AI Companion Privacy & Safety Index

Verified means we checked the fact directly against the vendor's official product page, pricing page, account-access information or published policy on the date shown in the row. A review-site price, a Reddit comment, a marketing screenshot or a user report can help us decide what to inspect, but it does not become an index fact unless we can re-find it on an official source.

That standard matters in AI companion apps because the category changes quickly. Pricing pages move behind signup walls, free tiers shrink, NSFW rules change, apps rebrand and some products disappear while old listicles keep recommending them. If a fact is not publicly reproducible, we leave the cell as not published or not publicly disclosed instead of laundering a guess into the table.

Because we looked and the answer is not published. "Not in policy" means the vendor has a policy, but the specific fact we were checking is not documented there. "Price not published" means we could not find a public pre-signup price on the vendor's own pages. We label the absence precisely rather than filling the gap with a third-party estimate.

The absence is still useful information. If an intimate companion app does not publish deletion details, retention language, data-sharing posture or a price before signup, the user has less leverage before handing over conversations, images, payment details or account data. The label is not a moral judgment; it is a transparency signal.

No. The row is based on the same structured facts whether a tool has an affiliate relationship with AI Haven or not. A partner does not get a stronger privacy score for hiding pricing, and a non-partner still gets credit for a clear policy, a self-serve deletion path or a public no-sale claim.

Affiliate links can affect how AI Haven earns money from outbound clicks, but they do not change the verified fields, the scoring labels or the evidence standard. The index is designed so the important claims can be checked against the vendor's own public surfaces rather than taken on trust.

The verified set is re-checked monthly, and sooner when a shutdown, rebrand, pricing change, privacy-policy change, NSFW-policy change or age-gate change is reported. Every row shows its own last-verified date because freshness matters more at the row level than at the page level.

Use the date as part of the decision. A recent verification is stronger than an old one, but no directory can guarantee that a vendor did not change a checkout, policy or moderation rule after the last pass. For any tool you are about to pay for, the safest workflow is to use this index as the shortlist filter, then confirm the live vendor page before subscribing.

They are two of the clearest public signals in a category built around intimate chat, fantasy roleplay and sometimes explicit media. An age gate does not prove a product is safe or compliant, but it does show whether an adult-positioned app takes any visible step to keep minors out before use. No visible age gate is a weak signal because it leaves more of the burden on the user.

NSFW policy clarity matters for a different reason: users should know whether explicit content is actually permitted before they create an account or pay. Some apps imply adult use in ads while keeping policy language vague; others clearly allow, limit or prohibit explicit content. The index separates those states so expectations are set by published rules, not by vibes.

Each label is a coarse, verifiable judgment — not a fake-precision score. Strong means the fact is clearly stated on the vendor's own page or policy. Partial means the topic is addressed but hedged, conditional, request-based or not specific enough to count as a clean disclosure. Weak means the opposite signal is present, such as no visible age gate on an adult-positioned app. Not publicly disclosed means we checked standard paths, footers, sitemaps, pricing surfaces and policy text, and the answer was not published.

The labels are meant to make comparison fast without pretending there is a mathematically exact privacy score. A Strong label usually means the user can compare that fact confidently. A Partial label means read the policy before relying on it. Weak and Not publicly disclosed are caution labels: they do not automatically make a tool unusable, but they do mean the vendor is giving you less clarity before signup.

Start with the facts that are hardest to unwind later: public pricing, cancellation path, account deletion, privacy policy, data sharing, chat or media retention, NSFW policy and age gate. A polished companion demo is not enough if the vendor hides the recurring price, does not explain deletion, or gives no public answer about how intimate data is handled.

A practical workflow is to test the free tier with low-risk prompts first, then check the row for anything marked Weak, Not publicly disclosed or price not published. If you plan to upload images, use voice, buy credits or maintain long-running roleplay history, the privacy and deletion fields matter more than a generic star rating.

Because stale recommendations are a real problem in this market. Companion apps can shut down, rebrand, change domains, remove public pricing or leave users searching for a product that no longer works. Keeping dead or caution-marked rows visible helps users understand why an old recommendation may be risky or outdated.

Breach and retention notes are included for the same reason: they are decision-grade context, not trivia. If a companion app has a documented breach, unusually long retention language, no public policy or no published price, that does not always answer whether the app is good at chat. It does answer whether the user should slow down before trusting it with intimate data.