How We Test AI Companion Apps

Where we test from

A residential Australian IP, country Victoria. No VPN, no proxy, no datacentre IP. The same connection any other Aussie reader would be using when they visit one of these platforms.

This matters more than it sounds. The Online Safety Act 2026 implementation is region-specific, and several platforms have rolled out AU-targeted behaviour we would not see from a US or UK IP. Promptchan shows a "your region requires age verification" message on AU IPs that other regions do not see. OurDream.ai routes AU users into a third-party AVS (go.cam) for NSFW unlock. DreamBF.ai has been observed serving AUD pricing on some routings. Testing from a real AU IP is the only way to surface this kind of region-gated behaviour.

For most reviews this is supplemented by a Playwright automation run from a country Victoria IP, en-AU locale, with a screenshot captured of the landing page. The audit is documented on the 22-app AU access audit hub.

What we test for on every review

Four data points minimum, plus everything that helps an Aussie reader make a real decision. The four minimums:

1. AU access status. Does the marketing site load from an Aussie IP? Fully open / button gate / hard verification / blocked.
2. Age verification flow. What the actual gate looks like. No gate / button click confirm / ID upload / biometric / third-party AVS. With screenshots wherever possible.
3. AUD pricing. Real prices in Australian dollars. Most platforms charge in USD with bank conversion; the review documents both. When a platform escalates intro pricing on auto-renewal, we publish the intro price and the recurring price separately.
4. OSA compliance posture. Where the platform sits against the Online Safety Act 2026 codes, written plainly. Self-declaration is non-compliant. Hard verification (gov ID, biometric AVS) meets the spirit of the Act. We say which.

Beyond those four, every review covers: product features, character library breadth, conversation quality, the signup flow with screenshots, the pricing page with screenshots, trust signals (operator transparency, Trustpilot, Reddit footprint, scam-scanner ratings where they exist), and honest negatives. Always honest negatives — a review without disclosed cons is doing its readers a disservice.

The eight steps every review goes through

Before a single sentence of the review is written, the following research is done. This process is documented in our internal review-build memory and applied identically to every platform:

1. Source-of-truth research. Fetch the platform's own site: pricing, FAQ, features, ToS, Privacy Policy. Verbatim quotes are pulled for the ToS and Privacy items that matter to AU consumers (refund window, auto-renewal terms, AI-training data clauses, content licences).
2. Reddit and YouTube sentiment. Search Reddit and YouTube for genuine user discussion. Verbatim quotes are captured with subreddit name and date. Astroturfed-pattern threads (multiple score-1 comments in marketing voice) are flagged but not quoted. If the platform has zero genuine community footprint, that is itself the finding and we say so.
3. Trust scanner cross-reference. Gridinsoft, Scamadviser, scam-detector.com when accessible. WHOIS lookups for domain age. Trustpilot listing verification (we have flagged multiple cases where the Trustpilot result Google returns is actually for a sibling product, not the platform reviewed).
4. SERP analysis. What is ranking for the platform's review queries, what those ranking pages are saying, and what content gaps they leave open. This is how we identify what to write that nobody else has.
5. AU IP testing. Visit the marketing homepage from a country Vic IP. Capture screenshots of the landing page, the signup flow, the pricing page, and any age-verification surface. Record the actual page behaviour, not the marketing claim.
6. First-person draft. Written by Matt, in plain Australian English. No em dashes. No AI patterns. Specific details over generic marketing copy. The draft includes honest negatives by default.
7. PAA + structured data. The "People Also Ask" questions surfaced in the SERP analysis become the visible FAQ at the bottom of every review. The same content is mirrored into FAQPage schema for AI search and rich result eligibility. Review schema, BreadcrumbList schema, and per-image alt text are added at the same time.
8. Affiliate link verification. The CrakRevenue (or direct) tracking link is confirmed live before the review goes up. Dead or paused affiliate links are swapped to either the new offer ID or the platform's direct URL with no tracking, and that change is disclosed in the review's affiliate disclosure paragraph.

If any step cannot be completed (a Privacy Policy that does not load server-side, a Trustpilot listing that does not exist, a Reddit thread that returns a verification wall), we say so in the review rather than fabricating around the gap.

How we rate

A 5-star scale weighted toward AU access friction, product fit, pricing transparency, trust signals, and operator credibility. The rating is set by the reviewer based on the actual testing, not pulled from any affiliate dashboard.

Some specifics worth knowing about the rating: a recently-registered domain (under six months) costs the platform on trust signals. An operator hiding behind Cloudflare with no named entity costs more. A ToS clause permitting user chat data to be used for AI training costs more again. A dark-pattern pricing pattern (exit-intent lifetime deals, $1 hidden fees, intro-to-recurring 4x escalations) does not automatically tank the rating but the platform has to make up the ground on product and operator clarity to land above 3.5. A genuinely free tier, a named registered operator, an actual organic community presence, and a published refund policy all push the rating up.

The rating never moves because of affiliate payout. The highest-payout offers on the CrakRevenue catalogue rank below platforms paying us less when the testing supports it. Get-Harder pays $50 PPS and rates 3.0. JOI pays $35 PPS and rates 4.5. That is by design.

How often we re-verify

Every review has a "Last verified" date in the meta line under the title. The date reflects when the AU access status was last manually confirmed.

• The 22-app audit is the broad baseline. Run via Playwright from a country Vic IP, captures the landing page, age gate behaviour, and any AU-specific routing. The most recent full audit was 28 April 2026. Tracked on the audit hub.
• Individual reviews are re-verified when one of three things triggers it:
  1. A platform rolls out a new gate or pricing change. Promptchan and OurDream.ai were both re-tested on 26 May 2026 after rolling out hard age verification.
  2. A CrakRevenue offer status changes. JOI's old offer 10163 was paused 19 May 2026; the JOI review and 18 other site pages were updated within 6 days.
  3. A reader flags a discrepancy. The DreamBF correction below is one example of how that flows through.
• The 90-day check. Any review older than 90 days gets a baseline re-check even without a trigger event. AU access can shift quietly under enforcement pressure.

What happens when we get something wrong

Corrections happen in-place with the change disclosed in the review body, not deleted and rewritten silently. The most recent example, published the same day this methodology page went live:

The principles for corrections: visible, in-line, dated, and with the original error stated plainly so a returning reader can see what changed and why. We do not silently swap copy on a page and pretend the prior version did not exist.

Affiliate disclosure

This site earns commission when readers sign up through our tracked affiliate links. The platforms reviewed pay between $30 and $55 USD per signup on CrakRevenue's pay-per-sale offers, or 28% to 50% lifetime revenue share on the revshare offers. Some links are direct (no tracking, no commission earned) where the affiliate offer has been paused or where we have not yet wired a tracked link — those are flagged in the per-review disclosure.

Every affiliate link uses rel="sponsored" target="_blank" per ACCC and FTC standards. The dedicated affiliate disclosure page lists every platform we have a commercial relationship with.

The rating and recommendation are decided before the affiliate payout is checked. We have published reviews directly recommending competitors over the affiliate-paying platform on the page when the competitor is the better fit. Five of the 16 reviews on the site explicitly recommend an alternative platform in their verdict.

Who Matt is and why the byline is a pen name

Matt is a pen name. The reviewer is a single Australian based in country Victoria who tests every platform first-hand from an Aussie IP. The pen-name choice is standard for writers in this category: the subject matter is adult AI companion content, the reviewer's personal identity is kept separate from the byline, and that separation is disclosed openly on the About page and at the bottom of every review.

What the pen name does not change: every review is written by the same Australian person doing the same first-person testing from the same country Vic IP. There is no offshore content farm, no AI-generated bulk drafting, no team rotating through the byline. One reviewer, one location, real testing.

What the pen name does not allow: there is no real photo on the About page and no individual professional credential publicly attached to the byline. We are honest about that. If you want to verify the testing claims, the screenshots, dated re-verification stamps, and the live 22-app Playwright audit do the work that a CV cannot.

What we do not do

• We do not republish auto-rewritten competitor articles, transcript-derived posts, or bulk AI output. Every review carries first-hand testing details that could not have been written without using the platform.
• We do not run paid link exchanges, reciprocal link networks, or PBNs. Editorial links earned through actual testing only.
• We do not fabricate Reddit quotes, Trustpilot ratings, or user numbers. If a platform has no real Reddit thread, the review says so explicitly. If a Trustpilot listing does not exist, the review says that too.
• We do not raise a rating because the affiliate payout is high. We do not lower a rating because the affiliate payout is low. The rating reflects the testing.
• We do not publish "best of" lists where every entry is paid placement. Where we publish a ranking (the best AI companion apps in Australia pillar is the main one), the ordering follows AU access quality and product fit, with affiliate-paying status disclosed up front and explicitly held separate from rank.

Frequently asked questions