✦ Editorial information / Independent editorial guide
Our methodology and evidence rules
What we check, how we distinguish documentation from testing, and how affiliate links affect our work.
Independent research ✧ Source-linked feature claims ✧ No fabricated hands-on tests
Companion Fieldbook is an independent editorial buying guide focused on AI companion products. Our job is to make confusing feature claims, trial conditions and privacy language easier to assess before a reader spends money or shares intimate information.
What our research currently consists of
This initial edition is based on public, primarily first-party documentation available as of 19 October 2026. We compare service positioning, memory architecture, billing rules, voice modalities, and the limits of what official claims can prove. We also supply repeatable tests readers can run themselves.
What we have not done
We have not purchased and tested every plan, audited provider code, independently verified memory percentages, measured voice-latency in a laboratory, or evaluated backend security. We do not label these document reviews as hands-on testing or assign an invented 9.8/10 “score”. We will clearly label any future first-hand tests with methodology and dates.
Our feature evidence levels
- Documented: supported by a current official help article or provider page, cited on the relevant review.
- Not established: evidence reviewed does not reliably prove the feature, limit or performance claim. This is not the same as saying the feature is absent.
- Reader-test recommended: a feature is advertised but subjective quality, recall or reliability needs real-world checking.
- Independently tested: reserved for future work with repeatable methods and recorded outcomes, never implied by our current content.
How our recommendations work
We weigh a buyer’s intended use, documented capabilities, complexity, privacy questions and payment terms. There is no fixed universal champion. The finder uses a visible editorial fit rule rather than pretending to measure engagement or memory accuracy. We include comparison links between competing services to help readers make their own choice.
Affiliate independence
Some outbound links may become affiliate links if we join and verify a program. Affiliate arrangements never change the feature matrix or create a higher score. Where a tracked link is configured, our pages disclose that we may earn a commission at no additional cost to the reader. An unconfigured link goes directly to the provider’s official site.
Corrections and dates
We display the source-review date. Changes to source pages do not automatically update our content: it is a snapshot of documented claims. Use the contact page to flag an outdated feature, factual mistake, or broken link. Substantive corrections should replace the claim itself, not just update the date.
Safety and scope
We write for adult readers, discuss AI products as software rather than human substitutes, and do not offer mental-health treatment advice. Never depend on a companion app for emergency assistance.
Want to verify something? See how we research or submit a correction.