---
title: 683 reviews told me which competitor to attack, and which one to leave alone
description: AnyDesk scores 1.6 on Trustpilot and 4.6 on Capterra. What 683 hand-classified reviews say about which competitor to attack, and which one to leave alone.
image: https://go.aiento.eu/hubfs/og-review-mining-2026.png
---

[![AIento](https://aiento.eu/wordmark.png)](https://aiento.eu) [Lab](https://aiento.eu/lab) [Work](https://aiento.eu/#work) [About](https://aiento.eu/#about) [Blog](https://go.aiento.eu/blog) [Contact](https://aiento.eu/#contact)

[All posts](https://go.aiento.eu/blog)

# 683 reviews told me which competitor to attack, and which one to leave alone

João Diogo 13 September 2026

AnyDesk scores 1.6 out of 5 on Trustpilot. It scores 4.5 on G2 and 4.6 on Capterra.

Same product, same year, three stars apart depending on where you look. Before you use a competitor's public rating to decide anything, that spread is worth ten minutes, because it is not an anomaly. It is the mechanism.

If I joined a remote-access vendor tomorrow, the first thing I would build is a daily feed of my competitors' one and two star reviews. Not to find the reviewers. To find out what to say, to whom, and in which month. Here is what came out of running it across TeamViewer, AnyDesk and Splashtop.

## The trigger, and why this one

Most triggers answer "who". Funding rounds, job changes, technographics, visitor de-anonymisation. Very few answer "what to say", which is usually what decides whether the email works at all.

Competitor detractor reviews are the only free source where your competitor's customers describe, dated and in public, exactly what they hate about the thing you are trying to replace. It is your competitor's objection-handling document, written by the people who lost the argument.

The build is small. Pull each competitor's review feed daily, diff on review ID, send anything new through one model call that fills a fixed schema, write it to a table. Vendor, date, stars, country, whether the vendor replied, complaint theme, whether the complaint is about money or the product, and whether the reviewer says they are leaving.

That last set of flags is the whole thing. Everything downstream is a query.

## The rating gap, and what it is actually measuring

|  | Trustpilot | G2 | Capterra |
| --- | --- | --- | --- |
| TeamViewer | 4.6 / 127,521 | 4.5 / 4,172 | 4.6 / 11,639 |
| AnyDesk | **1.6** / 1,385 | 4.5 / 1,101 | 4.6 / 1,797 |
| Splashtop | 4.2 / 390 | 4.8 / 519 | 4.7 / 735 |

Eleven of those twelve cells sit between 4.2 and 4.8. One reads 1.6.

AnyDesk is not rated badly across the market. It is rated badly on one platform, and that platform is the one where anybody can post without proving they are a customer. G2 and Capterra verify business identity and fill largely through solicitation. Trustpilot is self-initiated, and in this category that means consumers and prosumers.

AnyDesk enforces its commercial-use licensing aggressively, including a connection delay for accounts it suspects. The population that enforcement angers is precisely the population that writes unprompted Trustpilot reviews. Its paying business customers are on the other two platforms giving it 4.5 and 4.6.

The corroboration is that both G2 and Capterra independently list false-positive commercial-use detection among AnyDesk's and TeamViewer's named drawbacks. The same mechanism shows up in the business population, just without the fury, because a company that pays for a licence is never accused of not having one.

Trustpilot also tells you TeamViewer replies to 98% of negative reviews within about 48 hours and pays for a business profile. A public rating is an output of a review operation. Read it that way or it will mislead you.

## The three rates that survive normalisation

Every figure below is a share of that vendor's own detractors, so company size cancels out. That matters when one vendor has 127,521 lifetime reviews and another has 1,385.

**Complaint is about money, not the product.** TeamViewer 31%. AnyDesk 85%. Splashtop 57%.

**Reviewer states they are leaving.** TeamViewer 23%. AnyDesk 36%. Splashtop 43%.

**Vendor publicly replied.** TeamViewer 99%. AnyDesk 24%. Splashtop 36%.

One note on how those numbers were produced, because it changed them. The first pass used keyword rules and left 368 of 628 rows with no theme at all. Every one of those rows had text. The rules simply had no pattern for "disconnects constantly" or "slowwww motioonnnn". I re-read all 368 by hand. 337 resolved, 31 stayed genuinely unlabelable, and TeamViewer's commercial share moved from 16% to 31% while AnyDesk's moved from 59% to 85%. If the biggest bucket in your theme distribution is "other", you do not have a theme distribution.

## Two competitors, two opposite cold emails

**Against TeamViewer, lead on reliability.** Product reliability is 208 of 510 rows, 41%, the largest single theme in the whole dataset. Licensing is second at 84 and usability third at 84. The angry TeamViewer customer is watching a session freeze, not reading a contract.

The language is in the data and it is unusually consistent: sessions that drop mid-support-call, screens that connect but never render, the reconnect loop, the black screen. Those are not adjectives, they are the sentences people actually wrote, and any sysadmin who has lived through them recognises the description in about four words. An opener about pricing lands on someone who was not thinking about pricing.

**Against AnyDesk, lead on contract terms and do not mention the product.** Of 118 rows: 31 cancellation blocked, 28 licensing, 17 vendor conduct, 9 auto-renewal, 7 price increase, 4 debt collection, 3 refund refused. Product reliability is 9 rows, 8%.

So the opener is about what happens at renewal and how you leave, stated before the reader needs it. The September 2026 reviews are still the same complaint: a price moving from €70 to €275 against a promised freeze, charges continuing after cancellation, a €500 collection letter.

Same category, same buyer title, two messages that share nothing.

**One correction to that advice, which the G2 and Capterra data force.** The Trustpilot anger at AnyDesk is concentrated in individuals and very small businesses caught by commercial-use enforcement. If your ICP is mid-market IT, those reviews describe a population you are not selling to, and the contract-terms opener will land softly on a 200-person company whose AnyDesk account was never in dispute. Use the theme, check the segment.

## The competitor not to attack

This is the part for a CRO rather than a campaign owner.

Splashtop's own detractors are 57% commercial. Closer to AnyDesk's 85% than to TeamViewer's 31%. Two of its reviewers independently use the phrase "bait and switch" about a trial that shows monthly pricing and offers annual only at checkout.

So a Splashtop attack on AnyDesk's pricing and renewal terms is fought on ground where Splashtop is also exposed, and the prospect who checks finds the same grievance waiting. Against TeamViewer that exposure does not exist.

The read: attack AnyDesk on delivery, support and conduct, attack TeamViewer on reliability and licensing, and open no pricing argument until the trial-to-paid path is fixed. That is a pricing decision that came out of a scraper.

There is a second operational read in the reply rates. Any play built on "nobody is listening to you" fails against TeamViewer, which answers 99% of complaints in about two days, and works against AnyDesk, which answers 24%. Same campaign, different mechanism per competitor.

## What does not work, and the version that does

The obvious next step is to identify these people and email them. Three of 207 hand-classified reviews named the reviewer's employer. That is 1.4%.

A mapped ICP list does not rescue it either, and I checked properly rather than assuming. 97 of those 207 carry a real full name, but only 5 are confirmed business users. In Germany: 14 reviews, 7 full names, zero confirmed business users. The constraint is not the absence of a list. It is that this review population is consumer and prosumer, which is the same fact the G2 gap exposed.

Invert the source and it works. G2 hides names and exposes job title, company size band, industry and country on every reviewer, and every G2 reviewer is a business user. Those four attributes matched against an account list you already own narrow it to a candidate set without identifying anybody, which is also the only version that survives a conversation about Article 6. I have not built that yet. It is the next thing I would build.

And at theme level you need none of it. AnyDesk has 1,385 lifetime Trustpilot reviews against the 200,000-plus customers it claims, seven per thousand. The people who complained publicly are the visible edge of a far larger group who had the same experience and wrote nothing. You are not looking for 31 leads. You are looking for the sentence that makes 3,000 people recognise themselves.

## Timing

Renewal complaints carry dates. Anchor on the review date, project twelve months, allow thirty days either side. In this set that produced an inferred renewal window for 73 rows. It applies to renewal and billing complaints only, never product complaints, and it is labelled inferred rather than known.

Weak per individual, useful in aggregate, and the one output that needs no identity at all, because a calendar does not need a name.

## What this generalises to

Any company with named public competitors can run this in a week, and it produces three things revenue teams usually guess at. Which differentiator to lead with, per competitor. Which axis to stay off because you are exposed on it too. And when each competitor's base is most likely to be in a renewal conversation.

One rule if you build it: never read a single platform. One source told me AnyDesk's customers were furious. Three sources told me which of its customers were furious, and that is the difference between a campaign and a misfire.

The [data page](https://claude.ai/code/artifact/99dd8ccb-ccd3-4782-91d4-feac343b898a) is open. Filter by vendor, theme, country or source, and click through to any individual review, including the ones I labelled by hand.

If you are building something like this, twenty minutes is usually enough to work out whether the trigger is worth the maintenance at your company, or whether you are about to build a dashboard nobody opens.

---

## Questions

### Does a low star rating mean a vendor's customers are unhappy?

Not on its own. AnyDesk scores 1.6 on Trustpilot, 4.5 on G2 and 4.6 on Capterra in the same period. The platforms sample different populations: G2 and Capterra verify business identity and lean on solicited reviews, Trustpilot is self-initiated and consumer-heavy. Read a rating as an output of a review operation and a reviewer population, then check which population you actually sell to.

### Can you identify which company left a negative review?

Almost never. Three of 207 hand-classified reviews named the employer, 1.4%, and G2 returns none by design. A mapped ICP list does not fix it: of 97 reviews carrying a full name, only 5 were confirmed business users. Work at theme level instead, which identifies nobody.

### What is the highest-value output of competitor review mining?

Message selection. In this dataset AnyDesk's detractor complaints are 85% commercial while TeamViewer's are led by product reliability at 41%, so the two require opposite cold emails. Getting that backwards is the most expensive mistake available.

### How do you match review data to an account list without identifying anyone?

Use the attributes review sites publish rather than the names they do not. G2 exposes job title, company size band, industry and country on every reviewer. Matched against a known account list those narrow it to a candidate set for prioritisation, with no individual identified and no reviewer contacted.

### How often should the tracker run?

As often as something downstream changes because of it. Daily if a live sequence's copy or targeting responds to the theme mix. Monthly if the output is a report. Ingestion is cheap, maintenance is not.

The builds behind these posts run in the Lab, with what worked and what I would change next.

[Open the Lab](https://aiento.eu/lab)

Keep reading

## [8 September 2026 GTM engineer, RevOps ou agência: o que uma empresa B2B portuguesa deve fazer em 2026](https://go.aiento.eu/blog/gtm-engineer-revops-agencia-portugal)

## [8 September 2026 GTM engineer, RevOps hire, or an agency: what a German B2B company should do in 2026](https://go.aiento.eu/blog/gtm-engineer-vs-revops-vs-agency-germany)

João Diogo · Berlin · aiento.eu [LinkedIn](https://www.linkedin.com/in/joaomigueldiogo/) [GitHub](https://github.com/joaomigueldiogo) [Privacy](https://aiento.eu/privacy) [Imprint](https://aiento.eu/imprint)

```json
{
  "@context" : "https://schema.org",
  "@type" : "BlogPosting",
  "author" : {
    "@type" : "Person",
    "name" : "João Diogo",
    "url" : "https://go.aiento.eu/blog/author/joao-diogo"
  },
  "dateModified" : "2026-10-07T09:33:49.624Z",
  "datePublished" : "2026-09-13T17:21:54.000Z",
  "headline" : "683 reviews told me which competitor to attack, and which one to leave alone",
  "mainEntityOfPage" : {
    "@id" : "https://go.aiento.eu/blog/review-mining-which-competitor-to-attack",
    "@type" : "WebPage"
  },
  "publisher" : {
    "@type" : "Organization",
    "logo" : {
      "@type" : "ImageObject"
    }
  }
}
```

```json
{
  "@context" : "https://schema.org",
  "@type" : "FAQPage",
  "mainEntity" : [ {
    "@type" : "Question",
    "acceptedAnswer" : {
      "@type" : "Answer",
      "text" : "Not on its own. AnyDesk scores 1.6 on Trustpilot, 4.5 on G2 and 4.6 on Capterra in the same period. The platforms sample different populations: G2 and Capterra verify business identity and lean on solicited reviews, Trustpilot is self-initiated and consumer-heavy. Read a rating as an output of a review operation and a reviewer population, then check which population you actually sell to."
    },
    "name" : "Does a low star rating mean a vendor's customers are unhappy?"
  }, {
    "@type" : "Question",
    "acceptedAnswer" : {
      "@type" : "Answer",
      "text" : "Almost never. Three of 207 hand-classified reviews named the employer, 1.4%, and G2 returns none by design. A mapped ICP list does not fix it: of 97 reviews carrying a full name, only 5 were confirmed business users. Work at theme level instead, which identifies nobody."
    },
    "name" : "Can you identify which company left a negative review?"
  }, {
    "@type" : "Question",
    "acceptedAnswer" : {
      "@type" : "Answer",
      "text" : "Message selection. In this dataset AnyDesk's detractor complaints are 85% commercial while TeamViewer's are led by product reliability at 41%, so the two require opposite cold emails. Getting that backwards is the most expensive mistake available."
    },
    "name" : "What is the highest-value output of competitor review mining?"
  }, {
    "@type" : "Question",
    "acceptedAnswer" : {
      "@type" : "Answer",
      "text" : "Use the attributes review sites publish rather than the names they do not. G2 exposes job title, company size band, industry and country on every reviewer. Matched against a known account list those narrow it to a candidate set for prioritisation, with no individual identified and no reviewer contacted."
    },
    "name" : "How do you match review data to an account list without identifying anyone?"
  }, {
    "@type" : "Question",
    "acceptedAnswer" : {
      "@type" : "Answer",
      "text" : "As often as something downstream changes because of it. Daily if a live sequence's copy or targeting responds to the theme mix. Monthly if the output is a report. Ingestion is cheap, maintenance is not."
    },
    "name" : "How often should the tracker run?"
  } ]
}
```