How Verified Reviews Build Trust on a Companion Directory

Reviews on a companion directory are not the same as reviews on a restaurant booking site. The stakes are different, the incentives to fake them are higher, and the consequences of a wrong rating affect more than just dinner. A good review system on a directory like SilkDots has to do work that a restaurant site doesn't: gate who can review, weight reviews against trust signals, and detect when ratings stop reflecting reality.

This article explains how the SilkDots review system is designed and what each layer is trying to prevent.

The first gate: only verified clients can review

Open review systems get gamed within days. The first design choice on SilkDots is that reviews are only accepted from accounts that have evidence of an actual interaction with the listing being reviewed. That evidence is one of:

• A message thread with the provider
• A logged contact-button click on the provider's profile
• An explicit admin grant (rare, used for cases like a long-time client whose account predates the messaging system)

If none of those exist, the "leave a review" button is disabled. The label tells the user why.

The second gate: time

A review can't be left in the first hour after a contact event. This sounds small but stops a particular failure mode: someone clicking contact and immediately leaving a five-star review with no actual interaction.

The cooling-off window is short enough that genuine same-day reviews still flow through, but long enough to filter the most obvious gaming pattern.

Weighted scoring

A raw average of star ratings is misleading because not all reviewers contribute equally to truth. SilkDots' rating uses a weighted average where each review's weight depends on the reviewer's trust score at the time of the review.

Trust score is derived from:

• Account age
• Email verification status
• Login history (a brand-new account that left exactly one review has a low trust score)
• Reports against the reviewer (a reviewer who has had reviews repeatedly removed for being false has a very low trust score)
• A handful of behavioural signals about how the reviewer engages with the platform

A new account's first review counts, but it counts less than a long-standing account's review. Over time, as the account builds trust, its weight rises. This is invisible to the reviewer — there is no "your weight is X" display — but it shapes the public rating substantially.

Two numbers are visible on a profile: the raw average and the weighted average. The weighted average is what is used for ranking and surfacing.

Low-star reasons: structured signals

When a reviewer leaves a 1- or 2-star rating, they are required to pick from a short list of reasons:

• No-show
• Misleading photos
• Misleading description
• Rude or harassing
• Hygiene or safety
• Payment issue
• Other

These structured reasons do two things. First, they make the moderation team's job tractable — a 50-listing review of "Other" complaints isn't much of a signal, but a 50-listing review of "Misleading photos" is. Second, they let the platform act on patterns. A profile with five separate "Misleading photos" complaints in three months gets attention from moderation regardless of the overall star average.

Fake-review detection

Even with verification gating, fake reviews still get attempted. The detection layer looks for:

• Ring patterns: a cluster of accounts all leaving five-star reviews on the same listing within a short window
• Voice consistency: language and structure that repeats across reviews from supposedly different accounts
• Trust score collapse: an account that has been issuing rapid five-star reviews and is now flagged for other reasons

Detected reviews are flagged for moderator attention rather than auto-removed. Genuine review patterns can sometimes look statistical, so a human looks at the flag before action.

Provider responses

Providers can respond to reviews. The response appears under the review and is itself moderated — a provider can't use the response field to abuse a reviewer. The response option is intentionally limited because giving providers a louder voice than reviewers would defeat the system.

Removing a review

Reviews are removed only for clear violations:

• Personally identifying details that go beyond what is needed
• Threats or abusive language
• Reviews that turn out to be from accounts later proven fraudulent
• Reviews factually contradicted by platform evidence (for instance, a "no-show" claim against a session the platform has logged)

A review is not removed because the provider doesn't like it. A 2-star "misleading photos" review with no abuse and clear factual claims stays up.

What this looks like in practice

A provider with 30 reviews, mostly from accounts with long histories, will have a stable weighted average that closely tracks the raw average. A provider with 30 reviews where 25 are from fresh accounts that have done little else on the platform will have a weighted average noticeably below the raw average. The public sees both numbers; ranking uses the weighted one.

For clients, the practical takeaway: a profile with a moderate number of reviews from trusted accounts is more reliable than a profile with a flood of fresh five-star ratings. The displayed numbers tell you which is which if you read them.

Frequently asked questions

Can I leave a review without giving my real name? Reviews are tied to your account, not your real name. The reviewer name shown publicly is whatever your profile name is.

Can I edit a review after posting? Yes, within a 24-hour window. After that, edits require moderator approval.

Why does my star rating not match the displayed number? The displayed number is the weighted average, not your single review. As more reviews come in, your contribution becomes proportionally smaller.

Do providers see who left a review? The reviewer's display name is visible. Email addresses and other private details are not.

The compact version: reviews on SilkDots are gated by verified contact, weighted by reviewer trust, structured by low-star reasons, and moderated for fakes. Each layer is small on its own; together they make a rating system that mostly tracks reality.