Methodology

A working document describing how the PolyBuzz AI chat page is researched, sourced and kept current. Every material claim on that page traces back to one of the source categories below.

Source hierarchy

Sources do not carry equal weight, so I rank them from heaviest to lightest. These are the five categories behind every published claim:

1. Primary platform materials. The App Store and Google Play listings, the official FAQ, the community guidelines and the privacy policy set the prices, tiers, age rating, sign-in options and data practices I report.
2. Account checks. A free account on the website and the iOS app confirms what those materials describe: the sign-in options, the chat list, the creator's required fields, the blur filter and Pure Mode. Paid-tier features come from the store listings and the FAQ, not from a paid subscription.
3. Regulators and public records. California's SB 243, the New York governor's notice on AI companion safeguards and the FTC's September 2025 inquiry frame what the law expects of companion apps. They are cited for what the rules require, not as proof of how any platform complies.
4. User reports. App Store and Google Play ratings, Trustpilot reviews and community discussion show recurring complaints such as billing, ads and filtering. They are always labelled as reports rather than findings.
5. Third-party write-ups. Breakdowns such as Fast.io's pricing overview and Bittime's walkthrough of the character builder fill in details the official materials leave out. Wherever one is the source of a figure, it is credited by name.

What gets tested versus what gets cited

Anything that changes quickly and can be seen on a free account gets checked in the live product before publication. That covers the sign-in options, the chat list, the creator's fields and the filter's behaviour. Anything I cannot observe myself, such as ratings, review counts, regulatory dates and paid-tier features, is cited to its source with the date it was checked, and it is framed as that source's statement rather than my own conclusion.

Cross-checks for pricing and features

Prices are checked against the App Store in-app purchase list on the day of publication. PolyBuzz sells Standard, Premium and Ultimate at different monthly prices, plus a weekly Standard option and coin packs, so I compare the store list with the official FAQ and at least one independent breakdown before a figure goes live. Prices on the main page carry the date they were checked, and a change spotted after publication counts as a correction under the editorial policy.

Comparison methodology

The alternatives table compares PolyBuzz with Character AI, Janitor AI, Chai and Candy AI on three axes: what the free plan covers, the cheapest monthly plan and how much adult content is allowed. Each platform is described in the terms it uses on its own site rather than from hands-on use here, and the author note on the main page says so. Prices carry the date they were checked, because this category reprices often enough that an undated table goes stale within months.

Freshness and revision cadence

The main page is re-checked at least every 90 days. Out-of-cycle updates are triggered by a price or tier change, a new store rating, a change to the community guidelines or privacy policy, or a new law or regulator action affecting companion apps. When a page changes, its dateModified in the schema moves forward, while datePublished stays fixed so the record of first publication survives.

Corrections process

Corrections are handled openly, and anyone can send one. Send a note via LinkedIn (linkedin.com/in/melissablake1) with the claim, the correction and a source. A verified correction triggers an edit to the affected page, an entry in the correction log on the editorial policy page and a dateModified bump, with a turnaround target of seven working days.

What does not get done here

Some practices are ruled out on purpose. Listing them here heads off the most common misreadings of what this site does:

• No sponsored coverage. No platform mentioned on this site has paid for placement.
• No AI-drafted final copy. Language models help summarise long documents such as privacy policies during research, but the published prose is written by me.
• No untraceable figures. A number that cannot be traced to a named source does not get published.
• No clinical claims. Anything about mental health or companion-app use comes from cited research, not from my own conclusions.

Conflict-of-interest scan

I hold no equity in PolyBuzz or the companies behind it, in any competitor named here, or in the affiliate networks behind outbound links. If that ever changes, the change is logged on the editorial policy page the same day and the affected coverage is re-checked.

Last reviewed: October 6, 2026 · by Melissa Blake