How to Read an AI Companion App Review
Most published reviews in this category were written without sustained use of the product, by writers paid per published page or per install generated, from a template that gets the app name substituted into it. That is not a conspiracy; it is the predictable output of the economics. Reading them usefully means knowing what produced them, recognising the specific tells, and knowing what a review would have to contain to be worth anything — a list that turns out to be short.
This site does not publish reviews of these apps, for reasons that become obvious once the requirements are spelled out. What it can offer is the reading method.
Who writes them and what they are paid for
Four kinds of page dominate the results, and the incentives differ.
Affiliate roundups. The publisher earns when you install or subscribe. The commercial consequence is not usually a false claim; it is an ordering. Every app on the list is worth trying, the differences between them are described as matters of taste, and the app with the best commission sits at the top. Nothing is asserted that could be shown false.
Content-farm pages built for the search term. Written quickly against a keyword, often by someone with no interest in the category, from other pages about the same product. These are the source of the templated structure you will start noticing once you look for it.
Operator-adjacent material — press coverage seeded by the company, guest posts, and pieces built from a press kit. Frequently accurate about features and structurally unable to be critical.
Genuine independent writing, which exists, is a minority, and is usually recognisable because it is specific about something in a way the others are not.
Distinguishing the fourth from the first three is what the rest of this post is for.
The tells of a review written without use
Nine, and any two together are close to conclusive.
Feature lists that match the marketing page in order and wording. The giveaway is order. Someone who used the app describes what they noticed; someone who read the marketing describes what the marketing led with.
Universal praise for memory. Continuity is the most oversold property in the category and the hardest to test, since testing it requires a long gap. A reviewer enthusing about how well an app remembers, without describing a specific thing it remembered across weeks, has not tested it. Why the claim is structurally overstated is in how AI companion apps work.
Prices and tiers stated with confidence and no date. These change frequently. A page that states them without a date is not maintained, whatever its byline says.
No account of anything going wrong. Sustained use of any product in this category produces something to complain about — a refusal that broke the mood, a session where the character drifted, an outage, a change after an update. A review with no friction in it is not a report of use.
Numeric scores across dimensions that cannot be measured. “Conversation quality: 8.5/10” is a number attached to a subjective impression to borrow the authority of measurement.
Screenshots that appear in other reviews too, or that are visibly from the store listing.
Superlatives without a comparison set. “The most advanced companion AI” from a writer who does not say what they compared it against.
Silence on the exit. No mention of cancellation, export, or account deletion. These are the things that matter most to someone who actually committed to a product, and they are absent from nearly every review because the writer never committed.
A structure that repeats across the site’s other reviews, heading for heading. Open two of the publisher’s reviews side by side; this takes thirty seconds and it settles the question.
What an honest review would have to contain
Five things. Notice how few of them appear in what you have read.
A stated duration and depth of use, and enough incidental detail to make it credible. “Three weeks, most days, about twenty minutes” plus something that could only have come from that.
One long-gap continuity test, described concretely. A specific unremarkable detail mentioned once, asked about weeks later, with the actual result reported — including a disappointing one.
What the reviewer paid, to whom, and what happened when they cancelled. The billing channel and the exit process, from experience. Nothing else in a review is as useful, and this is the item that requires genuinely committing. Why the channel matters is in who you actually bought the subscription from.
Something the product does badly, in specific terms.
A dated snapshot and an acknowledgement that it is one. Because models, instructions and filters are revised server-side without notice — see what an app update can change without asking — a review of one of these products is a description of a fortnight, and an honest one says so.
Reading store reviews instead
Often more useful than published articles, provided you read them correctly.
Ignore the average and ignore the five-star reviews. Read the low-rated ones, and read them for events rather than sentiment: billing that continued after cancellation, features withdrawn, behaviour changing abruptly, support silence. Repeated dated specifics from unconnected people are the strongest evidence available to anyone outside the company.
Watch the shape. A cluster of short generic five-star reviews arriving together, alongside a thin tail of detailed complaints, is a recognisable pattern — one of several store-listing signals set out in spotting a lookalike companion app.
Note that store reviews survive a rename. Older ones may describe a differently-named product, so it is worth establishing whether an app has traded under another title before weighing anything written about it.
Why this site publishes no reviews of these apps
Because it cannot meet the five requirements above. It has not used these products for weeks at a time, has not subscribed and cancelled, and has not run long-gap continuity tests. Writing a review anyway would mean assembling one from the pages described in this post, which is precisely the mechanism that makes the category’s coverage untrustworthy — and adding another instance of it would be worse than adding nothing.
What is publishable honestly is method: how to translate marketing claims into tests, what to check before installing and before paying, and where the authoritative answers live. Those pages stay true as the products change, which is more than can be said for any review of a product revised weekly from a server you cannot see.
The uncomfortable conclusion is that the review you actually need does not exist for most of these apps, and the substitute is twenty minutes of your own use plus a policy document. That is less satisfying than a ranked list and it is considerably more reliable than one.