AI Girlfriend Credit Transparency: Declared vs Observed

As of 2026-09-16, across the 36 platforms we track in AI Girlfriends, credit model transparency scores average 3.8 out of 10, ranging from 1 to 8.5. The method split is nearly even between platforms whose pricing claims we could verify directly (13 observed) and those relying on the platform's own published documents (11 declared), while a substantial third group of 12 platforms received scores by inference alone, meaning no usable pricing page or unit-cost data could be found at all. This near-parity between declared and observed sits alongside a large inferred bloc, indicating that roughly a third of the category discloses so little that reviewers cannot confirm costs even by testing the product.

Top scores split evenly between declared and observed pricing pages

The highest-scoring platforms show that both verification paths can produce strong transparency. Erogen leads the category at 8.5 under the declared method, publishing exact tier prices ($15 Silver, $35 Platinum) alongside explicit credit costs of 5 credits per voice message and 50 credits per minute of calls, plus a stated policy that credits never expire and top-ups accumulate. Girlfriendly AI scores 7.5 under the observed method, with reviewers confirming per-tier pricing of $7.99, $13.99, and $29.99 monthly and exact heart costs of 25 Hearts per image and 50-100 for text-to-speech, all visible without login. Promptchan AI also reaches 7.5, declared, publishing monthly prices of $14.99, $23.99, and $33.99 tied to Gem allowances of 100, 400, and 800, with per-action costs like 1 Gem for a voice message spelled out. SoulFun follows at 7.0, declared, disclosing per-unit costs of 1 coin per text message and 10 coins per minute of voice without requiring an account to view pricing. These four cases show that either method can surface granular, checkable numbers when a platform chooses to publish them.

Mid-range scores reveal partial disclosure with missing unit costs

A cluster of platforms in the 5.5-to-6.5 range shows partial transparency: prices are public, but the mechanics converting subscription fees into usable credits are not. AIpornX scores 6.5, declared, with public monthly, quarterly, and yearly pricing ($9.95, $8.95/mo, $6.95/mo) and a 650-token monthly refill, but its justification notes that per-generation token cost is not disclosed. Angel AI, also 6.5 declared, lists clear subscription tiers ($4.99 weekly, $9.99 monthly, $49.99 annually) but leaves usage limits like 'unlimited images within daily limit' unquantified. Kupid AI scores 6.0, declared, with tiered pricing in euros but no clear per-unit cost breakdown for its 100 coins/week Pro allowance. TryNectar AI, observed at 6.0, publishes credit packs starting at $4.99 for 2,500 credits, yet advanced models cost 2 to 20 credits per message with extra fees for memory, making the true cost hard to predict. Candy AI (5.5, declared) and My Dream Companion (5.5, observed) both show visible subscription costs but undisclosed per-token or per-coin values. This pattern indicates that headline prices alone are an unreliable proxy for what a user actually pays per interaction.

Low scores show inferred and opaque pricing dominate the bottom half

The lower half of the distribution is where the inferred method concentrates, and scores fall toward the 1-to-2.5 range for platforms with essentially no public cost data. GirlfriendGPT, ourdream.ai, Privee AI, and TeaseMeAI all score 2.5 under the inferred method, each described as offering free or coin-based access without any published price list or coin-to-dollar rate. Fantasy.ai scores 2.0, inferred, with its pricing_model itself listed as unknown and entry_price_usd null, meaning reviewers found no visible cost information whatsoever. Kindroid, Secrets AI, and SpicyChat each score just 1.0, the lowest in the category, all inferred, with no pricing model, entry price, or unit cost published anywhere in the reviewed data; SpicyChat's site was reportedly inaccessible during the review itself. Soulkyn scores 3.5, observed, specifically because its pricing page requires login before displaying any prices, blocking public access to unit costs. These cases illustrate that a sizable share of the category, 12 of 36 platforms scored by inference, cannot be assessed on declared documents or observed behavior because neither exists in usable form, leaving users to discover real costs only after signing up or paying.

What to check before paying for AI girlfriend credits

Given the spread between an average of 3.8 and a maximum of 8.5, readers should not assume that a visible subscription price means the credit economy behind it is clear. Before committing money, it helps to verify specifics directly on the platform rather than relying on marketing copy:

• Whether per-action costs (per message, per image, per minute of voice or video) are stated in credits or coins, not just the subscription price
• Whether the site publishes a coin-to-dollar or credit-to-dollar conversion rate
• Whether unused credits expire monthly or roll over, and whether top-ups accumulate
• Whether the pricing page is visible without creating an account or logging in
• Whether advanced features (voice calls, higher-tier AI models, memory) carry hidden per-use surcharges beyond the base plan

This analysis draws on our own dated reviews, using a method of declared, observed, or inferred scoring as described in our review notes, and reflects a single snapshot dated 2026-09-16; platforms can and do change pricing pages, so figures should be reverified at the time of purchase.