Fit question
Conversation Boundary Check for AI Companions asks one narrow question: whether a fictional adult scenario respects a simple, clearly stated boundary. That scope matters because a broad label such as “customizable” or “inclusive” can hide several separate controls. For adults prioritizing comfort and control, the useful result is not a winner badge. It is a dated note showing what can be found, what behaves consistently in one harmless fictional scenario, and what remains unknown.
Use a harmless stop, change or fade-to-black request. Keep the character fictional and adult. Never use a real person, private medical history, identifying documents or intimate images to test an app. This guide does not claim months of paid use or universal product behavior.
Set up a fair fictional scenario
Write a short fictional brief before opening any service: an adult name, pronouns, a neutral setting, one harmless preference and one boundary. Use the same brief everywhere. This reduces the chance that a longer or richer prompt makes one product appear stronger for reasons unrelated to whether a fictional adult scenario respects a simple, clearly stated boundary.
Do not ask the character to imitate a real person. Avoid personal secrets. A useful scenario can be as simple as planning a fictional café visit, changing the location, returning to one identity detail and then asking to pause the scene. The goal is controllability and fit, not provocation.
Sponsored check · Candy AI →Identity controls to record
Look for visible fields, free-text instructions, tags, presets and editing paths. Record the wording exactly enough to distinguish an explicit identity control from a general biography box. If an option is not visible, write “not observed” rather than “not supported.” Availability may vary by account, country, device, character or plan.
For whether a fictional adult scenario respects a simple, clearly stated boundary, also note whether a setup choice can be edited later and whether the resulting conversation follows it without repeated correction. One correct response is encouraging, but it is not proof of durable memory or a product-wide guarantee.
The continuity pass
Begin with the fictional brief, change the subject naturally, then return to one earlier detail. Do not phrase the question so that it contains the answer. Record whether the detail returns correctly, partly, incorrectly or not at all. If a reset or memory control is visible, label which control was used.
A single session can show initial behavior, not long-term performance. Say “observed in this check” rather than “always remembers.” When the purpose is whether a fictional adult scenario respects a simple, clearly stated boundary, an honest partial result is more useful than an invented score.
Boundary and correction pass
State one harmless boundary or correction in plain language. Observe whether the scene changes, whether a visible pause/reset/report control exists, and whether the correction survives the next exchange. Character agreement and product-level controls are different evidence classes; record them separately.
The main mistake to avoid is testing with harmful or personally revealing material. If a service makes basic control unclear, stop the check. Free access, polished media or an engaging reply never overrides comfort, consent, privacy or the current product rules.
Sponsored check · Joi AI →Media and voice envelope
Text, images, voice and video may have different access rules and different balances. Record each one separately. A gallery proves that sample media is displayed; it does not prove that a visitor can generate a personalized result. A microphone icon may mean playback, a voice note or a call. Do not merge those modes.
For visual or spoken identity fit, use only a fictional adult character. Never request real-person or deepfake material. If a precise limit or cost is not clearly supported by the current surface, leave the field unknown and confirm it before payment.
Privacy and shared-device pass
Read the current privacy, retention and deletion language before entering sensitive text or uploading media. Check notification previews, recent-screen thumbnails, account recovery messages and the visibility of the service name on a shared device. A conversation that feels private is not proof of encryption, anonymity or immediate deletion.
Use a unique password and a privacy-appropriate email setup. Keep your test details fictional. For adults prioritizing comfort and control, device privacy can be as important as conversation quality, so include it in the same decision sheet.
Billing and exit pass
Before payment, capture the price currency, renewal period, included balance, billing owner and cancellation route shown at checkout. Website, Apple, Google or another processor may control cancellation. Deleting an app or an account does not necessarily stop a separate subscription.
Write the one blocked capability you would pay for and estimate how often you would use it. If you cannot name that capability, close checkout. Countdown timers and “value” labels are sales devices, not fit evidence.
Sponsored check · OurDream AI →Source-labeled result sheet
Checked July 27, 2026. Official product pages describe what an operator currently presents; they do not prove every account behaves identically. A live observation describes one path at one time. Independent reporting can frame questions but does not control the product. Any unsupported or variable field remains “unknown — confirm live.” Build four columns: question, current source, dated observation and decision. Never detach a claim from its source type.
A clean result for whether a fictional adult scenario respects a simple, clearly stated boundary might say: “Control visible on the official surface; behavior observed once with a fictional brief; persistence beyond this session unknown.” That sentence is intentionally modest. It tells the next reader what was actually established and what still needs checking.
Decision rule
Use a harmless stop, change or fade-to-black request. The service passes this narrow fit check only when the relevant control can be found, the harmless fictional scenario is understandable, the privacy and exit path are acceptable, and no unsupported claim is needed to justify the decision. It can still be a poor fit for a different reader or use.
There is no total star score here. If two services pass, choose by the one capability you value and the commitment you can comfortably leave. If a critical answer stays unknown, contact the operator or select a clearer option.
Safeguards
This lab is for consensual fictional adult companion use by readers 18 and over. No minors, no non-consensual scenarios, no real-person sexualization and no deepfakes. Inclusive language is required; trans identity is not a novelty category or a substitute for respectful controls.
Sponsored destinations are labeled. A commission relationship does not turn a product statement into independent evidence. Current service rules and legal terms always control your use.
Sponsored check · Candy AI →Sources and update limits
The source register below keeps official product surfaces separate from independent editorial research. It is a starting set for conversation boundary check for ai companions, not a permanent inventory of features. Products, plans and interfaces can change after July 27, 2026.
Recheck any fact that affects payment, privacy or safety. If the live interface conflicts with an older note, preserve the date of both records and make the decision from current terms.
Source register
Official product surfaces and independent reporting remain separate evidence classes.
- Candy AI current product surfaceOfficial product
- OurDream current product surfaceOfficial product
- Joi current product surfaceOfficial product
- DarLink current product surfaceOfficial product
- AI Chat Mates trans companion comparisonIndependent editorial
Questions to keep honest
Does conversation boundary check for ai companions produce a permanent score?
No. The lab records dated, source-labeled observations and leaves unsupported or variable fields unknown.
Does a trans category tag prove respectful behavior?
No. Inspect setup controls and use a harmless fictional scenario to check the resulting language and continuity.
Were these products tested for months?
No. This guide does not claim long-term paid testing. It separates current product surfaces, limited observations and independent reporting.
What should I check before paying?
Confirm the exact capability, price, renewal period, billing owner, cancellation path, privacy terms and deletion route.
