Review methodology
The short version: primary sources first, independent review outlets second, community threads for context only. Pricing and features get cross-checked against multiple sources before landing on the page. When a claim can't be verified, it's flagged rather than filled in from imagination.
Source hierarchy
Sources on this site rank in a specific order. The Joi AI pricing page at joi.ai, the T&C, the App Store listing for the iOS app and the platform's own product pages come first. These are treated as canonical for anything Joi AI itself controls: prices, feature availability, model naming (Mars, Moon, Saturn 5.2, Mars 2.2), Neuron costs and monthly allowances. If joi.ai and a review disagree, the platform's own page wins for what the platform charges and offers today.
Independent review outlets that published dated, dollar-figure coverage in 2026 come second. Names cited in the main coverage include aigirlfriendbase, aicompanionguides, ohgirlfriend and scribehow. These outlets went far enough to document Neuron pack prices, the HELLO50 code behaviour and specific feature costs. They get cited by name when their reporting materially shapes a claim.
Trustpilot reviews and public Reddit threads sit third. They're used to characterise the shape of user complaints and praise, not to establish specific numbers. A Trustpilot cluster mentioning "slow refunds" is fair to cite; a Trustpilot user claiming "refund took exactly 8 days" is not treated as generalisable data.
What gets verified vs what gets cited
Anything readers can check for themselves gets verified against the primary source. Prices, promo code existence, iOS listing publisher name, availability across regions, character library size range. These are one-click verifications; there's no excuse for getting them wrong.
Anything that requires live account access without direct testing gets cited from named independent reviews with a clear attribution and date. Video call latency at peak hours, refund response times in weeks vs days, image quality drift across specific prompt patterns: these require testing beyond desk research, and the site currently reports them through named third-party reviewers rather than fabricating measurements.
Anything neither verifiable nor third-party documented gets flagged as unavailable on the page rather than filled in. If a specific figure hasn't been published anywhere reliable, that's the fact worth reporting.
Cross-checks for pricing and features
Pricing lands on the page only after two independent confirmations. Joi AI's $17.77 monthly, $26.66 quarterly and $60.95 annual figures were checked against multiple mid-2026 reviews that documented them. The Neuron pack prices ($12.69 for 500, $38.09 for 2,000, $126.99 for 10,000) came from a single primary review with specific dollar figures; those are labelled as of July 2026 because pack prices change more often than subscription prices.
The HELLO50 code is handled carefully. Sources consistently report that HELLO50 brings the annual plan to around $24 total, but the math doesn't cleanly come out to a flat 50% off $60.95. The page reports what reviewers found without asserting the discount mechanic ourselves. If HELLO50 behaviour changes, the coverage updates.
Comparison methodology
The alternatives section in the Joi AI overview compares Joi AI with CrushOn AI, Ourdream, Lovescape, Secrets AI, Kindroid and Replika Pro. Prices for each are the current sticker on each platform's own pricing page as of mid-July 2026. Trade-offs are drawn from the same source hierarchy: primary product pages first, named third-party reviews second, community threads for context.
The comparison is deliberately narrow. It compares platforms Joi AI users consider switching to, based on comparison-thread patterns on Reddit and Discord. Broader "top AI companion" tables that rank thirty platforms without regard for which platforms Joi AI users would consider add no reader value and are avoided.
Freshness and revision cadence
The main page is reviewed monthly at minimum, and updated whenever a pricing change, feature launch or model update happens. The lastReviewed date in the schema and the visible "Last reviewed" line under the author byline reflect this. Coverage of the live video call feature specifically is flagged for more frequent review because it's newer and changing faster.
The homepage currently carries the October 6, 2026 review date. Older claims retained from prior versions get re-verified at each review pass; anything that no longer holds gets rewritten or removed rather than left in place.
Corrections process
Factual challenges go to [email protected] with the specific claim and a source. Verified corrections get applied within seven days and logged with a dated note on the editorial policy page. Corrections that materially change the shape of a claim (a price that was reported wrong, a feature described inaccurately) trigger a re-review of adjacent claims on the same page. Reader-facing conditions of using this coverage sit on the terms of service, and how correspondence with reader challenges is retained is documented in the privacy policy.
What does not get done here
No live account testing with fabricated measurements. If a claim depends on data from an active Joi AI subscription that hasn't been run through direct testing, it's cited from a named third party or flagged as unavailable. Manufactured "our team tested this for six months" language does not appear on this site because the honest attribution is more valuable than the marketing-style claim.
No star ratings, aggregate scores or "our verdict" tables. Trustpilot's 4.0 out of 5 is theirs; we cite it, we don't reinvent a competing number. Readers make their own call after reading the coverage; nothing gets summarised down to a single score they didn't ask for.
No paid placement or sponsored content dressed up as editorial. If Joi AI or any competitor pays this site directly for placement, that placement carries a clear label. See the editorial policy under sponsored content policy for the specifics. About context and who edits sits on who runs the site. Author profile is on the senior editor page. Adult framing of the coverage is documented on the the 18+ gate and enforcement, and federal record-keeping compliance for editorial-only coverage sits on the 18 U.S.C. § 2257 statement.