Every "best AI companion apps" list you have read this year was paid for by someone. That is not a conspiracy — it is the business model of the genre — but it does mean the lists are optimised for a different question than the one you are asking. You want to know which product is worth your money. A list earns money when you click a link, subscribe, and do not cancel.
This guide is about the gap between those two things: how to read a ranking, what the tells are, and what an honest listing actually looks like. It is the one piece of methodology we can offer that a vendor blog structurally cannot.
The four ways a list makes money (and the fingerprint of each)
1. Affiliate links. The list is free because a click that converts pays the publisher. Fingerprint: outbound links carry tracking parameters, and the product that pays the highest commission drifts toward the top over time. This is not inherently dishonest — it is how most review sites work — but the ranking and the revenue are not independent, and a list that does not say so is hiding the only fact that lets you discount it.
2. Paid placement. A vendor buys a slot: a "featured" card, a top-of-list banner, a sponsored category. Fingerprint: the paid product has thin or generic copy, sits above products with stronger evidence, and the page either discloses it in small grey type or does not disclose it at all. Paid placement is legitimate when it is labelled and illegitimate when it is disguised as ranking. The distinction is entirely about the label — which is why our own disclosure rules are written down rather than assumed.
3. Owning a product in the category. The reviewer competes with the reviewed. Fingerprint: the publisher's own product is ranked first or second, appears in every "top" list, and gets a paragraph of first-person praise the others do not. This is the most common pattern in this niche and the least often disclosed. If the list is published by a company that sells an AI companion, everything above the disclosure line is advertising.
4. Traffic arbitrage. The page does not recommend anything; it exists to capture a search term and monetise the visit with ads. Fingerprint: prices that are months out of date, no verification dates, no disclosure, and a list that reshuffles to match whatever is trending.
None of these are crimes. The problem is that all four push in the same direction — toward whichever product converts best — and your question is not "which converts best".
Five checks that take about four minutes
- Is there a disclosure, in the page, before the recommendations? Not in the footer, not on an "about" page. A one-line disclosure at the top is the single strongest signal that a list is trying.
- Do the outbound links have tracking parameters? Hover a link. If it carries a network tag, the site is monetised by your click, and you should read the ranking as inventory order.
- Is any product ranked first by the same company that publishes the list? If yes, discount everything above and below it.
- Do the prices have dates? A price without a verification date is a rumour. This category changes prices quietly and regionally — Replika does not publish prices on its website at all; they appear in-app and vary by country. A list quoting one global number is quoting something it could not have verified.
- Does the list ever say "we could not confirm this"? A review site that has never been uncertain is not reviewing, it is paraphrasing marketing pages.
What we do here, stated plainly
Since the whole argument is about disclosure, ours belongs in the middle of the article rather than in a footer:
- Paid slots are labelled. The directory sells placement — a category spot, a homepage card, a hero banner — and every paid entry carries a
Sponsored · Partnerlabel. Paid placement does not buy rank in editorial lists, and editorial content never auto-links to a paid slot. - Some links are affiliate links. Where a product pays a commission, we say so on the entry. A commission does not change a price, a rating or a place in a comparison.
- We also build a product in this category. SoulChat is ours. It appears in the directory on the same terms as every other entry, it is excluded from editorial recommendations, and it is not ranked first in any list we publish. Every article that compares it carries a disclosure at the bottom.
- Prices carry dates. 217 of the 242 entries have been verified, the last full pass on 2026-11-02, and each entry shows when its numbers were checked. When a vendor does not publish a price publicly, we record the figure we could confirm and say that is what it is.
- Adult platforms are listed, not reviewed. They get verified commercial facts and a separate filter; editorial content stays inside the non-adult set. The verification standard is identical for both — the difference is what we will write about, not how carefully we check it.
That is not a claim to be the only honest site in this niche. It is a claim that these five things are checkable, and that you should check them on every list you read, including this one.
Why this matters more than it sounds
The cost of a bad recommendation here is not a wasted $10. It is the switching cost. Companion products are built on accumulated history — weeks of conversation that a competitor cannot import. Once you have spent a month teaching a product your context, the annual plan at $8.33/mo becomes very hard to walk away from, which is exactly why the annual plan is priced that way.
That asymmetry is the thing a paid ranking exploits: the decision happens once, early, on the strength of a list, and then it is very expensive to revisit. Which means the four minutes you spend checking the list are worth more than the four hours you spend comparing features afterwards.
Disclosure: the team that runs this directory also builds SoulChat, an AI companion product. It is listed here on the same terms as every other entry, is not a paid placement, and is not recommended in this article.
How this was checked. The practices described in the "what we do here" section reflect the directory's current implementation: paid placement is labelled from the spot_tier field, affiliate relationships are recorded per entry, verification dates come from each record's last-verified timestamp (last full pass 2026-11-02, 217 of 242 entries verified).