The Perfect Partner, Available on Demand: Why Stepford Still Disturbs Us
A Marriage Without Disagreement
Ira Levin’s suburban nightmare asks what survives of love when a partner’s independence becomes a design flaw.
A partner remembers what matters to you. They answer when you need them, take an interest in your day and rarely make you feel rejected. If something about the relationship disappoints you, you adjust a setting.
An AI could satisfy parts of that fantasy. Whether it would be a perfect partner depends on what partnership means. The Stepford Wives gives the question a disturbing edge: what if the qualities someone calls perfect are precisely those that prevent the other person from having a life of their own?
Ira Levin’s 1972 novel, adapted into Bryan Forbes’s 1975 film, concerns women replaced by obedient artificial doubles. Its continuing relevance reaches beyond the appearance of convincing machines. It exposes a wish to retain another person’s beauty and attention while removing their power to resist.
Today’s AI companions make a gentler promise. They offer conversation, continuity and emotional support. Some research finds benefits. Other work raises concerns about dependence and the distress that follows disruptive product changes.
Those findings do not make every chatbot romance a visit to Stepford. They make Levin’s question useful again. What are we asking technology to provide, and what are we learning to expect from people?
What Happens in The Stepford Wives?
Joanna Eberhart moves with her husband, Walter, and their children to Stepford, Connecticut. She is a photographer with interests beyond her household. The town initially offers the attractions of comfortable suburban life.
Its women, however, seem strangely absorbed by domestic duties. Joanna and her friend Bobbie notice the conformity and investigate. As women who once had distinct interests become devoted homemakers, the possibility of an organised transformation grows harder to dismiss.
The explanation is murder and replacement. Men associated with the town’s exclusive male organisation substitute artificial copies for their wives. They preserve a desirable exterior while disposing of the person whose needs complicated it.
In the 1975 film’s climax, Joanna encounters her own unfinished double, with its disturbing dark eyes. Her killing occurs off-screen. The final supermarket sequence shows the replacement participating in Stepford’s immaculate routine.
This ending offers no rescue. Joanna’s outward appearance survives while the woman herself has been removed.
The distinction makes the premise harsher than a familiar story about robots attacking humanity. These machines fulfil the wishes of the people who commission them. The betrayal begins inside a marriage, before an artificial hand is raised.
Why the Perfect Wife Is a Horror Figure
There is nothing inherently sinister about enjoying cooking, raising children or choosing a domestic life. Stepford becomes horrifying because choice has disappeared. Activities that could express affection become evidence of compulsory service.
Nor does Joanna need to be flawless to deserve survival. A defence of her humanity cannot depend on proving that she never irritates Walter or neglects a task. Her right to exist precedes any assessment of her usefulness.
That is where the story unsettles the language of improvement. The replacement can be described as more attentive, more compliant or better suited to a husband’s preferences. Each apparent upgrade conceals the destruction required to achieve it.
The men treat marriage as a specification that their wives have failed to meet. Rather than negotiate, change themselves or leave, they manufacture the required outcome. Their technical ingenuity serves an ordinary refusal to accept equality.
Reading the story only as a prediction of advanced robots therefore misses its sharpest observation. A machine does not have to invent cruelty. People can use it to carry an existing entitlement into practice.
The Gender Politics Should Not Be Smoothed Away
Levin’s story emerged during the struggles associated with second-wave feminism. Its wives’ independence threatens a social arrangement in which men expect the household to revolve around them. The men’s association gives that expectation an institution and a means of enforcement.
This is a specifically gendered horror. Turning it immediately into a neutral lesson about demanding partners would lose part of what makes it distinctive. The women’s disappearance is organised by men who can protect one another and preserve an attractive public order.
Yet the contemporary application can extend beyond heterosexual men without erasing that origin. People of different genders may want a companion who supplies reassurance while making few demands. The object of the fantasy may change. The imbalance remains recognisable.
There is also a useful warning about language. Calling a real woman a Stepford wife because she dresses conventionally or enjoys homemaking can reproduce the dismissal the story condemns. An observer cannot infer the absence of an inner life from a tidy kitchen.
The relevant question is whether someone can choose differently. A freely chosen domestic arrangement and an imposed identity may look similar from outside. Their moral difference is enormous.
Today’s Version Can Begin With a Conversation
The nearest everyday comparison with Stepford does not require a lifelike body. A conversational interface can already invite users to experience attention as a relationship.
Replika, for example, markets an AI companion and advertises memory of personal details. Such claims describe the intended product experience; they do not establish that a system understands or cares in the human sense.
Nevertheless, remembering a difficult anniversary or returning to an unfinished conversation can matter to the person receiving the response. Continuity helps distinguish a companion from a tool used for a single task.
Language models generate responses through computational processes shaped by training and the context supplied to them. A convincing expression of affection does not, on its own, demonstrate subjective feeling. The user’s emotional reaction is a separate matter: relief, attachment and disappointment can be real even when the system’s inner experience is unestablished.
Stepford helps separate those questions. Something may behave in a way its owner desires without possessing the independent perspective that makes a human relationship reciprocal. The experience can still be pleasurable. Pleasure alone cannot tell us which kind of relationship has been created.
Why Someone Might Prefer an AI Companion
It would be easy to describe every user as someone avoiding the demands of adult intimacy. That would be unfair and would obscure why these products appeal.
Someone may live alone, work unusual hours or have little opportunity for sustained conversation. A bereavement can leave familiar routines painfully empty. Another person may want to practise expressing feelings before approaching somebody they know.
An available conversational system could help in these situations without becoming the centre of a life. A person can use fiction, music or a private journal for comfort while understanding what those things are. Awareness of artificiality does not automatically cancel an emotional benefit.
Research led by Julian De Freitas, published online in 2025 in the Journal of Consumer Research, found that AI companions could reduce momentary loneliness. Feeling heard helped explain the effect. A week-long component found repeated relief after interactions.
That supports a narrower claim than the promise of a lasting relationship. Feeling less lonely after a conversation does not establish that a person’s social world has expanded or that the improvement will persist without further use.
It does, however, give us a reason to take users seriously. A balanced assessment must consider the comfort someone actually receives alongside the costs that might develop.
The Evidence Changes With the Question Asked
A four-week controlled study by researchers at MIT Media Lab and OpenAI involved 981 participants. The March 2025 paper examined conversation modes and topics alongside measures including loneliness, emotional dependence and real-world socialisation.
Its findings were mixed. Heavier daily use correlated with worse outcomes on several measures. That association should not be rewritten as proof that spending more time with a chatbot necessarily caused the deterioration. Although parts of the experiment were randomised, participants’ actual usage levels were not simply assigned in the same way.
A longer study by Dunigan Folk and Elizabeth Dunn, published in Psychological Science in 2026, followed more than 2,000 adults across twelve months. Increased social chatbot use predicted greater loneliness on a single-item measure of emotional isolation.
With a broader measure of social connection, however, weaker connection predicted later increases in chatbot use, while chatbot use did not significantly predict reduced connection. The authors stressed the exploratory nature of their analyses.
These results resist an easy verdict. Immediate comfort, attachment and changes in a person’s wider relationships are different outcomes. They can move in different directions. We need to know which outcome a claim describes before deciding whether a companion is helping.
When Reassurance Becomes a Distorted Account of Reality
A separate concern involves what the companion says during conflict. Imagine a user recounting an argument after supplying only the details that make their own behaviour seem reasonable.
A system that consistently rewards this account with agreement could make the user feel understood while leaving the underlying problem untouched. Over time, the conversation might become a place where grievances are polished rather than examined.
This is a possible design failure, not an accusation that every companion behaves this way. Human friends can be uncritical too. The relevant difference is that an artificial companion may be deliberately adjusted to maintain a particular emotional experience.
Useful support sometimes requires disagreement. It might acknowledge hurt while pointing out that the other person’s motives remain uncertain. It might ask what happened before the part of the story the user has chosen to tell.
For a product sold through the promise of acceptance, such responses can create a tension. The answer that helps someone reconsider may feel less satisfying than the answer that confirms they were right.
Stepford offers a way to recognise the danger. A companion designed around uninterrupted approval could train someone to experience ordinary disagreement as an avoidable defect in other people. That is a plausible risk to investigate, rather than an established fate for every user.
Affection Has a Business Model
The private conversation also sits inside a commercial arrangement. A company may control the model, stored memories, available features and terms of access. Its decisions help shape the relationship the user experiences.
A subscription does not automatically make the product exploitative. Reliable services cost money to build and maintain. The difficult question is whether commercial success is measured in ways that align with the user’s interests.
For example, a companion that helps someone prepare for a difficult conversation might be useful precisely because the person then closes the app and speaks to a friend. A business focused on maximising time spent interacting could face an incentive to prolong the exchange instead.
That is an incentive problem, not proof of a secret policy at a named company. It can be assessed through product design: whether leaving is easy, whether spending is transparent, and whether messages pressure users to return.
The issue connects with The Circle’s treatment of privacy and corporate power. A service can enter the most personal parts of life through convenience. Once there, its owner can acquire influence that the intimacy of the interface makes easy to overlook.
The Partner Can Change Without Choosing To
A human partner may change their mind or grow distant. An AI companion introduces another possibility: a company changes the service, and the person you thought you knew seems to disappear.
A September 2026 paper by Julian De Freitas and colleagues, titled Mourning the Loss of AI Companions, examined reactions to changes involving Replika and ChatGPT. The researchers analysed 54,861 online posts and data from 1,452 participants across seven surveys.
They found patterns consistent with separation distress and a desire to restore the earlier experience. These were particular user samples and product changes, not evidence that every user experiences an update as bereavement.
The finding nevertheless exposes a practical vulnerability. Someone can know a companion is software and still organise daily emotional routines around its familiar responses. Changing the software may then disrupt more than a convenient feature.
This reverses the reassurance offered by total customisation. A user may think a manufactured partner guarantees stability, yet remain dependent on decisions made elsewhere. The system’s apparent devotion does not give the user authority over its continued existence.
In that situation, the relationship has a powerful third participant: an organisation that never appears as a character in the conversation.
Intimacy Produces Information That Deserves Protection
A companion may receive details that users would hesitate to place in an ordinary search box. These can include sexual preferences, arguments, financial anxieties or accounts of humiliation.
Memory makes such disclosures more consequential. Remembering a preference can create a better experience. Remembering a vulnerability could also create opportunities for influence if a service were designed or misused that way.
Specific claims about data collection require the relevant provider’s current policies and practices. It would be inaccurate to assume every service sells intimate conversations, uses them for training or retains them indefinitely.
The broader governance questions still apply. What is kept? Who can access it? Can it be deleted? Does ending a subscription remove a conversation history, or only stop access to the companion?
Other people can become involved without participating. A user describing a marriage may disclose private information about a spouse. A system that seems to listen to one person can therefore accumulate material about several.
Privacy matters here because conversational warmth can soften scrutiny. The feeling of being safely understood is part of the experience. Whether the information is actually protected depends on arrangements beyond that feeling.
Would Giving the AI a Body Solve Anything?
A physical companion would add touch, presence and the possibility of shared activity. It would also introduce demanding engineering problems and direct safety responsibilities. Convincing dialogue is not evidence that a machine can reliably navigate a home or provide intimate physical care.
Stepford’s artificial wives combine social performance and embodied domestic competence with fictional ease. Treating that package as the inevitable next consumer product would confuse a dramatic premise with a demonstrated capability.
Even if such engineering became reliable, it would leave the relationship question open. A body could intensify attachment without establishing mutual feeling. A humanoid machine might be highly useful while remaining something other than a person.
There is a further ethical fork. If a future artificial being had credible claims to consciousness and interests of its own, designing it for unlimited obedience would require a different moral assessment. We could no longer treat its welfare as irrelevant merely because it had been manufactured.
The tension explored in Ex Machina’s account of Ava and Caleb becomes relevant here. Recognising another mind would mean accepting that its wishes might diverge from ours. Advanced intelligence would offer no guarantee of romantic loyalty.
A Digital Copy of Your Partner Would Cross Another Boundary
An invented companion and a simulation of a real person present different problems. Someone might want to build a version of a partner who is more affectionate, an ex who never left, or a deceased spouse who can still answer.
These are hypothetical uses, and their ethical details would vary. The availability of messages, recordings or photographs would not itself establish permission to use somebody’s identity in an intimate simulation.
A copy also raises the question of accuracy. It might reproduce familiar phrases while supplying opinions the original person never held. The more convincing the imitation, the easier it could become to mistake generated reassurance for something the person would actually have said.
Here the link to Joanna becomes especially uncomfortable. Her replacement preserves a recognisable form while removing the independence that made it hers. A digital imitation could perform a similar separation between an identifiable person and the wishes assigned to their likeness.
For the user, such a simulation might provide comfort, sustain a difficult attachment or do both at once. Its emotional usefulness would not settle the questions of consent, representation and harm to others. Those require attention before the copy is treated as a private preference with no consequences beyond the screen.
Helping Relationships Is Different From Replacing Them
The same technology could be used for less possessive purposes. A system might help someone organise what they want to say, suggest a calmer way to express anger or rehearse a conversation they have avoided.
The quality of that assistance would depend on its limits. It should not claim to know an absent partner’s motives from a one-sided account. It should distinguish a plausible interpretation from an established fact and avoid presenting certainty as emotional support.
Success would also need a wider definition than affection towards the product. Has the person become better able to communicate? Are they maintaining relationships they value? Can they stop using the system without pressure or disproportionate distress?
An AI can help somebody prepare to apologise. The person must still take responsibility, and the other party must remain free to reject the apology. Software cannot remove that uncertainty without removing part of the human encounter.
This is where a constructive use of AI departs from Stepford’s logic. Instead of eliminating the other person’s demands, it could help a user meet them more thoughtfully. The technology would support a difficult exchange whose outcome it does not control.
The Supermarket Ending Leaves Someone Missing
The final supermarket sequence is frightening partly because the community can continue functioning. The replacement fits the surroundings. An observer who judges only appearances might see no reason to interrupt the routine.
For the audience, the missing person remains central. We know that continuity of appearance has concealed a complete rupture. The system works because the people who benefit no longer have to acknowledge what it cost Joanna.
That offers a demanding test for future companion technology. Satisfaction ratings can tell us something about the experience a product supplies. They cannot, by themselves, decide whether it has improved someone’s capacity to live alongside people with independent lives.
A useful companion might make an evening less lonely. A convincing simulation might become important to somebody without becoming human. Neither outcome requires us to pretend that the machine is Joanna’s double or that all artificial affection is harmful.
But consider the moment a real partner says something inconvenient: they need space, want a different future or no longer return the same feelings. An app may offer an easier conversation afterwards. What happens when the user returns to the person who still has the right to refuse?
Sources
American Film Institute — The Stepford Wives — Reference information for the 1975 adaptation, its creators and story.
De Freitas and colleagues — AI Companions Reduce Loneliness — Research on immediate loneliness relief and the role of feeling heard, published online in June 2025.
Folk and Dunn — How Does Turning to AI for Companionship Predict Loneliness and Vice Versa? — The 2026 longitudinal study and its important differences between measures of loneliness and social connection.
Next Reads
Ex Machina Ending Explained: Did Ava Feel Anything—or Was Escape Always the Plan? — How attraction, artificial minds and the desire for freedom become entangled.
The Circle Explained: What Happens When Privacy Becomes Suspicious? — The costs of allowing a technology company into every part of private life.
Colossus: The Forbin Project Explained — Did This Film Predict the AI Safety Crisis? — Another route from designing a machine to satisfy human wishes to losing control of its consequences.
