How comparison sites really work

Can you trust AI companion comparison sites?

Search for the best AI companion apps and you get a wall of ranked lists. In this category most of those rankings are ordered by what each app pays for a signup, not by how good it is. Here is how that machine works, how to spot it in ten seconds, and what to do instead.

The short answer

Mostly no. In this niche the ranked list is an advertising format wearing a review’s clothes: the publisher earns a cut of every subscription it sends, so the order tends to follow the payout rather than the product. A ranking is evidence that somebody is being paid. It is not evidence that anything is good.

Two things actually tell you whether an app fits you: using it yourself on a free tier, and asking people who already live in it.

How do AI companion comparison sites make money?

Affiliate commission. A reader clicks a link, subscribes, and the site that sent them takes a cut — usually for as long as that subscription lasts. This is not a rumour: the networks that run these programmes publish the rates on open pages. One large AI companion app is offered through a network at 40% lifetime revenue share, or up to $44 per sale, to anyone in the world who opens an account.

Now do the arithmetic from the publisher’s side. A list of ten apps where the top slot pays 40% forever and the eighth slot pays nothing is not a list of ten apps. It is one recommendation and nine pieces of set dressing. Nobody has to be lying for the order to come out wrong — the incentive does it on its own. The same machine drives AI girlfriend site reviews, "top 10 companion" roundups and best-chatbot lists: one affiliate layer, many wrappers.

There is nothing wrong with earning a commission. What the US Federal Trade Commission’s Endorsement Guides ask is that you say so: disclose the relationship "clearly and conspicuously on your site, so readers can decide how much weight to give your endorsement". The Guides draw the line at influence — if the money does not affect "what products you review, what you say about them, or the placement or prominence of those reviews", there is nothing to disclose. If it does, there is.

How can you spot an affiliate-ranked list?

Five checks. None of them takes longer than the page took to load.

  1. Look at where the link actually goes

    Hover the button, or long-press it on a phone, and read the URL. A plain link to the product’s own domain is a link. A URL carrying a tracking identifier is a paid referral. So is a link that goes to a redirector domain you have never heard of before bouncing you onwards.

    ?ref=a1b2c3 · ?aff_sub5=… · ?aff_unique5=… · ?clickid=… · t.some-tracker.com/370286/…

  2. Search the page for the word "commission"

    Use ctrl-F. A site that earns commission and says so somewhere is playing straight, and you can read its list knowing what it is. A ranked list with no occurrence of "affiliate", "commission" or "disclosure" anywhere on the page, whose every button is a tracked link, has made a decision about you.

  3. Ask what the score is made of

    A rating out of five with no method behind it is decoration. Look for the boring specifics: how many hours, on whose account, paid for by whom, with screenshots of the actual output. "We tested 25 platforms" with no evidence of a single session is a sentence, not a test.

  4. Read the weaknesses

    Every product has real ones. A review that awards 9.5 out of 10 and lists "occasionally slow to load" as the only downside was not written by somebody who used it. An honest write-up names a flaw that would actually stop some readers from buying — and then explains who it does not matter to.

  5. Check who publishes it

    Articles titled "what Reddit actually recommends" are often published by one of the apps in the list, which is, remarkably, number one. Scroll to the footer and work out whose site you are standing on before you weigh its verdict. A vendor writing about the category is fine. A vendor pretending to be a referee is not.

We ran these checks ourselves on 30 August 2026, on two "best AI companion apps, ranked" articles from the same publisher. On one, eight of ten outbound links carried affiliate tracking parameters; on the other, four of ten. The words "affiliate", "commission" and "disclosure" appeared nowhere in either page. We are naming a pattern, not a publisher — run the check yourself on whatever list you are reading and see what you get.

Does a number one ranking mean it is the best AI companion app?

It means it converts. Ranking and quality are read off different instruments: a ranking measures what the list owner earns, quality is whether the thing still holds up in your third week with it. They come apart hardest exactly where you need them not to. A companion platform is judged on whether it remembers, on how it handles a scene going somewhere you did not plan, on whether it is still interesting once the novelty is gone. None of that survives a bullet list, and none of it is what a commission rate is tracking.

What should you do instead?

  1. Use the free tier for twenty minutes

    Every serious platform in this category has one, and twenty minutes is enough to tell. Bring the same scenario to each app you are weighing so you are comparing like with like, and start with the thing you actually want out of it rather than the demo character on the front page.

  2. Ask people who actually use it

    Not a ranked list — a room full of people mid-conversation. Ask about the one thing you care about, watch who answers with specifics, and pay attention to what the regulars complain about. A community that lets its users complain in public is telling you more than any score out of ten.

  3. Read the threads, not the listicles

    Search something like "character ai alternative reddit" and read the discussion itself rather than an article summarising it. Weight the person who is three months in and still annoyed about one specific thing above the person posting a neat top ten — the first one has used it, and the second one might be selling it.

What to test in the first twenty minutes

  • ✓ Memory: tell it something specific, change the subject, come back to it four messages later.
  • ✓ Message limits: find out what happens on the free tier when you hit the cap, and what the cap becomes when you pay.
  • ✓ Content limits: find out where they are before you are invested, not during a scene.
  • ✓ Consistency: push the story somewhere your character should push back on, and see whether they stay themselves.
  • ✓ Editing: can you fix a bad reply, or change a character after you made them, or is the first draft permanent?
  • ✓ Everything that is not text: images, voice, video — included, metered, or a second purchase?
  • ✓ Exit: can you get your characters and your conversations out again if you decide to leave?

We are a vendor. Read this page knowing that.

Soulkyn is one of the products those lists rank, and this page is published by Soulkyn. We make money when people subscribe. That is a real bias and you should weigh it exactly as hard as you weigh anybody else’s.

The reason the incentive still differs is in what each side is asking you to do. A commission-ordered list needs the click, today, from this page — its whole revenue depends on you not going away to check. We are asking you to go away and check: open a free tier, run your own twenty minutes, then go and ask a room full of current users what annoys them about us. If we were optimising for the click, sending you to an unmoderated chat full of people who already pay us would be a strange way to do it.

And where we cannot show our work, we say so. Our comparison pages set out what Soulkyn ships, specifically enough that you can check each line; they do not score other people’s products, because we cannot audit somebody else’s platform from the outside. We deleted a twenty-six-service scoreboard off this site for exactly that reason — it had no methodology, and a comparison you cannot check is not a comparison.

Questions people ask about AI companion review sites

Are AI review sites trustworthy?

Some are. The ones that are will tell you how they tested, who paid for the accounts, and what each product is bad at — and they will say plainly that they earn a commission. Treat a ranked list with none of those three as an advertisement, because that is what it is.

Why do different comparison sites keep picking the same number one?

Because the same affiliate programmes pay the best rates to everybody, so independent publishers chasing the same revenue converge on the same order. Sites also copy each other. Agreement between two lists is only evidence if they reached it independently — and in this niche they usually did not.

What does an affiliate link look like?

A normal link goes to the product’s own domain and stops there. A referral link carries an identifier — ref, aff_sub, aff_unique, clickid, a long utm_campaign — or it goes through a tracking domain that redirects you onward. Hover it and read the status bar; on a phone, long-press and copy the link, then look at it.

Is a comparison site allowed to hide that it earns commission?

The US Federal Trade Commission’s Endorsement Guides say a paid relationship should be disclosed clearly and conspicuously, so that readers can decide how much weight to give the endorsement, and they treat payment that affects the placement or prominence of a review as exactly that kind of relationship. We are not lawyers and enforcement is not our business — but you can use it as a reader’s standard: no disclosure, less weight.

Where can I get honest opinions about AI companion apps?

Communities where people are mid-use rather than mid-sale: the platforms’ own Discord servers, subreddits and forums. Search things like "character ai alternative reddit" and read the thread rather than the article about the thread. Look for the long, specific, slightly annoyed posts — those are written by people who are still there.

You are a vendor telling me not to trust vendors. Why believe this page?

Do not. Believe the free tier and the people in the community instead. Everything on this page is a check you can run yourself in under a minute — hover a link, search the page for the word commission, look for a stated method. If the checks hold up, the page did its job whether or not you ever open Soulkyn.

Where the numbers on this page come from

Two published sources plus our own checks, all read on 30 August 2026. Both links open the original pages so you can read them in full rather than take our summary for it.

Go ask the people who already use it

The Discord is open and it is not curated into an advertisement. The fastest honest read on whether Soulkyn suits you is someone who has been in it a year — including whatever it is that still annoys them.

Ask users on DiscordTry it yourself insteadCompare

Keep exploring

Pages people read next.