How we review AI companion apps

This page says exactly how a review on AI Companion Desk is put together, so you can decide how much weight to give it. The short version: we research from published sources, we score on five fixed criteria, and affiliate commissions never move a score.

What a review is based on

• The product's own pricing page. Every tier, credit system and renewal price, checked on the date shown in the review. If pricing changes, the review is updated and the date moves.
• Official documentation and changelogs. What the app says it can do: memory, image and voice generation, character tools, content policy.
• Public user feedback. Recurring themes across public reviews and communities, not single complaints. We look for what people cancel over, what breaks after the first month, and what surprised them positively.
• Public product pages. Where a review shows screenshots, they are captures of the app's public landing, pricing or feature pages, labelled with the page and the capture date. They are not screenshots of a private session.

What we do not claim

We do not describe hands-on testing we have not done. You will not find invented "tested for 30 days" claims, made-up message counts or fictional spending figures here. Where a review is research-based, it says so in its How we evaluate section. If that changes for a specific app, the review will state what was done and when.

The five criteria

Every app is scored on the same five things, so scores are comparable across reviews:

1. Conversation features. Model quality as documented, roleplay depth, voice, image and video generation, and how restrictive the content filter is according to the product's own policy pages.
2. Memory. Whether the app remembers context across sessions, how long, and whether memory is a paid tier. Taken from documentation, changelogs and recurring user reports.
3. Customization. Character builder depth: appearance, personality, scenario and backstory controls, and whether custom characters are gated behind a plan.
4. Value for money. The full price structure: free tier limits, subscription tiers, credit systems, renewal price and what quietly runs out first. Checked against the live pricing page and dated.
5. Privacy controls. Account requirements, data retention statements, payment descriptor, and deletion or export options as published by the provider.

The overall rating is out of 5. A high score means the app does what it says at a fair price with clear policies; a low score usually means an aggressive credit system, a misleading free tier, or a privacy posture we would not accept ourselves.

How rankings are ordered

Best-of lists are ordered by the criteria score, then by fit for the use case in the title. They are not ordered by commission. When the strongest app in a category has no affiliate program, it still ranks first.

Affiliate links

Some links are affiliate links, marked with rel="sponsored". If you subscribe through one we may earn a commission at no extra cost to you. Commissions fund the site; they do not change a score, a ranking position or a verdict. See About for the full disclosure.

Updates and corrections

Reviews carry a published date and, when changed, an updated date. We re-check pricing and policies when an app ships a major change. If you find an error, the fastest way to get it fixed is the contact details on the About page.

Who writes this

Reviews are published under the byline Adrian Wells, the pen name of the site's lead editor. He tracks pricing, features and policy changes across 40+ AI companion platforms.