Trustpilot - surely enough is enough?

As regular readers will know, we monitor all review channels on a continuous basis. For the benefit of our clients and their customers. Our only loyalty is to both of these constituencies. If a new and better solution to reviews emerges, or a currently great review solution makes a misstep, we need to know so we can advise our clients accordingly. We have no contract or financial arrangements with any of these solutions, and we never will. 

Trustpilot - again

One such solution is Trustpilot, the Danish-based and London-quoted review site. It makes much of its AI fake-review detection. 

We recommend you park this video for later and read on (we have viewed all 40 minutes on your behalf); in this article we are just going to test one sentence from Anoop Joshi, Chief Trust Officer at Trustpilot, and that is his assertion that 'it is incredibly difficult to spot what is genuine and what is fake'. and remeber when you read our findings that we have no access to any of the multitude of data points that Trustpilot's software is able to pick up

So we decided to put their much-vaunted AI to the test by giving two of our moderators a challenge: find us 20 definitely fake or fraudulent reviews on Trustpilot in half an hour. Here are the results.

Oh, and while you're at it...

Now another, this time based in the UAE...

Red Flags

Almost too many to count, some apply to all, some to individual reviews...

• All reviews were written on the same day
• None of the reviews are of businesses in the reviewer's home country
• The spelling and grammar are suspect
• The content of the review bears no relation to the product/service reviewed
• The content of the review is patently irrelevant 
• The spectrum of reviewed businesses is untypical of a genuine reviewer: parking services in San Francisco, a T-shirt retailer in South Carolina, an AI tech platform (!) based in SF, L Ron Hubbard's publisher in LA, a quilt manufacturer in the Italian Dolomites, a Tennis coaching business in Queensland, a music label in Berlin, a baby stroller rental company in Orlando, and the list, as you can imageine, goes on in a similar - and similarly unlikely - vein.

Take the first review written by 'Ganesh Awate': it purports to be of Hollywood Mirrors, a makeup mirror supplier based in South Yorkshire. The review begins 'very comfortable...' A 'comfortable' mirror? His second review is of Photon Brothers, a solar panel installer from Denver, Colorado. What would you expect a review of solar panels to say? 'Great installation'? 'These panels saved me money'? No such thing. Instead, we find gibberish.

Fazullah Mohammed - who has written 262 reviews on Trustpilot - is a 'verified user'...

...and has written 282 reviews, all the same (nonsense), this one for Mochi health is particularly interesting...

...because it even elicited a thank you from the business, which is intriguing, since they almost certainly paid for the review in the first place, which leads one to wonder...

...just how many more of those nearly 15,000 reviews are genuine?

We took a closer look at both these businesses and make the following comments:

The great upside of AI is that once developed and implemented, it is effectively free (for the business); on the other hand, human moderators, such as those here at HelpHound are highly trained and experienced salaried specialists. But we have always been very much aware that the reviews we moderate are used by consumers to make potentially life-changing or decisions that will have long-term impacts on their financial and even physical well-being; reviews such as those you see above would never be relied on by anyone in their right mind when choosing any kind of business or service, but that's the core issue: Trustpilot reviews are virtually invisible once posted, they are rarely returned in Google searches - Google reviews are what the majority of consumers see; it is the overall score that businesses are looking for, to bolster the kind of marketing you see above

1. Our moderators found these reviews, undoubtedly all fake and paid for, within minutes (the individual reviews within seconds), yet they evaded Trustpilot's AI moderation
2. These reviews are not difficult to spot with the naked eye
3. We use our own AI for basic functions such as identifying foul language in reviews, but we would not dream of using it to moderate reviews 
4. None of the above reviews would make it through HelpHound moderation. 

In addition, Trustpilot already admit they remove millions of fraudulent reviews. This raises another question: just how long do these reviews remain on the site before removal?

Both the company listings and reviewer accounts remain live on Trustpilot as of today's date; we will monitor them going forward to see what action, if any, is taken.

Conclusion

Adrian Blair, current CEO of Trustpilot is quoted in the article above as saying 'More and more people are using us every month because they see us as a trusted source. A lot of people are saying just that in their Trustpilot reviews.

“And why is that? Because we go to enormous lengths to ensure that the content on Trustpilot is trustworthy.

“We have more than 350m reviews on our platform. Every one of those reviews has metadata attached to it.

None of the above reassures us, or should reassure users - business or consumer - of Trustpilot. Or the CMA. 

The fact that 'more and more people are using [Trustpilot]' is simply a factor of its sales success, not because consumers 'see [Trustpilot] as a trusted source'. If the likes of Octopus Energy (740,000 reviews on Trustpilot and counting) are using Trustpilot of course 'more and more people are 'using' it. But we need to define 'using'.In the main this 'using' is simply the act of writing a review when requested by the company, not, as in the case of Google, reading those reviews (Trustpilot reviews are surprisingly hard to find in search, unless one goes directly to their site, which we are sure few people do - except, as perviously mentioned, to write a review.

'Enormous lengths'? We feel we have already covered that point.

'Metadata attached'? We've said it before: metadata -  time of writing the review, location of the review writer's IP, IP address, device, etc. - might be useful to support real-world moderation, but in and of itself, it is very nearly useless. 

This headline should ring alarm bells for any CEO. More so, the readers' comments, which are mostly in this vein...

The 'likes' only serve to reinforce the comments

We - and many others - have been highly critical of many aspects of Trustpilot's business model, modus operandi and systems for many years now. Because the whole universe of reviews stands and falls on their credibility. 

Being able to trust reviews is vital, maybe not so much for online retail (you can return most products bought online these days - irritating, but you remain unharmed), but for an oncologist? A wealth manager? A family lawyer? We are sure you are getting our drift

Companies such as Yelp, in the US (no longer trading in the UK and EU thanks to compliance issues too many to mention in this article), that have prioritised traffic over moderation will, eventually come to the attention of the "woefully underfunded and therefore underresourced" (we quote) CMA - which recently announced AI all of its own to track down companieds that host fake reviews. 

And when they do, we hope they wish they had used some of their investors' cash and/or profits to moderate the reviews they expect consumers to trust and rely on to choose really important services.

More from the web...

• Fake Review Report

• Money Saving Expert

• An interesting Reddit thread

• Grizzly research's report