This page explains how we capture and verify the pricing data behind every review and comparison on TopAiChatGirls. It is the companion to our scoring rubric: the rubric explains what we judge, this page explains where the numbers come from and how we keep them honest. We publish it because most “AI girlfriend pricing” pages quote a single promotional figure and move on — we don’t.
How we capture pricing
For each platform we review, we record pricing from two independent classes of source: (1) the vendor’s own pricing or help page, and (2) at least one independent 2026 review or pricing tracker. Every figure is stored with its source URL and the exact date we fetched it. When sources disagree, we keep both numbers and cite both sources — that divergence is itself a finding (a promotional “starting at” price is rarely the price you actually renew at).
Our current pricing dataset covers 6 platforms — Candy AI, Replika, Character.AI, Kindroid, Nastia, and Nomi — and was last fully priced on 14 June 2026 against 18 cited sources. Reviews that quote these numbers carry a “Last priced” date so you can see how fresh the figure is.
How we derive cost-per-use figures
Some of the most useful numbers — the true monthly cost on an annual plan, the saving versus paying monthly, the cost of each included image — aren’t printed on any vendor’s pricing page. We calculate them from published tier limits and label them as derived, never as test results. The arithmetic is reproducible:
- Annual effective monthly = annual total ÷ 12. Example: Candy AI Premium at $71.88/yr works out to $5.99/mo, not the $12.99 monthly sticker.
- Annual saving vs monthly = how much the yearly plan cuts off the monthly rate. Replika Pro is the steepest we found: $19.99/mo drops to an effective $5.83/mo on the year — a 71% saving.
- Cost per included image = monthly price ÷ the hard monthly image cap, where the vendor publishes one. Candy AI Premium’s 100 included images at $12.99 work out to about $0.13 per image.
Current pricing snapshot
The headline figures below are the ones our reviews draw from. Every entry is cross-checked against vendor and independent sources; where a platform’s pricing is genuinely in flux, we say so in the right-hand column rather than pretending a single number is settled.
| Platform | Monthly (sticker) | Best annual (effective/mo) | Pricing finding to know |
|---|---|---|---|
| Candy AI | $12.99 | $5.99 (54% off) | Token top-ups can push real cost to $30–$80+/mo for heavy media users; Ultra is $16.99/mo. |
| Replika | $19.99 (Pro) | $5.83 (71% off) | Biggest monthly→annual saving we found. Ultra tier added in 2026 at $29.99/mo ($119.99/yr). |
| Character.AI | $9.99 (Plus) | $7.92 ($94.99/yr, 20% off list) | No NSFW/mature tier at any price — the main reason users migrate away from it. |
| Kindroid | $13.99–$14.99 (Standard) | — | Range reflects stackable Ultra/Max add-ons; mature roleplay is permitted on Standard. |
| Nastia | $11.99 (Basic) / $15.99 (Unlimited) | — | High source divergence: some 2026 sources cite $19.99/mo or $129/yr. Verify the tier at signup. |
| Nomi | $8.33–$15.99 | — | Cheapest entry point to NSFW-capable chat in this set, across tiers/billing cadence. |
What the data tells us across platforms
- Sticker price rarely equals real cost. Candy AI advertises a low promo rate but a token system can multiply the bill; Character.AI’s “$9.99” is only the start once you account for its missing NSFW tier.
- Annual billing is where the real discount lives. Replika Pro (71% off) and Candy AI (54% off) reward the yearly commitment far more than most readers expect.
- Highest pricing opacity: Candy AI (token-driven real cost) and Nastia (sources disagree by up to $8/mo) — both get an explicit “real cost vs sticker price” callout in their individual reviews. We dig into Nastia’s tiers and how it stacks up in our Nastia vs Replika comparison.
What this page does not claim yet
We hold ourselves to a hard rule: we never publish a number we didn’t measure or verify. Pricing is publicly verifiable, so it’s here. Several of our scoring axes are behavioural and require a paid account used over a full month — those measurements are in progress and are deliberately absent from the pricing dataset rather than estimated:
- Response latency (timed live sessions per platform)
- 30-day memory retention accuracy (details seeded in week one, recalled in weeks three and four)
- Photo-quality and facial-consistency scoring (at least 20 generations per platform, graded)
- Personality consistency across 100+ messages
When a behavioural axis is measured, it appears in the relevant review with the test window and date attached — never as a guess. If you ever see a figure on this site without a source or a test date behind it, treat it as a bug and tell us.
See also: How we score · Best AI girlfriend apps in 2026 · Our honest verdict