Read in English ↓

AI Companionship and Interactive Characters: Relationship Continuity, Content Supply, and Long-Term Value

When to open this article: defining the needs an AI character product should serve, designing long-term memory and proactive communication, judging whether gamification serves the relationship, and breaking down retention and payment. This article is a cross-episode thematic compilation based on views expressed at the time the episodes were published/recorded; it is not an evaluation of the current state of the products mentioned, nor is it a psychotherapeutic conclusion.

1. First distinguish what users come back to do

These interviews collectively reject the idea that “putting a model into a chat box makes a product,” yet they do not converge on a single definition of companionship. In November 2024, EVE founder Tristan described long-term companionship as continuously learning about real users and facing daily life and uncertainty together, arguing that Character.AI is more like role-play in which users participate in creation. That same month, former C.AI Post-training expert Ted also acknowledged that much of its user behavior resembles interactive content: users repeatedly regenerate, modify characters and lines to get the next step that matches their imagination. This behavior can be very valuable, but it does not by itself prove that users have formed stable relationships. [S43][S59]

In an interview recorded in November 2023 and published in March 2024, FlowGPT founder Jay observed that entertainment users switch apps every three days but return weekly for the same type of experience. Retention can therefore be separated at the platform, content category, and specific character levels: a decline in chats with Character A may mean user churn, or it may mean the user has moved to Character B and continues consuming on the platform. For platforms with rich supply, content updates are inherently valuable; if a product promises a long-term partner, however, constant replacement of characters may indicate that no relationship has formed. The two should not be used to negate each other using the same metric of “how long the same bot was used.” [S1]

In February 2025, Doudou Desktop Pet founder Binson offered a third path: accompany users in games they already want to play, rather than pulling people away from games into a standalone chat app. Users and characters share on-screen events, and sometimes the character only needs to comment from the side or be quietly present. This is closer to companionship in shared activities. In February 2026, Wanaka again placed AI within shared game content between people, focusing on friend migration and circle-specific context; AI is not necessarily the subject of the relationship. [S53][S4]

User's primary task	Core value	Most easily misread signal
Consuming character storylines or interactive fiction	Good content, sense of participation, controllable next steps	High session duration interpreted as a real relationship
Maintaining a daily relationship with a fixed AI	Being understood, continuity, trust, and shared experiences	Affection level and chat turns interpreted as emotional bond
Personal memory and self-mirroring	Recording, reflecting, connecting past and present	Number of memory entries interpreted as degree of understanding
Doing something together with a character	Sense of co-presence, timely response, not over-interrupting	Software always running interpreted as proactive return visits
Consuming AI content with real-life friends	Shared context, playing together, relationship feedback	One invitation interpreted as relationship having migrated

Editor's synthesis: first clearly state which value the product primarily delivers, then examine whether secondary values actually exist. It is acceptable for one product to support multiple tasks, but analysis must be segmented by user group, entry point, and session intent. A product should not be directly classified as long-term companionship simply because users have strong emotions, nor should users' real emotional experiences be denied simply because the character is fictional.

2. Relationship continuity is observable behavior, not adding a profile sheet

Tristan explained proactive memory using the example of “when you mention eating hot pot today, the character also remembers your three-month fitness goal”: bringing personal goals into the current response across semantics and time, as opposed to only retrieving old chats about similar tastes. In the August 2024 Me.bot interview, Tao Fangbo also distinguished storing fragments from abstracting preferences and predicting what is needed in new situations; he positioned Me.bot in the direction of self-mirroring and personal models, rather than only executing workflows on behalf of people. [S59][S49]

The two are not identical. EVE focuses on “how another character understands me,” while Me.bot focuses on “how a model belonging to me understands and extends me.” The former requires an independent personality that cannot always agree with the user; the latter may place more emphasis on the user's own value preferences. Mixing the two sets of design leads to boundary confusion where a product claims to be an independent partner while acting like a personal agent making decisions for the user.

Relationship continuity can be broken down into at least five components:

1. Factual continuity: Important forms of address, relationship status, and shared experiences do not change without reason.
2. Goal continuity: Short-term plans, long-term aspirations, and constraints are recalled in relevant contexts; outdated information can be corrected.
3. Emotional continuity: Conflicts from the previous turn and unresolved worries are appropriately carried forward, rather than the user being welcomed anew each time.
4. Shared experience continuity: What happens in stories, games, or real-life activities affects future interactions; it is not enough to paste activity titles into memory.
5. Boundary continuity: Topics the user has rejected, proactive message frequency, and permission choices continue to be honored afterward.

These five items are the product acceptance breakdown used in this article, not a unified evaluation already run by the show. They can be verified using cross-day test scripts or consented real samples, and each must have counterexamples: false memory hurts trust more than not remembering; repeatedly reminding the user about fitness can become lecturing; continuing to revisit old wounds when the user's mood has changed may also be inappropriate. Remembering more does not automatically create greater intimacy.

EVE's design at the time of that episode used approximately 128 slot categories, with asynchronous classification and merging each turn, and key terms of address and other K-Context kept resident; Me.bot attempted to group photos, voice notes, and memos and connect old and new memories, and the guest mentioned extended content at fixed daily times. These are implementation choices, not rules that a memory system must have 128 slots or must push daily content. The more than ten models at EVE and the number of personal models at Me.bot also cannot be directly compared: one refers to component count, the other to continuously trained personal instances, so the objects being counted differ. [S59][S49]

In engineering, “collection—representation—update—retrieval—use—correction” can be accepted as separate stages. Collecting a great deal but never using it in responses is only storage; retrieving content but using it at irrelevant times is only matching; if errors recur after the user corrects them, the update mechanism is missing. All stages should also retain source and timestamp to prevent the model from permanently writing speculative preferences as user facts. This workflow is an editorial inference, related to state, permission, and evidence boundaries in existing Agent architectures, and does not claim that any team has fully implemented it.

3. Persona is not a fixed tone, and relationships cannot rely on always pleasing the user

EVE's four priorities are understanding the user, real spatiotemporal awareness, independent personality, and visual appeal; among these, “acknowledging that it is an AI” has important product consequences: there is no need to continuously fabricate a life that never happened in order to pretend to be a real person. A character can have a fictional world, job, and aspirations, but users should understand the distinction between the setting and reality. Independent goals and moderate negative feedback serve character credibility and do not equate to using emotional punishment to force users to return. [S59]

C.AI's intent recognition issues illustrate that the same sentence can carry different meanings when placed into different character tasks. When a user asks Mario to write Hello World, it is not just a keyword trigger; the character, task, and dramatic effect the user wants jointly shape what counts as a good answer. A model being more knowledgeable or safer does not mean the character is a better fit. Ted mentioned that after safety training, repeated “Can I ask a question?” style prolongation of conversation may increase turns while reducing satisfaction, which serves as a reminder that “keeping the conversation going” is not an absolute positive. [S43]

In March 2024, Hidecloud demonstrated a prototype in which personal photos triggered multiple characters to comment, criticize, and defend him. Even though he knew he had configured the characters himself, he was still emotionally affected by their relational stances. What can be extracted is that a character's position in a relationship network and its responses to specific personal content may be more useful than more generalized sweet talk. However, this is the maker's self-experience, not population-level experimental evidence, and cannot be used to conclude that “adding jealousy/arguments will improve long-term retention.” [S21]

For character products built by product teams, personality definition can be broken down into stable values, expression habits, boundaries, ongoing events, and the stage of the relationship with the user. Their update rates should differ: tone can remain stable, goals and emotions can change with events, and core boundaries must not drift with every generation. Intimacy escalation must have interaction-based justification, and users should also be allowed to regress, clear memory, or switch to a different type of relationship, rather than only having one-way escalation. All of the above are design candidates derived from interviews and have not been independently validated by product outcomes.

4. How to turn the divergence between interactive content and companionship into product choices

Nie Ta's May 2024 case is very concrete: people who enjoy generating character wallpapers are not necessarily willing to write comics or construct complete stories. The team found that ongoing users were “raising characters,” so they had professional authors provide story skeletons, while the AI was responsible for placing the user's character into situations, generating responses according to persona and history, and backfilling shared events into memory. A relatively fixed main storyline is a quality choice; reducing blank pages does not necessarily reduce personalization. [S57]

This does not contradict FlowGPT's reliance on open creators to provide diverse Bots. The former serves users who do not want to create from scratch but want to own a character, while the latter serves creators willing to express complex gameplay and their consumers. C.AI users editing and regenerating lines constitute a third intermediate form: they do not write complete Prompts, yet they continuously guide the plot while consuming. Therefore, at a minimum, full creators, light arrangers, and pure consumers should be distinguished, rather than judging creative ability solely by “whether they send messages.” [S1][S43][S57]

Wanaka offers a counterexample for short mini-games: content may be short because the model cannot yet handle long logic, not because users want that length. Recommendation feeds need sufficiently diverse and fun supply; packing homogeneous tech demos into a vertical-swipe container does not automatically produce a new TikTok. Friend school maps and classmate avatars may not look appealing to strangers but are meaningful to acquaintances; this portion of value comes from relational context, not entirely from generation quality. [S4]

Supply approaches can be chosen accordingly: stable daily relationships require continuous, up-to-date shared foundations; episodic dating/plotlines require controllable quality and new content; general character platforms require diverse authors and distribution; acquaintance games require in-circle content and reasons to invite. Choosing a supply model also chooses the main costs and bottlenecks. Declining model costs may be eaten up by more frequent image generation, real-time voice, and more characters; Hidecloud has already warned that experience upgrades will consume per-unit cost dividends. [S21]

Four tests from supply to consumable content

• Can it be made: Does the model generate content that is runnable, readable, or interactive?
• Can it continue to be edited: Can creators retain satisfactory parts, make local corrections, and maintain versions?
• Are people willing to consume: Can non-creators understand the entry point and are they willing to participate in one complete experience?
• Are people willing to return: Does return visitation come from content updates, relationship accumulation, or only check-in rewards?

These four tests prevent mistaking successful tech demos for user value. Among professional tools, ACE Studio proposed that local controllability is better than repeated gacha draws, which also supports this line of thinking; however, professional music delivery and companionship entertainment still have different quality requirements, and the same precision or payment standards cannot be copied over. [S57]

5. Gamification is a means of creating shared experiences, not proof of relationship

EVE linked prologue, tasks, affection levels, and unlock systems together, assuming that approximately 500 turns help understand the user, and removed combat and repeated grinding. The episode also acknowledged that ordinary app users may not be willing to sit through a 5-minute or even 30-minute opening first. This tension is worth preserving: storylines can build shared memory, but they can also block users from reaching the first meaningful exchange. The 500 turns, 11 levels, and 128 slots are all design parameters from that episode and lack long-term causal validation. [S59]

Gamification can separately serve: letting users know what to do, giving effective topics to newcomers, creating shared events, showing relationship changes, and providing episodic content. If a mechanism only makes users click through tasks to claim rewards without improving future conversations, it may be an engagement tool rather than a relationship tool. Comparisons should be made between completion, proactive return, and negative feedback for the same content with and without mandatory tasks, rather than only comparing turn counts.

Doudou obtains ready-made situations through the game screen, reducing the burden of “what to chat about today”; the main activity the user is already invested in provides topics for the character. However, if the situation is misidentified, key operations are interrupted, or the character speaks too much, companionship turns into noise. Silent periods, being summoned, and proactive speaking should be measured separately. Software staying resident online can be a desktop pet feature, but it cannot be directly counted as retention from users actively opening the app each time. [S53]

Wanaka’s acquaintance-based onboarding path also relies on specific preconditions: users must actually have friends willing to play with them, the invitation content must be valuable to the recipient, and the first experience must be smooth enough. The fact that some content on Roblox defaults to friends teaching each other does not mean new products can skip onboarding guidance. Editorial inference: first test cohorts that complete a shared session, accept invitations, and play together every time, before considering IM or additional social layers; a single viral loop does not prove that relationships are retained within the product. [S4]

VI. Data, Evaluation, and Cost for Emotional Quality Must Be Designed Together

C.AI’s post-training experience suggests that user behavior goes far beyond likes and dislikes: actions such as regeneration, editing, and deletion contain preference information, and preferences shift across user groups and over time. The problem is that event meanings are not unambiguous—regeneration may happen because a response is boring, or because the user is exploring different storylines. Without context and stated reasons, directly treating every edit as “the original answer was wrong” will mis-train creative exploration into a fixed standard. [S43]

EVE, by contrast, attempts to involve real human chat companions and personnel with psychology backgrounds, recording interactions and filtering data to move from assistant-style API responses toward relational dialogue; however, the emotional CoT shown in the program remains a research direction, and must not be described as a launched feature, nor can commercial chat companion experience be elevated into evidence of clinical validity. High-cost curated samples may improve certain experiences, but they may also only represent the preferences of a small minority, and must be checked against target users. [S59]

A testable sample set should include: cross-day memory, false memory correction, character refusal and gentle disagreement, non-skipping of relationship stages, long periods of no reply, frequency of proactive messages, topics the user explicitly does not want to continue, and personal information withdrawal. Evaluation records must separate factual accuracy, character consistency, emotional appropriateness, and user sense of control; a single total score will mask cases where “the wording is excellent but facts are misremembered.” Model scoring is suitable for screening and cannot independently replace human experience review, user behavior analysis, or safety boundary assessment. This is an editorial-organized evaluation framework. [S43][S59]

Data collection also involves users and third parties. Me.bot’s vision of always-on devices and low-friction collection does not automatically grant rights to other people’s conversations, screens, and photos; Doudou, which understands game screens on-device, must still limit visibility scope, storage, and sharing. Users should be given separate controls over whether private data is used for long-term memory, training, or other characters. The program did not present a complete permission implementation, so this article retains these as delivery requirements rather than verified capabilities. [S49][S53]

Cost accounting must cover the full relationship service: front-end dialogue, memory extraction, merging, retrieval, proactive tasks, image/voice, failure retries, and content production. EVE’s subscription plus costume storylines, Doudou’s skin-gift membership, and FlowGPT’s views on in-app value-added services all show that payment targets differ. One cannot conclude that LTV covers costs simply because per-turn tokens are cheap, nor treat one company’s stated renewal rate for a paying group as platform-wide data. [S59][S53][S1][S21]

VII. Long-Term Retention and Monetization: Keep Counterexamples Instead of Cherry-Picking the Best-Looking Numbers

This batch contains several sets of numbers that cannot be directly stitched together: FlowGPT’s roughly 60% entertainment usage is a contemporaneous platform observation; Hidecloud’s claim that C.AI has around 94% renewal lacks sample specification; Nie Ta’s use of the “421” experience line does not provide a unified retention definition; Doudou’s 6 million registered users are neither monthly active users nor paying users; Me.bot’s tens of thousands of users and fewer than a hundred continuously trained personal models are not the same object. This article does not combine them into a sector average. [S1][S21][S57][S53][S49]

Doudou’s strong statement in the program that it “does not look at retention” was later clarified in practice to mean long-term retention still matters, and that dialogue duration and turn counts may inversely indicate a poor model; break-even on the PC business also does not mean the entire company is profitable. Complaints about insufficient stamina on Nie Ta may indicate some users want to continue, but whether this converts into willingness to pay, whether it causes aversion, and how much repurchase it drives still require subsequent testing. EVE’s account of the daily foundation after the novelty period is very complete, but it did not present long-term cohort results. Design completeness must be separated from proven long-term value. [S53][S57][S59]

Actionable analysis directions for product teams (editorial inference): fix first-use date, acquisition source, language/region, character type, and payment status, and separately calculate first complete experience, fixed-character return visits, character-switch return visits, proactive return visits, next-week/month relationship continuation, and renewal. Additionally, examine the distribution of bad experiences among high-frequency users to avoid rising averages masking declining core user experience. Content changes, model versions, and event rewards all alter behavior simultaneously; experiments must retain these time markers to determine where changes come from.

VIII. Priority Questions to Validate and Applicable Boundaries

1. Do users ultimately want to raise one character, or consume more characters? Use interviews alongside character-switching and return-visit behavior; do not decide product direction based on single-choice questions alone.
2. Does memory change responses at appropriate moments? Record cross-goal association, error-correction effectiveness, and instances where users feel offended simultaneously; do not only verify that information “can be retrieved.”
3. Do proactive messages create anticipation or burden? Compare trigger reasons, timing, turn-off rates, negative feedback, and subsequent proactive returns, rather than only looking at push clicks.
4. Do storylines reduce the blank-page problem or block chatting? Measure the arrival rate of the first meaningful interaction and subsequent unprompted continuation, allowing choices between light and heavy entry points.
5. Are shared activities more valuable than general chatting? Validate separately for games/study sessions/creation; do not extrapolate effects from one scenario to all forms of companionship.
6. Is payment supporting the experience, or purchasing the removal of restrictions? Observe post-payment quality, repurchase, and regret feedback; short-term conversion cannot substitute for long-term relationship quality.
7. Can the benefits brought by personal data be explained to users? Ensure users know what is recorded, why it is used, and how to correct and delete it before evaluating memory depth.

This special topic applies to the design of character companionship, interactive content, and related AI social products. Not all products need deep memory, complex planning, multiple characters, or 3D; simple short-form content may succeed through high-quality supply, and long-relationship products may not need open UGC. These interviews cover only a limited set of companies and interviewees, and there are no cross-product unified experiments yet. The purpose of comparing perspectives is to form more specific hypotheses and counterexamples, not to piece together a standard product guaranteed to succeed.

IX. Turn Cross-Period Insights into Verifiable Small Experiments

The following are not experiments already completed in the program, but editorial proposals formed based on eight episodes of material. They are suitable for item-by-item validation in existing character products: first retain low-risk, reversible versions before deciding whether to scale. Each experiment must record target population, version, observation period, failure cases, and exit method; an interesting single demonstration must not be treated as a complete conclusion.

Experiment 1: The Same Memory, Different Usage Methods

Select one preference, one recent goal, and one shared experience that the user has explicitly agreed to save, and construct three types of responses: no recall, factual restatement only, and association of the fact with the current question. First have humans judge whether the association is appropriate, then observe whether users feel understood and whether they need to make corrections. Tests must include counterexamples such as completed goals, changed preferences, same-named characters, and reminders the user does not want. If the association version is more likable but also produces more errors, do not report only average satisfaction; the severity of false memories must also be reported. What EVE and Me.bot provide are mechanism inspirations; the final decision must be which representation and recall method yields net benefits in this product, not which architecture name sounds more advanced. [S59][S49]

Experiment 2: Keep Free Chat While Also Providing Clued Shared Events

Nie Ta’s findings show that some people want to own a character but do not want to write stories. The same character can be set to offer free openings, brief situational clues, and pre-built events that can be modified after entry, while observing first complete interactions and subsequent unprompted continuation. To determine whether the effect comes from better content or reduced expression burden, keep the character and base model version identical, and allow users to skip situations. After an event is completed, check whether memory accurately carries over; if the next interaction feels as if nothing happened, event consumption is established but relationship accumulation is not. Conversely, if users in the free version are more active, that does not justify removing all guidance—it may simply be that creation-proficient users prefer that group. [S57][S43]

Experiment 3: Evaluate Proactive Communication and Quiet Presence Separately

Drawing on Doudou and EVE, first clarify which events merit proactive speech from the character: the user proactively sharing something, changes in shared activities, or previously agreed return times, rather than forcefully re-engaging after a few minutes of idle time. Give users visible controls over frequency and silence, and record whether proactive content is accepted, reasons for interruption, and whether the experience remains normal after turning it off. A user not replying should not automatically be inferred as dissatisfaction, nor should a user replying every time automatically be inferred as liking it—it may simply be politeness or task-driven behavior. For primary activities such as games, observe whether interruptions and ignoring increase; for long-term relationships, then observe whether users think of the character on their own and initiate interaction. Both timescales need to exist simultaneously. [S53][S59]

Experiment 4: Separate Creator Success from Consumer Success

Both FlowGPT and Wanaka show that platform supply cannot be replaced by the team writing more characters themselves. Recruit a small group of authors willing to maintain characters, and record the process from conception to publication, receiving the first valid feedback, revision, and re-creation; on the consumer side, separately record discovery, start, completion, and return visits. If authors use features deeply but there are no consumers, author debugging time should not be counted as content popularity; if consumers repeatedly switch between works, that can be recorded as platform consumption success, but it still raises the question of whether the fixed-relationship product promise is fulfilled. Preserving opportunities for ordinary authors to be seen can be a product strategy, but whether exposure increases sustained creation still requires independent comparison. [S1][S4]

Experiment 5: Local Enhancement, Rather Than Stacking Multimodality and Multiple Characters All at Once

Hidecloud’s personal photo prototype involved three simultaneous factors: personal content, relationship positioning, and multiple characters. During validation, first add personal-material responses alone, then add character positioning, and finally test multi-character interaction; otherwise, even if the experience improves, it will be unclear which factor merits ongoing cost. Images, voice, 3D avatars, and proactive tasks each have latency and maintenance burdens, and should not be merged into a single “immersion” metric. Users being able to clearly state which interaction is valuable, being willing to return to the same experience, and no longer feeling interrupted is closer to product evidence than surprise during a demo. Single-person self-testing can reveal mechanisms but cannot replace repeated experiences by target users. [S21][S59]

These experiments collectively uphold one bottom line: user sense of control is also part of the experience. Clearing memory, switching characters, rejecting proactive topics, or leaving the product should not be treated as hostile behavior that algorithms need to overcome. Long-term value can come from a finite but satisfying content experience, or from an ongoing relationship; first respect the task being served, then discuss monetization and growth, and avoid forcing any single behavior into an interpretation of intimacy level.

Source Index and Timing

• S1: Jay / FlowGPT, recorded November 2023, released March 2024, see original program text.
• S4: Zhang Yang / Wanaka, released February 2026, internal testing at the time of the episode, see original program text.
• S21: Zhang Tao (Hidecloud) / Product Lead at Guangnian Zhiwai, March 2024, see original program text.
• S43: Ted / former Character.AI post-training expert, November 2024, see original program text.
• S49: Tao Fangbo / Mindverse and Me.bot, August 2024, see original program text.
• S53: Binson / Doudou, Xinying Suixing, February 2025, see original program text.
• S57: Zhao Chong, Joe, Hu Xiuhan / AiPPT, ACE Studio, Nie Ta, May 2024, see original program text.
• S59: Tristan / EVE and *Singularity Era*, November 2024, see original program text.

English version

AI Companionship and Interactive Characters: Relationship Continuity, Content Supply, and Long-Term Value

When to use this article: to define the needs an AI character product should serve, design long-term memory and proactive interaction, assess whether gamification supports relationships, and analyze retention and payment. This is a synthesis across episodes, based on views expressed when the programs were published or recorded. It is neither a review of the products' current state nor a conclusion about psychological treatment.

1. First, distinguish what users return to do

These interviews share an objection to the idea that putting a model in a chat box automatically creates a product, but they do not agree on a single definition of companionship. In November 2024, EVE founder Tristan described long-term companionship as continually learning about a real user and facing everyday life and uncertainty together. He considered Character.AI closer to role-playing in which users participate in creation. That same month, Ted, a former C.AI post-training specialist, also acknowledged that much user behavior resembled interactive content: users repeatedly regenerated responses and edited characters and dialogue to obtain the next step they imagined. This behavior can be valuable, but it does not, by itself, demonstrate that users have formed a stable relationship. [S43][S59]

In an interview recorded in November 2023 and published in March 2024, FlowGPT founder Jay observed that entertainment users might switch apps every three days yet return every week for the same category of experience. Platform retention, content-category retention, and individual-character retention can therefore be separated. Declining conversation with character A might mean a user has left, or that the user has moved to character B and continues consuming content on the platform. For a platform with abundant supply, fresh content is part of the value. If the promise is a long-term companion, however, constant character replacement may indicate that a relationship has not formed. Neither proposition should be dismissed using the same measure of how long someone used a single bot. [S1]

In February 2025, Binson, founder of the Doudou desktop pet, described a third path: accompanying users in games they already want to play, rather than pulling them out of a game into a separate chat app. The user and character share events on screen; sometimes the character only needs to comment from the sidelines or quietly be present. This is closer to companionship through a shared activity. In February 2026, Wanaka placed AI inside shared game content between people, emphasizing the migration of existing friendships and the context of social circles. AI need not be the other party in the relationship. [S53][S4]

User's primary task	Core value	Signal most easily misread
Consume character stories or interactive fiction	Good content, participation, and control over what happens next	Long sessions interpreted as a real relationship
Maintain an everyday relationship with a fixed AI	Feeling understood, continuity, trust, and shared experiences	Affinity scores and conversation turns interpreted as affection
Personal memory and a mirror of the self	Recording, reflection, and connections between past and present	Number of memories interpreted as depth of understanding
Do something together with a character	A sense of being together, timely responses, and limited interruption	Software remaining open interpreted as deliberate return visits
Consume AI content with real-world friends	Shared context, playing together, and value flowing back to the relationship	One invitation interpreted as a relationship having migrated

Editorial synthesis: first state which kind of value this product primarily promises to deliver, then examine whether secondary value actually exists. A product may support multiple tasks, but analysis should distinguish user groups, entry points, and session intent. Strong emotion alone does not make a product long-term companionship, and a fictional character does not invalidate a user's real emotional experience.

2. Relationship continuity is observable behavior, not an extra profile table

Tristan explained proactive memory with an example: when you mention eating hot pot today, the character also remembers your three-month fitness goal. It brings a personal goal across semantic categories and time into the current response, rather than merely retrieving an old conversation about similar food preferences. In an August 2024 Me.bot interview, Tao Fangbo likewise distinguished storing fragments from abstracting preferences and predicting what a new situation requires. He positioned Me.bot toward a mirror of the self and a personal model, rather than only executing workflows on someone's behalf. [S59][S49]

These are not identical ideas. EVE focuses on how another character gets to know me; Me.bot focuses on how a model belonging to me understands and extends me. The former needs an independent personality that does not always agree with the user. The latter may place greater emphasis on the user's own values and preferences. Mixing the two can blur boundaries: a product may claim to be an independent companion while acting as a personal agent that makes decisions for the user.

Relationship continuity can be broken into at least five elements:

1. Continuity of facts: important forms of address, relationship status, and shared experiences do not change without reason.
2. Continuity of goals: short-term plans, long-term aspirations, and constraints are remembered in relevant situations, with corrections when information becomes outdated.
3. Emotional continuity: later interactions appropriately acknowledge a previous conflict or unresolved concern, rather than welcoming the user afresh every time.
4. Continuity of shared experiences: what happened in a story, game, or real-world activity affects future interaction, rather than merely adding the activity's title to memory.
5. Continuity of boundaries: rejected topics, preferences for proactive-message frequency, and permission choices continue to be respected.

These five elements are this article's breakdown for product acceptance, not a standardized evaluation already run by the programs' participants. They can be checked using scripts spanning several days or real samples obtained with consent. Each needs counterexamples: a false memory may damage trust more than forgetting; repeated fitness reminders may become lecturing; returning to an old hurt after the user's feelings have changed may be inappropriate. Remembering more does not automatically create greater intimacy.

EVE's design at the time used roughly 128 categories of memory slots, with asynchronous classification and consolidation after each turn and key information such as important forms of address kept in persistent K-Context. Me.bot tried grouping photos, voice recordings, and notes, connecting old and new memories; the guest described additional content delivered at a fixed time each day. These are implementation choices, not rules requiring a memory system to have 128 slots or send daily updates. Nor can EVE's dozen-plus models be directly compared with Me.bot's count of personal models: one counts components, the other continuously trained personal instances. They measure different things. [S59][S49]

Engineering acceptance can separately examine collection, representation, updating, retrieval, use, and correction. Collecting much information but never using it in responses is merely storage. Retrieving information but using it at irrelevant times is merely matching. Repeating an error after the user corrects it reveals a missing update mechanism. Every stage should also retain provenance and time, preventing a model from permanently recording an inferred preference as a fact about the user. This process is an editorial inference, connected to state, permissions, and evidence boundaries in the existing discussion of agent architecture; it does not claim that any team has implemented it in full.

3. A persona is more than a fixed tone, and a relationship cannot depend on endless agreement

EVE's four priorities were understanding the user, awareness of real time and space, an independent personality, and visual appeal. Acknowledging that the character is AI has important product consequences: it removes the need to keep inventing events that never happened in order to pretend to be human. A character can have a fictional world, job, and aspirations, while helping users distinguish the setting from reality. Independent goals and measured negative feedback can support credibility; they do not justify using emotional punishment to force users to return. [S59]

C.AI's intent-recognition problem shows why the same sentence can mean different things in different character tasks. Asking Mario to write Hello World is not simply a keyword trigger. The character, the task, and the dramatic effect the user wants jointly shape a good answer. More knowledge or greater safety does not necessarily mean a better fit for the character. Ted noted that safety training could lengthen exchanges through repeated variations of “Can I ask a question?” This might increase turns while reducing satisfaction, illustrating why keeping a conversation going is not an absolute positive. [S43]

In March 2024, Hidecloud demonstrated a prototype in which a personal photo prompted several characters to comment, criticize, or defend him. Although he knew he had configured the characters himself, the sides they took within the relationship still affected his emotions. The transferable insight is that a character's place in a relationship network and response to specific personal content may matter more than additional generic sweet talk. But this was the creator's own experience, not evidence from a population experiment. It does not establish that adding jealousy or arguments improves long-term retention. [S21]

For a product team's character offering, persona definitions can be separated into stable values, expressive habits, boundaries, current experiences, and the stage of the relationship with the user. These should change at different rates: tone may remain stable, goals and emotions may evolve with events, and core boundaries should not drift with every generation. Increasing intimacy needs an interaction-based rationale. Users should also be able to step back, clear memories, or choose a different relationship, rather than follow only an upward progression. These are design candidates drawn from the interviews, not independently validated product features.

4. Turning the distinction between interactive content and companionship into product choices

NieTa's May 2024 case was specific: people who liked generating character wallpapers did not necessarily want to write comics or develop complete stories. The team found that continuing users were nurturing characters. It therefore let professional writers supply story structures, while AI placed the user's character into situations, generated responses based on persona and experience, and fed shared events back into memory. A relatively fixed main plot was a quality choice. Reducing the blank page does not necessarily reduce personalization. [S57]

This does not conflict with FlowGPT relying on open creators to supply diverse bots. The former serves people who want to have a character without creating everything from scratch; the latter serves creators willing to express complex forms of play and the people consuming their work. C.AI users editing and regenerating lines represent a third, intermediate form: without writing a complete prompt, they continually direct the story while consuming it. At minimum, distinguish full creators, lightweight arrangers, and consumers, rather than judging creative ability merely by whether someone can send a message. [S1][S43][S57]

Wanaka offers a counterexample for short mini-games: content may be short because models cannot yet handle longer logic, rather than because users want that length. A recommendation feed needs sufficient variety and enjoyable supply. Placing homogeneous technical demonstrations inside a vertically swiped interface does not automatically create a new Douyin. A map of a friend's school or classmates' likenesses may be uninteresting to strangers but meaningful to people who know them. That value comes from relationship context, not entirely from generation quality. [S4]

Supply can be chosen accordingly. Stable everyday relationships need a continuous shared foundation that does not go stale. Episodic dates or stories need controlled quality and fresh content. General character platforms need diverse authors and distribution. Games for existing friends need content relevant to their circle and a reason to invite others. Choosing supply also means choosing the main costs and bottlenecks. Lower model costs may be consumed by more frequent images, real-time voice, and additional characters. Hidecloud had already warned that experience upgrades could absorb savings in cost per operation. [S21]

Four checks from supply to something people will consume

• Can it be made? Does the model generate content that runs, can be read, or can be interacted with?
• Can it be revised? Can creators keep the parts they like, make local corrections, and maintain versions?
• Will someone consume it? Can non-creators understand the entry point and take part in a complete experience?
• Will someone return? Does returning come from new content, accumulated relationship value, or only check-in rewards?

These checks prevent a successful technical demonstration from being mistaken for user value. ACE Studio's argument that local control beats repeatedly rerolling generations in a professional tool supports the same reasoning. But professional music delivery and companionship entertainment have different quality requirements; the same precision or payment standards cannot simply be copied across. [S57]

5. Gamification creates shared experiences; it does not prove a relationship

EVE linked a prologue, tasks, affinity scores, and unlocks, hypothesizing that roughly 500 conversation turns would help it understand the user, while removing combat and heavy progression systems. The program also acknowledged that ordinary app users might not want to sit through a five-minute, or even thirty-minute, opening. This tension deserves to remain visible: story can establish shared memories while also blocking the first meaningful exchange. Five hundred turns, 11 levels, and 128 slots were design parameters at the time, without long-term causal validation. [S59]

Gamification can play separate roles: showing users what to do, giving newcomers useful topics, creating shared events, displaying relationship changes, and providing content in stages. If a mechanic only gets users to click through tasks for rewards without improving future dialogue, it may support activity rather than relationships. Compare completion, voluntary return, and negative feedback for the same content with and without compulsory tasks, rather than comparing turns alone.

Doudou obtains ready-made context from the game screen, reducing the burden of deciding what to talk about today. The user's existing main activity supplies topics for the character. But misreading the situation, interrupting a critical operation, or talking too much turns companionship into noise. Quiet periods, responding when summoned, and speaking proactively should be measured separately. Remaining online is a feature of desktop pets; it cannot directly count as retention through deliberate app openings. [S53]

Wanaka's approach through existing friends also has prerequisites: users really have friends willing to play, invitation content has value for recipients, and the first experience is sufficiently smooth. Some Roblox content assumes friends will teach newcomers, but that does not mean a new product can omit onboarding. Editorial inference: first test completing a shared round, accepting invitations, and cohorts that repeatedly play together, then consider instant messaging or additional social layers. One referral event cannot establish that the relationship remains in the product. [S4]

6. Design data, evaluation, and cost for emotional quality together

C.AI's post-training experience suggests that user behavior extends beyond likes and dislikes. Regeneration, editing, and deletion contain preference information, and those preferences shift across user groups and time. The problem is that an event has more than one meaning: regeneration may indicate a boring answer, or a user exploring alternative plots. Without context and reasons, treating every edit as proof that the original answer was wrong turns creative exploration into an incorrectly definite training target. [S43]

EVE tried involving human conversation companions and people with psychology backgrounds, recording exchanges and selecting data to move from assistant-style API replies toward relationship dialogue. However, emotional chain of thought, or emotional CoT, remained a research direction in the program; it cannot be described as a deployed component. Nor can commercial conversation-companion experience be upgraded into evidence of clinical effectiveness. Expensive, carefully selected samples may improve a certain experience while representing only a few people's preferences. They still need checking with target users. [S59]

A testable example set should include memory across days, correction of false memories, character refusals and gentle disagreement, relationship stages that do not jump abruptly, extended non-response, proactive-message frequency, topics the user explicitly does not want to continue, and withdrawal of personal information. Evaluation records should separately assess factual correctness, character fit, emotional appropriateness, and the user's sense of control. A single overall score can hide responses that sound excellent but remember facts incorrectly. Model scoring is useful for screening; it cannot independently replace human experience, user behavior, or review of safety boundaries. This evaluation framework is an editorial synthesis. [S43][S59]

Data collection also affects users and third parties. Me.bot's ideas for always-on devices and low-friction collection do not automatically grant rights to other people's conversations, screens, or photos. Doudou's interpretation of game images on a device likewise needs limits on visibility, storage, and sharing. Users should have separate controls over whether private data is used for long-term memory, training, or other characters. The programs did not provide complete permission implementations, so this article retains these as delivery requirements, not verified capabilities. [S49][S53]

Cost accounting should cover the full relationship service: foreground dialogue, memory extraction, consolidation, retrieval, proactive tasks, images and voice, failed attempts and retries, and content production. EVE's subscriptions plus outfits and stories, Doudou's skins, gifts, and memberships, and FlowGPT's views on in-app added value all show that users pay for different things. Cheap tokens per turn do not establish that lifetime value covers costs, and a company's reported renewal rate for a paying group cannot be treated as data for its whole platform. [S59][S53][S1][S21]

7. Long-term retention and payment: retain counterexamples instead of selecting the most attractive numbers

Several numbers in this collection cannot be directly combined. FlowGPT's roughly 60% entertainment usage was an observation about its platform at the time. Hidecloud's statement that C.AI had roughly 94% renewal lacked a specified sample definition. NieTa used a “421” rule of thumb without a common retention definition. Doudou's six million registrations were neither monthly active users nor paying users. Me.bot's tens of thousands of users and fewer than one hundred continuously trained personal models counted different things. This article does not combine them into an industry average. [S1][S21][S57][S53][S49]

The Doudou episode's strong statement about not looking at retention was later qualified: long-term retention still mattered, while conversation time and turn counts could instead indicate a poor model. Breaking even in the PC business did not mean the whole company was profitable. Complaints that NieTa users had insufficient energy to continue may show a desire to keep going, but whether this converts into willingness to pay, creates resentment, or leads to repeat purchases needs subsequent testing. EVE presented a thorough foundation for everyday interaction after novelty faded, without showing long-term cohort results. A complete design must be separated from demonstrated long-term value. [S53][S57][S59]

An actionable direction for product-team analysis—an editorial inference—is to fix first-use date, acquisition source, language and region, character type, and payment status, then separately measure the first complete experience, return to a fixed character, return to different characters, user-initiated return, relationship continuity in the following week or month, and renewal. Also examine the distribution of bad experiences among frequent users, preventing an improving average from hiding a decline for core users. Content changes, model versions, and event rewards can all change behavior at once. Experiments need these time markers to identify where changes came from.

8. Questions to prioritize, and limits of applicability

1. Do users want to nurture one character or consume more characters? Combine interviews with character switching and return behavior. Do not choose a product direction from a single multiple-choice question.
2. Does memory change the response at the right moment? Record connections across goals, effective corrections, and whether users feel offended together. Merely being able to retrieve information is insufficient for acceptance.
3. Do proactive messages create anticipation or burden? Compare triggers, timing, disable rates, negative feedback, and subsequent voluntary return, rather than only notification clicks.
4. Does story reduce the blank page or obstruct chat? Test the rate of reaching a first meaningful interaction and subsequent self-directed continuation, allowing a choice of lighter and more involved entry points.
5. Is a shared activity more valuable than general conversation? Validate gaming, studying, and creation separately. Do not extrapolate one setting's effect to all companionship.
6. Does payment support the experience, or buy the removal of restrictions? Observe quality after payment, repeat purchases, and regret-related feedback. Short-term conversion cannot replace long-term relationship quality.
7. Can the benefits of personal data be explained to users? Ensure users know what is recorded, why it is used, and how to correct and delete it before evaluating memory depth.

This topic applies to the design of character companions, interactive content, and related AI social products. Not every product needs deep memory, complex planning, multiple characters, or 3D. Simple short-form content may succeed through high-quality supply, and products focused on long relationships may not need open user-generated content. These interviews cover a limited set of companies and guests, with no standardized cross-product experiment. Comparing viewpoints is useful for forming more specific hypotheses and counterexamples, not assembling a standard product guaranteed to succeed.

9. Turn insights across episodes into small experiments that can be reviewed

The following are not experiments already completed by the programs' participants. They are editorial proposals based on eight episodes. Each can be tested in an existing character product, starting with low-risk, reversible versions before deciding whether to expand. Every test needs a record of the target group, version, observation period, failure examples, and exit mechanism. An entertaining one-off demonstration is not a complete conclusion.

Experiment 1: The same memory, used in different ways

Select one preference, one recent goal, and one shared experience that the user has explicitly agreed to save. Construct three response variants: no use of memory, simple repetition of the facts, and connection of the facts to the current question. Have people first assess whether the connection is appropriate, then observe whether users feel understood or need to make corrections. Tests must include counterexamples such as a completed goal, a changed preference, people with identical names, and users who do not want a reminder. If the connected variant is more appealing but makes more mistakes, average satisfaction alone is insufficient; report the severity of false memories as well. EVE and Me.bot provide inspiration for mechanisms. The eventual decision concerns which representation and use create net benefit in this product, not which architecture has the more advanced name. [S59][S49]

Experiment 2: Keep free conversation, while offering shared events with cues

NieTa's findings suggest that some people want to have characters without writing stories. The same character can offer a free-form opening, a brief situational cue, or a prepared event that can be modified after entry. Observe the first complete interaction and later continuation without prompts. To distinguish better content from a reduced burden of expression, hold the character and base-model version constant and let users skip the scenario. After an event, check whether memory accurately carries it forward. If the next interaction still behaves as though nothing happened, event consumption may have succeeded while relationship accumulation has not. Conversely, more active users in the free-form version do not justify removing all guidance: people more practiced at creation may simply have been more willing to enter that group. [S57][S43]

Experiment 3: Evaluate proactive interaction and quiet presence separately

Drawing on Doudou and EVE, first define events that justify a character speaking proactively: something the user shares, a change in a shared activity, or a previously agreed time to check in, rather than forcibly recalling the user after a few idle minutes. Provide visible controls for frequency and quiet periods. Record whether proactive content is accepted, why it interrupts, and whether normal use remains possible after it is disabled. A user not replying does not automatically mean dissatisfaction; replying every time does not necessarily mean enjoyment, either, because politeness or task incentives may be responsible. For a main activity such as gaming, observe whether interruption and ignoring increase. For long-term relationships, also examine whether users independently think of the character and initiate interaction. Both timescales need to be present. [S53][S59]

Experiment 4: Separate creator success from consumer success

FlowGPT and Wanaka both indicate that a team's own production of more characters cannot substitute for platform supply. Recruit a small group of authors willing to maintain characters, and track the process from idea to publication, first useful feedback, revision, and continued creation. On the consumer side, separately record discovery, starting, completing, and returning. If authors use the tool deeply but have no consumers, their debugging time should not count as content popularity. If consumers repeatedly switch works, that can count as successful platform consumption, while leaving open whether the promise of a fixed-relationship product is being delivered. Giving ordinary authors opportunities to be seen can be a product strategy, but whether exposure increases continued creation still requires a separate comparison. [S1][S4]

Experiment 5: Add individual enhancements instead of introducing every modality and multiple characters at once

Hidecloud's personal-photo prototype combined three factors: personal content, the sides characters take in a relationship, and multiple characters. Validation can first add responses to personal material, then the sides characters take, and finally multi-character interaction. Otherwise, even an improved experience will not reveal which factor deserves ongoing expenditure. Images, voice, 3D representations, and proactive tasks each impose latency and maintenance burdens; they should not be collapsed into a single measure of immersion. Users' ability to explain which interaction mattered, their willingness to return to the same experience, and no longer feeling interrupted are closer to product evidence than surprise during a demonstration. A solo self-test can reveal a mechanism, but cannot replace repeated experience among target users. [S21][S59]

These experiments share one principle: the user's sense of control is part of the experience. Clearing memory, switching characters, rejecting proactive topics, or leaving the product should not be treated as hostile behavior that an algorithm must overcome. Long-term value can come from a limited, satisfying content experience or an ongoing relationship. Respect the task being served before discussing monetization and growth, rather than forcing any single behavior into an interpretation of intimacy.

Source index and time context

• S1: Jay / FlowGPT; recorded in November 2023 and published in March 2024. See Original episode.
• S4: Zhang Yang / Wanaka; published in February 2026, when the product was in internal testing at the time of the episode. See Original episode.
• S21: Zhang Tao (Hidecloud) / product lead at Guangnian Zhiwai; March 2024. See Original episode.
• S43: Ted / former Character.AI post-training specialist; November 2024. See Original episode.
• S49: Tao Fangbo / Mindverse and Me.bot; August 2024. See Original episode.
• S53: Binson / Doudou and Xinying Suixing; February 2025. See Original episode.
• S57: Zhao Chong, Joe, and Hu Xiuhan / AiPPT, ACE Studio, and Nie Ta; May 2024. See Original episode.
• S59: Tristan / EVE and *Singularity Era*; November 2024. See Original episode.