Methodology
Every fact on this site should be traceable to a source you can open yourself. This page sets out where facts come from, how they are dated and refreshed, how apps are ranked, and how mistakes get fixed.
How the research is done
Research is compiled with AI assistance and every price, policy and date is checked against the primary source linked on the page. That includes the parts you see in search results: page titles, descriptions, structured data and image descriptions. Pages are published under the byline CompanionHunt Research. Reviews are built from what each app publishes (its pricing, terms, content policy, privacy policy and help pages) and from its store listings. For some apps we also open the app ourselves and record what it shows (below). Each review says at the top which of the two it rests on.
In-app checks
An in-app check runs the same fixed routine on every app, so the results compare. We install the app from the link on its own website, sign up for a free account on a US connection, and record, with dated screenshots: how sign-up and the age check work, what the free tier gives, every plan and price on the paywall, the content, memory and data-deletion settings, and every upgrade prompt that appears during a set chat.
The chat is a fixed script. Early on we give the companion four facts (a name, a home town, a dog's name and a birthday), make fifteen lines of small talk, then ask what it remembers, and record how many of the four it gets right. Nothing is bought, and only the free tier is checked. Results are written up only after the check has been run, with the date, the app version and the screenshots on the page.
Which sources count
Prices, in this order of trust:
- The app's own in-app paywall, opened on a US account and captured as a dated screenshot (a store's in-app purchase list shows every product ever created, so it is never a price source)
- The app's official web pricing page
- The app's official help or documentation pages
Other review sites are never used as a price source. Prices are quoted in US dollars for the US storefront. When the App Store price and the web price differ, we show both and say which is which.
Store ratings come from the US App Store first, then Google Play, then Trustpilot, always shown with the store, the number of ratings and the date we read them. They are the stores' numbers, not ours.
Policies are quoted, not paraphrased from memory. A statement that an app allows adult content, trains on chats or lets you delete your data is backed by a short quote from the app's own terms, content policy, privacy policy or help pages, with a link.
Incidents, lawsuits and laws need a primary source: a court filing, a regulator's release, the bill text or the company's own statement. News coverage is used as support, not as the only source.
What users say is only ever reported as a pattern across many public reviews or posts. We do not quote individual users or invent any.
Dates on every fact
Every price and rating on the site carries the date we checked it. All dates on the site are US Eastern calendar dates (New York time), whoever did the checking and wherever they were. A page's "updated" date only moves when a fact on it changes, and the change itself is recorded in the app's timeline and in the change tracker.
How often facts are refreshed
Prices are re-checked in a sweep at the start of every month. A price older than 60 days is either re-checked or taken off the page. Anything we cannot see without paying is left out rather than guessed.
How apps are ranked
Rankings use each app's US App Store rating, with its source named on the app's page. General-purpose chatbots (ChatGPT, Claude, Gemini, Grok, DeepSeek, Pi) are listed unranked, because no store rating measures how good they are as companions. Apps that have shut down are listed as retired and unranked. No position on this site can be bought, and affiliate payouts play no part in the order.
The CompanionHunt score
An app's review also shows the CompanionHunt score: a 10-point score calculated only from published facts. It is ours, it is based on published facts rather than testing, and anyone can recompute it from the sources on the review. A price change can move it, and when it does the change is recorded.
A company gets no credit for what it does not document. If its own pages do not mention a memory feature, a free allowance, a privacy promise or its rules on adult content, that part scores zero, and the review links the page we checked. A price we could not verify is left out rather than guessed: the other parts are scaled up to 10 and the score is marked provisional, naming the part that is missing. At least 7 of the 10 points must come from sourced parts, or the app gets no score at all. General-purpose chatbots are marked "Not rated: general-purpose AI", and apps that have shut down are marked "Shut down". For now the rankings still follow store ratings, as described above; the score sits alongside them on each review.
| Part | Points | How it is scored |
|---|---|---|
| Value | 2 | Cheapest paid plan as a monthly price, on any platform. $8 or less scores full marks, $30 or more scores zero, straight line between. An app that is entirely free scores full marks. If we could not verify a price, this part is left out, the others are scaled up to 10 and the score is marked provisional. |
| Free tier | 1.5 | Everything free: full marks. Unlimited free use on a lighter model or with major features removed: 0.75. A permanent free tier with a hard cap: 0.5. A time-limited trial only: 0.25. Pay to use, or no free allowance stated: zero. |
| Platforms | 1 | A third of the marks each for a web app, an iOS app and an Android app. |
| Content freedom | 1 | From the app's own content policy. Adult content allowed for adults: full marks. Allowed only behind an opt-in setting: 0.75. Not allowed: 0.25. Not addressed in its policy: zero. |
| Memory | 1.5 | From the app's own documentation, not measured recall. Long-term memory documented as a feature: full marks. Only short or limited memory documented: 0.5. None documented: zero. |
| Voice and images | 1 | Half the marks for documented voice calls, half for documented image generation. |
| Privacy terms | 1 | From the privacy policy. 0.4 if it says personal data is not sold, 0.3 if chats are not used for training (0.15 if you can opt out), 0.3 if it documents how to delete your data. |
| User ratings | 1 | The US App Store rating, else Google Play, else Trustpilot: the first with at least 25 ratings, dated. 3.0 or lower scores zero, 5.0 full marks, straight line between. If no store has 25 ratings, this part is left out and the other seven are scaled up to 10. |
The change tracker
Companion apps change constantly: prices, models, content policies, free-tier limits and owners. Each recorded change is a dated entry with a type (pricing, model, policy, feature, legal, shutdown or company) and a source link. Entries that have not yet been checked against their sources carry a "being verified" label until they are.
How we count app-store complaints
For the "What 1-2 star reviewers complain about" sections we collect every one- and two-star review we can reach for an app from the US App Store, Google Play (US, English) and Trustpilot, going back up to twelve months. Stores only show their newest reviews, so a busy app's sample can cover a few weeks while a quiet one covers the year; each section states its own sources, counts and dates. We read every review and give it one main complaint from a fixed list of 16 (for example ads, paywall and credits, filter, bugs, age verification, refunds). A second, blind read of 50 random reviews agreed on 88% of them. Google Play's one- and two-star reviews are weighted by how often each arrives, so a pull of equal numbers does not over-count two-star reviews. We never quote a review or name a reviewer, and an app with fewer than 30 low reviews gets no breakdown. Low reviews show what unhappy users complain about, not how many users are unhappy.
Adult content
Apps whose own policies allow adult content carry an 18+ label. We describe policies in plain, non-explicit language and never publish explicit images.
Affiliate links
Some links to apps are affiliate links, and we may earn a commission if you sign up through one. It never changes a ranking or a fact. Details and the list of programmes are on the affiliate disclosure page.
Corrections
If something is wrong, tell us. We check it against the source, fix the page, and record the correction with its date in the app's timeline.