautions before applying this to reviews. Asch's task had an obvious right answer, and the pressure came from people physically in the room. Online, the pressure is weaker, but the uncertainty is much greater: there is no "correct line" for whether a dog trainer is good. That is the situation in which Cialdini expected social proof to matter most.
Do this: ask where your visitors are most uncertain. Put social proof there, not in a separate "Testimonials" page that the uncertain visitor never opens.
The towel study, precisely
If you have read anything about social proof, you have met the hotel towels. The study is worth getting right, because it is often summarized with the wrong numbers and without the most interesting part.
Noah Goldstein, Robert Cialdini and Vladas Griskevicius ran two field experiments in a mid-sized, mid-priced hotel in the US Southwest, published in the Journal of Consumer Research in 2008. Guests did not know they were in a study. Room attendants simply recorded whether guests reused at least one towel.
Experiment 1 compared the hotel industry's standard card ("Help save the environment") with a card saying that almost 75% of guests who were asked took part in the towel reuse program. On the analyzed sample of 433 observations, reuse was 35.1% with the standard card and 44.1% with the descriptive norm card.
Experiment 2, with 1,595 observations, is where it gets interesting. It tested the standard card against four versions of the same "75% participated" norm, each attributed to a different group.
All four norm messages combined (44.5%) beat the standard message (37.2%). But the winner was the strangest one. "75% of the guests who stayed in this room" produced 49.3% reuse, significantly more than the other three norm versions combined (42.8%). The authors called this a provincial norm: the behavior of people in your immediate situation, even a group as meaningless as "previous occupants of room 313".
Two details are usually left out.
First, the effects are modest. The best sign moved reuse by about twelve percentage points over the standard card, in one hotel. Second, a 2014 replication by Gerd Bohner and Lena Schlüter in two German hotels found that both standard and norm messages increased reuse compared with no message, but the norm messages were not more effective than the standard one, and the proximity effect was inconsistent. Social norms matter, but the size of the effect depends on context.
And one more detail for anyone tempted to borrow the technique. The researchers' 75% figure came from the company that supplied the cards, and in Experiment 2 it was attributed to a study "conducted in Fall 2003". In a lab or a research protocol, that is a design choice. On your website, a social proof number has to be true. More on that in the law section.
Do this: make your social proof as close to the visitor's situation as you can. "Most of my clients are first-time dog owners in Gdańsk" is a provincial norm. "Thousands of happy customers" is not.
What reviews do to sales
The most quoted numbers on reviews come from the Medill Spiegel Research Center at Northwestern University, which partnered with the review platform PowerReviews and published How Online Reviews Influence Sales in 2017.
The first number needs its context. It comes from one high-end gift retailer: about 15.5 million page views of 1,800 products by 7.8 million users over a year, tracking sales as each product accumulated reviews. The report adds that the benefit of extra reviews diminishes quickly after the first five. Lower-priced products came from two other retailers' consumer-goods data across 111,460 SKUs.
This is observational data from retailers, analyzed by academics in partnership with a review vendor that has an interest in the result. It is good evidence that reviews matter, especially for expensive or risky purchases. It is not a promise that adding five reviews to your page will multiply your enquiries by 3.7.
The price finding is the one most relevant to small service businesses. A dog-training course, a website, a renovation: these are higher-priced, high-uncertainty decisions. That is exactly where the report found reviews made the most difference.
Do this: if your service has no visible reviews at all, getting the first handful is probably worth more than polishing anything else on this page.
Why 5.0 looks suspicious
Now the counterintuitive part. Across the Spiegel data, purchase likelihood typically peaked for products rated between 4.0 and 4.7 stars, then fell as ratings approached 5.0. Products rated 4.7 to 5.0 were less likely to be bought than those rated 4.2 to 4.7. In no category was the best-performing rating a perfect 5.0. The underlying paper, by Ewa Maslowska, Edward Malthouse and Stefan Bernritter (2017), is titled, fittingly, "Too Good to be True".
The explanation is about credibility. Real experiences vary. A wall of perfect scores suggests either very few reviewers or a curated selection, and shoppers know it. The report cites earlier PowerReviews research that 82% of shoppers specifically seek out negative reviews, a vendor figure, but consistent with the pattern.
Volume works the same way in reverse. Nielsen Norman Group's Jen Cardello described a usability session in which a participant decided an article was unpopular because "only" 1,000 people had shared it. Her summary: "The most significant risk with using social proof is the perception that too few people approve." A tiny number can be weaker evidence than no number.
Two trainers, same price. Which one would you contact first?
Trainer B
All reviews shown. Collected by email invitation after every completed course; only clients who booked can review.
The trainers and ratings are fictional. Trainer B's distribution adds up to 480 reviews averaging 4.6.
The demo is a thinking tool, not an experiment: nobody has tested these two cards. But it combines three things the evidence does support. Ratings just under perfect read as more credible, small counts can look like weak support, and showing the full distribution and the collection method gives visitors a way to check rather than simply believe.
Do this: show your real average, your real count and your negative reviews. Answer the negative ones calmly. A short, fair reply to a two-star review can say more about benevolence than ten five-star ones (the trust chapter explains why).
Logos, numbers or words
Reviews are only one kind of social proof. Small businesses also use client logos, counts ("400 dogs trained") and testimonials. Which works best?
Honest answer: there is no good study comparing them head-to-head on landing pages that I could verify. What we can do is ask what each one proves, using the trust questions from the previous chapter. Each form answers a different doubt.
Strong when visitors recognize the names and the clients resemble them. Weak or confusing when nobody knows the logos. Only show clients who agreed to it.
"400 dogs trained since 2018" signals experience. Small or round numbers can backfire. Every number must be true and current.
Show the experience in the client's own words. Persuasive when specific and attributed. Read as curated, because they are.
Google, Trustpilot or industry platforms. Harder to fake, so more credible. NN/g found people trust external sources more than company-sponsored content.
The last point comes from Nielsen Norman Group's long-running credibility research: users check review sites, social media and news before trusting a company, and they trust "these external sources more than company-sponsored content". Linking to an independent profile, even with imperfect ratings, signals that you are not hiding anything.
Do this: use at least one form of proof you do not fully control. A link to your Google profile with its real rating is worth more than a carousel of hand-picked quotes.
Specific testimonials win
Tomasz's testimonials have the most common problem: they are praise, not evidence. "Best trainer ever!" tells the visitor that someone was happy. It does not tell them whether their problem, with their dog, is the kind Tomasz solves.
In eye-tracking work summarized by Nielsen Norman Group's Therese Fessenden, people paid attention to the testimonial author's job title and affiliation, and reacted most strongly to testimonials that started with a reservation and ended in confidence: the "I didn't think this would work for me, but..." pattern. That research was about business buyers, but the logic carries over. Identity makes a quote checkable. A doubt overcome mirrors the doubt the reader has right now.
"Best trainer ever!! Highly recommend!"Anna K.
"Thank you so much Tomasz, amazing!"Satisfied customer
"I was sure Bruno was too old to change. He's eight and lunged at every bike. After six sessions we walk the Motława embankment without drama."Anna Kowalska, owner of a springer spaniel · Gdańsk-Oliwa · 6-week course, 2026
Figure 3. The same client, before and after asking better questions. Fictional example.
You get specific testimonials by asking specific questions. Instead of "Would you leave us a review?", try three prompts: What were you worried about before you started? What changed? Who would you recommend this to? Use the client's own words, ask permission to show their name, and do not edit the meaning.
Do this: rewrite your request for testimonials as those three questions, and send it to your last five happy clients this week.
Fake reviews are now illegal
Here is why Tomasz's café-owning friend was more right than she knew. Over the last few years, the EU, the US and the UK have all moved from general rules against misleading advertising to specific rules about reviews.
This section is general information, not legal advice. Which rules apply depends on where you and your customers are. Check your own jurisdiction or ask a lawyer.
Since 28 May 2022, Article 7(6) of the Unfair Commercial Practices Directive treats as material information "whether and how" a trader that provides access to consumer reviews ensures they come from consumers who actually used or bought the product. Leaving it out can be a misleading omission. Two blacklisted practices were added to Annex I: point 23b, claiming reviews come from real users without taking reasonable and proportionate steps to check, and point 23c, submitting or commissioning false consumer reviews or endorsements, or misrepresenting reviews or social endorsements, to promote products. Fines in coordinated cross-border actions must reach at least 4% of annual turnover, or €2 million where turnover data is unavailable.
Source: Directive 2005/29/EC, consolidated text, Art. 7(6), Annex I points 23b–23c, Art. 13. Not legal advice.
Tomasz's habit of hand-picking the nice ones is the subtle trap. Selecting reviews is not automatically illegal in the EU, but presenting a curated selection as if it were the whole picture can be "misrepresenting consumer reviews", and he now has to say how the reviews he shows were collected.
The FTC's Trade Regulation Rule on the Use of Consumer Reviews and Testimonials prohibits: writing, selling or knowingly buying fake reviews and testimonials, including from people who do not exist or never used the product (§465.2); giving incentives conditioned on a positive or negative review (§465.4); undisclosed reviews and testimonials by officers, managers and employees (§465.5); company-controlled "independent" review sites (§465.6); suppressing reviews through groundless legal threats or intimidation, or displaying only the good ones while implying they are all of them (§465.7); and buying or selling fake followers and views (§465.8). Because it is a rule, knowing violations can bring civil penalties. The current maximum is $53,088 per violation: the FTC announced in September 2026 that there would be no inflation adjustment for 2026 and that 2025 levels continue to apply.
Source: 16 CFR Part 465 on eCFR; Federal Register, 22 August 2024; 16 CFR 1.98; FTC notice 2026-18853. Not legal advice.
Note what §465.4 does and does not forbid. Asking every customer for a review, even with a small thank-you, is not banned in itself. Offering the thank-you only for a positive review is. The FTC's revised Endorsement Guides of June 2023 (16 CFR Part 255) add guidance, though not binding rules, on incentivized reviews, employee reviews and procuring, suppressing, boosting or editing reviews in ways that distort what consumers think.
In force since 6 April 2025, paragraph 13 bans submitting or commissioning fake consumer reviews or reviews that conceal they were incentivized, publishing reviews or review information in a misleading way, and publishing reviews without reasonable and proportionate steps to prevent and remove fake or misleading ones. "Misleading" expressly includes removing negative reviews while publishing positive ones, giving positive reviews more prominence, and omitting that a reviewer was paid. Under the same Act, the CMA can now decide breaches itself and fine businesses up to 10% of global turnover or £300,000, whichever is greater.
Source: DMCC Act 2024, Schedule 20 (commenced by SI 2025/272); CMA guidance on direct enforcement. Not legal advice.
The UK wording is the clearest of the three on curation. Under paragraph 13, giving "greater prominence to positive consumer reviews over negative ones" is an example of publishing in a misleading way. A testimonials section with only the best quotes can still be fine if it is clearly presented as a selection and sits alongside access to the full, unfiltered reviews.
What enforcement looks like
Fashion Nova
The FTC alleged the retailer automatically posted four- and five-star reviews while holding lower-rated ones for approval, and from late 2015 to November 2019 never posted hundreds of thousands of them. It agreed to pay $4.2 million; the order was finalized in March 2022.
FTC press release, 25 January 2022First warning letters
On 22 December 2025, the FTC sent warning letters to ten companies over possible violations of the reviews rule, flagging fake reviews and incentives for five-star reviews, and reminding them of penalties of up to $53,088 per violation.
FTC business blog, 22 December 2025TruHeight
The FTC charged the supplement seller with violating the Reviews and Testimonials Rule, alleging several thousand five-star reviews written by employees posing as customers until at least November 2024, plus free or discounted products in exchange for five-star reviews. The final order of 15 July 2026 includes a $4 million judgment, suspended after payment of $750,000.
FTC, 13 April 2026 · 15 July 2026CMA sweep and investigations
After a website review of over 100 businesses in 2025, the CMA wrote to 54 that might not comply with its fake reviews guidance, for example because they had no published policy on fake or incentivized reviews. In March 2026 it opened investigations into Autotrader, Feefo, Dignity, Just Eat and Pasta Evangelists. As of its September 2026 update these were ongoing, and the CMA stresses it has made no finding of a breach.
CMA case page · press release, 27 March 2026The pattern across all four is worth noticing. Regulators are not only chasing fake review factories. They are looking at ordinary business habits: hiding the bad ones, rewarding only the good ones, asking staff to chip in, offering a discount for five stars. The CMA's sweep found that more than half the businesses it checked lacked a clear, accessible policy. A one-paragraph review policy on your site is a cheap way to be on the right side of that line.
What to change tomorrow
Show the real numbers
Display your actual average and count, link to the full set on an independent platform, and stop filtering out the critical ones. A 4.6 from many people is credible; a 5.0 from a few looks curated.
Write a short review policy
Say how you collect reviews (for example, "we invite every client after the course ends"), whether you check that reviewers are real clients, and that you publish negative reviews too. This is what UCPD Article 7(6) asks for, and the CMA expects such policies.
Never pay for a sentiment
Ask everyone, not just the happy ones. If you offer a thank-you, offer it regardless of the rating, and require incentivized reviews to say so. No reviews from staff or family without disclosure.
Ask for evidence, not praise
Use three questions: what worried you, what changed, who would you recommend this to. Attribute with name, context and date, with permission.
Make proof provincial
Place the most relevant testimonial beside the decision it supports, and choose ones from people like the visitor: same city, same problem, same kind of business.
Tomasz added a link to his Google profile: 4.7 stars from 63 reviews, including the man who thought it was expensive (and Tomasz's polite reply). He swapped three of his testimonials for longer ones with names and dogs' ages. Enquiries did not double overnight. But the next client said, without being asked, "I liked that you had a couple of grumpy reviews. It made the rest believable."
What the research supports
Social influence is one of psychology's most replicated findings, and Asch's basic effect has been reproduced recently. Field evidence on specific norm messages is more mixed: the towel study's provincial-norm effect did not replicate cleanly in Germany. The Spiegel review figures are large-scale but observational, from a few retailers in partnership with a review vendor, and should not be read as predictions for your site. The "too perfect" rating effect is consistent across their categories but comes from product reviews, not services. There is little rigorous research comparing logos, numbers and testimonials directly. The legal rules, by contrast, are clear and in force.
Sources
- Cialdini (1984). Influence: The Psychology of Persuasion (first published as Influence: How and Why People Agree to Things). William Morrow. Consulted for the social proof principle and its conditions.
- Asch (1956). Studies of independence and conformity: I. A minority of one against a unanimous majority. Psychological Monographs, 70(9). Figures as reported in Franzen & Mader (2023).
- Franzen & Mader (2023). The power of social influence: A replication and extension of the Asch experiment. PLOS ONE. Full text consulted for the original 36.8% figure and the replication results.
- Goldstein, Cialdini & Griskevicius (2008). A Room with a Viewpoint: Using Social Norms to Motivate Environmental Conservation in Hotels. Journal of Consumer Research, 35(3), 472–482. Full text consulted for all percentages, sample sizes and sign wording.
- Bohner & Schlüter (2014). A Room with a Viewpoint Revisited: Descriptive Norms and Hotel Guests' Towel Reuse Behavior. PLOS ONE, 9(8), e104086. Abstract consulted for the replication results.
- Medill Spiegel Research Center (2017). How Online Reviews Influence Sales. Full report PDF consulted for the 270%, 190%/380%, 15% and 4.0–4.7 findings and data descriptions, including the cited Askalidis & Malthouse (2016) and Maslowska, Malthouse & Bernritter (2017) papers.
- Cardello (2014), Nielsen Norman Group. Social Proof in the User Experience. Consulted for the low-count backfire observation.
- Harley (2016), Nielsen Norman Group. Trustworthiness in Web Design: 4 Credibility Factors. Consulted for the finding on external sources.
- Fessenden (2019), Nielsen Norman Group. What B2B Designers Can Learn from B2C About Building Trust. Consulted for testimonial attribution and the reservation-then-confidence pattern.
- Directive 2005/29/EC as amended by Directive (EU) 2019/2161. Consolidated text on EUR-Lex. Article 7(6), Annex I points 23b–23c and Article 13 consulted.
- US Federal Trade Commission. 16 CFR Part 465 and final rule, 89 FR 68034 (22 August 2024). Rule text and effective date consulted.
- US Federal Trade Commission. 16 CFR 1.98 and Civil Penalty Inflation Adjustments notice (15 September 2026). Consulted for the $53,088 figure and the cancelled 2026 adjustment.
- US Federal Trade Commission (2023). Updated Endorsement Guides announcement. Consulted for the 2023 changes to 16 CFR Part 255.
- Digital Markets, Competition and Consumers Act 2024. Schedule 20 on legislation.gov.uk. Paragraph 13 and commencement notes consulted.
- Competition and Markets Authority. Online consumer reviews case page, Feefo case timetable and direct enforcement guidance. Consulted for the website review, investigations, status and penalty caps.
- FTC enforcement: Fashion Nova (2022), warning letters (2025), TruHeight (2026). Press releases consulted.
Sources checked 18 October 2026. Enforcement summaries describe allegations and orders as stated by the agencies; investigations listed as ongoing had reached no finding at that date. Only publicly available material was consulted. The trainers, testimonials and ratings in examples are fictional; the figures are original illustrations.
Trainer A
Reviews chosen by the owner from messages received. Not verified.