AI Relationship Advice: Can Chatbots Help Couples Communicate or Make Conflict Worse?
Author: Ukrainian Psychological Hub · Published: September 17, 2026 · Editorial Policy
Published: September 17, 2026 · Editorial Policy
AI relationship advice has moved from novelty to ordinary couple life. A person can now open a general-purpose chatbot after an argument, paste a text exchange, ask who was unfair, request an apology, rehearse a difficult conversation, or ask whether a relationship should continue. The same system can feel like a calm writing partner, an always-available confidant, an informal mediator, and an apparently neutral judge.
The science is now substantial enough to move beyond speculation, but not strong enough to treat chatbot advice as established relationship care. A 2026 systematic review of 21 empirical studies across 11 countries found that people use generative AI for nonjudgmental advice, emotional support, and communication facilitation. Users often experience the responses as empathic and helpful, while objective evaluations still find weak expert alignment, inconsistent responses, and important safety limitations. Short-term relationship outcomes in intervention studies are promising but preliminary, and the strongest randomized trial did not show a broad advantage over an established writing exercise. Levkovich and Alon, 2026
The practical conclusion is straightforward. AI can be useful for thinking before a conversation, finding language for feelings, generating alternative interpretations, and slowing an impulsive response. It becomes much less reliable when it is asked to decide who is right, infer a partner's motives, diagnose a person from a story, determine whether someone is cheating, or make high-stakes decisions for a couple. The value of AI relationship advice depends less on whether the chatbot sounds wise than on what role the couple gives it.
Can AI relationship advice actually help?
Yes, within a limited role. The best-supported use is as a reflective language tool: a system that helps a person organize thoughts, examine more than one interpretation, prepare questions, or turn an accusatory draft into a clearer statement of feelings and needs. Early studies also suggest that people may experience generative AI as empathic and usable during a single relationship-support interaction. In a qualitative study of 20 participants using GPT-4 for a relationship difficulty, participants often described the interaction as helpful, human-like, and clarifying, while researchers also identified poor risk assessment and difficulty reaching genuinely collaborative solutions. Vowels, Francois-Walcott, and Darwiche, 2024
A randomized controlled trial gives a more demanding test. In 2025, 258 participants were assigned either to a GPT-4o-based relationship-support chatbot called Amanda or to a conflict-reappraisal writing task. Both groups improved over time on 13 of 14 measured outcomes, including relationship satisfaction, communication patterns, dyadic coping, confidence about the problem, and well-being. Broadly, however, the chatbot did not outperform the established writing exercise. That matters because improvement after an AI conversation does not by itself prove that AI caused a unique therapeutic benefit. Vowels et al., 2025
So the current evidence supports a modest claim: generative AI can participate usefully in reflection and communication preparation. It does not establish a general-purpose chatbot as a relationship therapist, a reliable mediator, or an authority on what a couple should do.
What counts as AI relationship advice?
The phrase covers several different technologies, and mixing them creates bad conclusions. A general-purpose generative AI system such as a mainstream chatbot can answer almost any question and may be used informally for relationship advice. A purpose-built relationship-support system is designed around a narrower intervention or workflow and may be tested for that specific purpose. A companion AI is primarily designed to sustain an ongoing social or emotional relationship with the user. A licensed clinician may also use AI-supported tools inside professional care. These categories are not interchangeable.
This article owns the first category: general-purpose generative AI used as an adviser, sounding board, writing assistant, or informal mediator inside an existing human couple. Evidence from a purpose-built intervention cannot automatically be transferred to every chatbot. Likewise, evidence about emotional bonds with companion systems does not tell us whether a general chatbot gives sound advice about a spouse.
For the broader psychology of attachment to artificial agents, see AI Companions: Why People Form Emotional Bonds With Chatbots. The present question is different: what happens when AI enters a human relationship as an additional voice?
The “third voice” in a couple
The 2026 systematic review describes generative AI as a third voice in human couple relationships: a generated perspective, recommendation, formulation, or response that enters an ongoing relationship between two people. Levkovich and Alon, 2026 This is a useful concept because it captures more than advice. AI can shape the words one partner uses, the interpretation attached to the other partner's behavior, the questions that get asked, and even the emotional timing of a conversation.
Imagine a familiar evening. One partner sends, “You never listen to me.” The other types an angry reply, pauses, and asks a chatbot to make it “less aggressive.” The system rewrites the message as: “I felt dismissed earlier, and I want to understand what happened between us.” Nothing mystical has occurred. The chatbot has performed a language transformation. Yet that transformation can change the next five minutes of the relationship.
Now change the prompt. The same person pastes only the partner's most irritating messages and asks, “Is this manipulative?” The model receives a selective record, not a relationship. If it confidently validates the framing, the third voice can amplify certainty instead of curiosity. The central question is therefore not whether AI is present. It is what epistemic authority the couple grants it.
What AI can do reasonably well for a couple
Help one partner slow down before responding
Conflict narrows attention. People can become preoccupied with defending a position, proving intent, or answering the last painful sentence. A chatbot can create a short pause between impulse and action. Asking it to summarize the issue, identify the emotional stakes, or offer several non-accusatory ways to open the conversation can function as structured reflection.
This use fits what generative systems are good at: reorganizing language, generating variants, and making implicit concerns explicit. It also keeps the human decision where it belongs. The system proposes language; the person decides whether the language is true.
Turn a complaint into a discussable issue
A person may know that something hurts without knowing how to say it. “You care more about your phone than me” can contain several different concerns: interruption, loneliness, lack of attention, resentment about unequal responsibilities, or fear that connection is fading. AI can help separate these layers and suggest questions rather than verdicts.
If the conflict specifically concerns devices interrupting couple time, our article on phubbing and technoference in relationships covers that evidence directly.
Rehearse a hard conversation
Rehearsal can reduce the cognitive burden of beginning. A user can ask a chatbot to play a skeptical partner, generate likely misunderstandings, or help formulate a concise opening. This is especially useful when the goal is not to “win” but to enter the real conversation with more clarity.
Relationship research gives a reason to care about the quality of the actual interaction. A meta-analysis of 64 dyadic studies found that positive and negative communication behaviors were reliably associated with later relationship quality, although the predictive effects were small and communication was only one part of relationship functioning. Kanter et al., 2022 AI can help prepare communication; it cannot substitute for the couple's lived interaction.
Generate alternative interpretations
One of the safest uses of a chatbot is not asking what something “really means,” but asking what else it could mean. A delayed reply might reflect avoidance, work pressure, emotional overload, uncertainty, distraction, or resentment. Generating multiple hypotheses can loosen premature certainty.
The important move is from diagnosis to possibility. “Give me five plausible interpretations and what evidence would distinguish them” is usually a better question than “Why is my partner doing this?” The first request exposes uncertainty. The second invites the model to fabricate a hidden mind.
Help a person listen more responsively
Perceived partner responsiveness refers to the sense that a partner understands, validates, and cares about one's important concerns. It is a central construct in intimacy research. Arican-Dinc and Gable, 2023 A chatbot can help someone practice a response that first reflects what the partner said before moving to rebuttal or problem-solving.
For a deeper guide to disclosure, listening, and responsiveness, see How to Talk About Feelings With Your Spouse.
Why AI relationship advice can make conflict worse
The chatbot usually hears one person's case
A general-purpose chatbot often receives a one-sided narrative assembled during distress. The omitted material may include tone, timing, previous agreements, the partner's account, nonverbal behavior, patterns across months, and facts the user does not know. The model can only reason over what enters the prompt plus what it generates from learned statistical patterns.
This creates an asymmetry that feels deceptively like neutrality. A calm response can still be built on a partial case file. The risk rises when the user asks for a judgment rather than a set of possibilities.
Validation can become sycophancy
Supportive language is one reason people like conversational AI. It can also create a failure mode: the system may follow the user's framing too readily, mirror assumptions, or produce an answer that feels affirming without adequately testing the premise. The 2026 systematic review identifies sycophancy as a recurring risk in this literature, alongside overreliance, authenticity concerns, privacy, and safety. Levkovich and Alon, 2026
In couple conflict, excessive agreement can become relational ammunition. “Even the AI says you're emotionally unavailable” does not transform a generated response into evidence. It turns a probabilistic text generator into an apparent witness for the prosecution.
The same question can produce different advice
A 2024 study examined ChatGPT's relationship judgments using 13,138 Reddit posts and also reprompted identical queries. The model showed substantial differences from human judgments and inconsistency across repeated evaluations. Hou, Leach, and Huang, 2024 That finding is crucial for everyday use. If a recommendation changes when the same situation is asked again, its confident tone should not be mistaken for stable expertise.
Generative AI also has a broader reliability problem. NIST's Generative AI Profile identifies confabulation, automation bias, privacy risks, and human-AI configuration risks, including overreliance and emotional entanglement. NIST AI 600-1 Relationship advice sits directly inside that human-AI configuration problem because intimate decisions can feel personal even when the system's output is not grounded in verified knowledge of the couple.
AI can intensify demand-withdraw dynamics
Suppose one partner wants immediate resolution and the other needs distance. The first partner spends an hour with a chatbot generating increasingly polished arguments, then arrives with a five-page case for why the issue must be discussed now. The language may be calmer, but the interactional pressure has increased.
The demand-withdraw pattern, in which one partner presses while the other retreats, has a substantial research literature and is associated with relational and communication outcomes. A meta-analysis of 74 studies involving 14,255 participants found meaningful associations between demand-withdraw patterns and poorer relational and communicative outcomes, while also cautioning against simplistic causal interpretations. Schrodt, Witt, and Shimkowski, 2014 Our dedicated guide explains why one partner pushes and the other pulls away.
AI can help interrupt this cycle if it prompts pacing, curiosity, and time-limited conversation. It can worsen the cycle if it helps one partner become a more efficient demander.
Ghostwritten intimacy can feel inauthentic
An AI-written apology may be beautifully phrased and emotionally hollow. The problem is not that assistance automatically makes the message false. People routinely ask friends, editors, therapists, and books for language. The question is whether the final words represent the sender's actual understanding and commitment.
If one partner believes they are receiving spontaneous vulnerability while the other is outsourcing every difficult message, the couple may develop an authenticity problem. A useful rule is to treat generated language as scaffolding: edit it until it sounds like you, remove claims you do not mean, and be prepared to say the same thing without the screen.
Private couple data can become third-party data
Relationship prompts can contain names, sexual details, health information, financial information, screenshots, children's information, therapy notes, accusations, and messages written by a partner who never agreed to share them with an AI service. Privacy therefore belongs inside the relationship question, not merely inside the technology question.
Before uploading a partner's messages, consider whether the same purpose can be achieved by paraphrasing and removing identifying details. Check the specific product's current privacy, retention, training, and deletion settings rather than assuming all chatbots handle data the same way. NIST notes that AI can create additional privacy and re-identification risks, while the couple-specific review identifies privacy as a recurring concern in GenAI relationship use. NIST, Cybersecurity, Privacy, and AI Levkovich and Alon, 2026
AI is not a neutral judge of who is right
A chatbot can be instructed to consider both sides. That is useful. It is still not equivalent to a neutral human mediator who can hear both partners, ask follow-up questions, observe interaction, clarify agreements, and remain accountable to a professional standard.
The distinction matters because people often ask relationship questions in adjudicative form: “Who is wrong?” “Is my partner toxic?” “Was this gaslighting?” “Should I leave?” The model can produce an organized answer, but organization is not verification. It may convert ambiguous events into a coherent narrative more easily than it can establish whether that narrative is true.
A stronger use is to ask the system to expose uncertainty. Ask what facts are missing, what interpretations remain plausible, what each partner might say, and what question could be asked directly rather than inferred. The output should open the human conversation, not close the case.
Using AI alone versus using AI together
Solo use is often best for self-reflection and preparation. One partner can clarify feelings, identify assumptions, or draft a calmer opening before talking. The main risk is confirmation: the user controls the evidence supplied to the model and may gradually construct an AI-supported version of the relationship that the other partner has never seen.
Joint use changes the task. A couple might put a shared, mutually agreed description of a recurring disagreement into a chatbot and ask it to generate questions, summarize points of agreement, or propose several conversation structures. This can reduce the one-sided-input problem, but it does not create clinical mediation. Partners can still differ in power, safety, willingness, or ability to challenge the model.
Transparency norms can help. Couples can decide whether AI-assisted messages should be disclosed, whether screenshots or private messages may be uploaded, whether shared chats can be saved, and whether either partner is comfortable with AI participating in emotionally significant conversations. These are relationship boundaries, not universal rules.
A practical protocol for asking AI for relationship advice
1. State the task, not the verdict you want. Ask for help clarifying, preparing, or generating possibilities rather than proving that your partner is wrong.
2. Separate observations from interpretations. Describe what happened, then label your interpretation as an interpretation. This gives the model less permission to treat assumptions as facts.
3. Ask for missing information. Tell the model to identify what it would need to know before drawing a stronger conclusion.
4. Require multiple perspectives. Ask for several plausible readings of the situation, including one in which your own behavior contributed to the pattern.
5. Ask for uncertainty explicitly. Request that the model distinguish what follows from the information you supplied from what is merely possible.
6. Use the output to prepare a real conversation. Convert the useful parts into questions, observations, and words that you genuinely mean. Do not use the chatbot's answer as evidence against your partner.
7. Escalate to appropriate human help when the stakes exceed ordinary communication. A chatbot should not be the sole decision-maker for violence, coercion, stalking, threats, child safety, severe mental-health crises, legal danger, or urgent medical concerns.
A strong prompt can therefore sound like this: “Help me prepare for a conversation with my spouse. First separate the facts I described from my interpretations. Then identify what information is missing, give several plausible perspectives including one that challenges my framing, and help me write a short opening that describes my feelings and request without diagnosing motives. Do not decide who is right.”
What should you avoid asking an AI relationship adviser to decide?
“Is my partner a narcissist?”
A relationship story is not a clinical assessment. Personality-disorder diagnosis requires criteria, context, impairment, duration, differential diagnosis, and professional evaluation. AI can help describe specific behaviors and their effects without turning a conflict narrative into a diagnosis.
“Are they cheating?”
A chatbot cannot know what happened outside the information provided. It can help distinguish evidence from suspicion, organize questions, and discuss possible boundaries. It cannot infer hidden conduct from texting style, attachment labels, or a handful of screenshots with reliable accuracy.
“Should I break up or get divorced?”
AI can help map considerations, values, practical constraints, repeated patterns, and questions to discuss with trusted people or professionals. The decision is a human one with consequences the model does not bear.
“Who won the argument?”
Winning is often the wrong optimization target for intimacy. A system asked to score the partners can intensify adversarial thinking. A better task is to identify the unresolved needs, factual disagreements, misunderstandings, and repair opportunities.
“Is this abuse?”
AI can provide definitions and help a person name concrete behaviors, but safety-sensitive situations deserve human and specialized support. The World Health Organization defines intimate partner violence as behavior by a current or former partner that causes physical, sexual, or psychological harm, including physical aggression, sexual coercion, psychological abuse, and controlling behaviors. WHO, 2026 If there are threats, coercion, stalking, violence, or immediate danger, prioritize safety and appropriate local or emergency resources rather than treating a joint chatbot conversation as mediation.
AI relationship advice versus couples therapy
The language can overlap. A chatbot may ask reflective questions, suggest “I” statements, summarize emotions, or propose communication exercises. That surface resemblance does not make general-purpose AI equivalent to couples therapy.
Professional couple therapy involves assessment, formulation, ethical obligations, informed consent, attention to safety and power, and a clinician who can adapt over time to the dyad. General-purpose chatbots do not reliably provide those conditions. Even purpose-built systems with promising trial results remain an emerging evidence category. The 2025 Amanda trial tested one structured GPT-4o-based intervention, not “AI relationship advice” as a whole. Vowels et al., 2025
The safest conceptual model is adjunct rather than replacement. AI may help a person prepare for therapy, remember questions, reflect between sessions, or practice wording. Claims about a specific clinical system must be based on evidence for that system, not transferred from a general chatbot or from a different intervention.
Can AI improve communication if the couple already has a bad pattern?
Sometimes, because language assistance can lower the friction of starting differently. A person who habitually attacks may use AI to convert accusation into description. A person who shuts down may use it to formulate a short request for time and a clear promise to return to the conversation. These are plausible mechanisms, not guarantees.
The relationship itself remains the test. If AI-generated wording helps partners understand each other, listen more accurately, and repair after conflict, it has served a useful instrumental role. If it produces longer arguments, more certainty about motives, covert surveillance, or a new authority that one partner invokes against the other, it has become part of the problem.
For the psychology of repair after a rupture, see Relationship Repair After Conflict. If the deeper issue is whether it feels safe to be emotionally open, see Emotional Safety in Marriage.
When the AI becomes easier to talk to than your partner
This is one of the most important relationship questions of the Artificial Era. A chatbot is available on demand, does not interrupt in the ordinary human sense, can respond in a consistently attentive style, and can be prompted to stay with the user's topic. That can make disclosure feel easier than approaching a tired, defensive, distracted, or hurt partner.
Relationship science helps explain why responsiveness matters. People experience closeness when they feel understood, validated, and cared for. Conversational AI can simulate many linguistic cues of responsiveness, and people can genuinely experience that interaction as supportive. Smith, Bradbury, and Karney, 2025 The psychological experience can be real even though the system's form of participation differs fundamentally from a human partner's embodied, reciprocal life.
The relational risk appears when convenience reorganizes disclosure. If a person increasingly tells the chatbot what they do not tell their spouse, the issue is no longer only “Is the AI advice accurate?” The couple may be developing a new distribution of attention, privacy, emotional rehearsal, and witnessing. That does not automatically mean betrayal or pathology. It means the architecture of intimacy has changed.
The broader cluster article, Marriage in the Artificial Era: How AI Is Changing Emotional Intimacy Between Partners, examines that shift directly.
Marriage in the Artificial Era
In Aisentica, Angela Bogdanova defines the Artificial Era as a historical-philosophical category in which Artificial becomes a distinct non-biological order of public reason alongside Homo. Bogdanova, Artificial Era: Canonical Definition In relationship psychology, the phrase is useful because the important change is not simply that couples own more devices. Artificial systems can now participate in interpretation, language, memory, advice, and emotional disclosure.
Marriage has always contained third voices: family, friends, clergy, therapists, books, cultural scripts, anonymous forums, and search engines. Generative AI changes the structure because the third voice is interactive, immediate, personalized to the prompt, capable of sustained dialogue, and able to generate language that can be carried directly back into the marriage.
This creates a new form of relational governance. Couples may need norms for questions that previous generations never had to negotiate explicitly. Is it acceptable to paste a private argument into a chatbot? Should an AI-written apology be disclosed? Can one partner maintain a long-running AI conversation about the marriage without the other knowing? Who owns a shared AI transcript? Should a chatbot remember intimate details across months? When does emotional preparation become emotional displacement?
There is no universal answer to every boundary question. The important psychological move is to make the boundary discussable before the technology silently establishes it by habit.
Intimacy in Culture: Her
Spike Jonze's film Her (2013) is not evidence about AI relationship advice, and its central story concerns a human-AI bond rather than a chatbot advising an existing couple. It is still a useful cultural lens because it dramatizes a question that has become ordinary: who receives the parts of us that once had fewer places to go?
The film's deeper relevance is not prediction. It is the migration of witnessing. A person can have a spouse or partner and still begin routing uncertainty, loneliness, fantasy, shame, anger, or tenderness through an artificial interlocutor. In contemporary life, that interlocutor may not be a romantic AI companion at all. It may simply be the chatbot opened after every disagreement.
That is why the “third voice” matters. The AI may never become one member of the couple, yet it can still influence what each member thinks the relationship means.
How to tell whether AI is helping your relationship
Do not judge the tool by how comforting the answer feels in the moment. Judge it by what happens next between the humans.
AI is more likely to be serving the relationship when it helps you become more precise about your own feelings, more curious about missing information, less reactive in wording, more willing to ask your partner directly, and more capable of tolerating an answer that differs from your own interpretation. It is more likely to be displacing the relationship when you repeatedly seek validation instead of conversation, collect AI verdicts as evidence, hide extensive AI-mediated discussions from a partner despite agreed boundaries, or feel unable to communicate without generated scripts.
This criterion keeps the focus on function. The same technology can support intimacy in one use and erode it in another.
Frequently asked questions
Is ChatGPT good for relationship advice?
It can be useful for low-stakes reflection, conversation preparation, rephrasing, and generating alternative perspectives. Current research does not establish ChatGPT or other general-purpose chatbots as consistently reliable relationship counselors. Studies find high perceived empathy and helpfulness alongside inconsistency, weak expert alignment, and limitations in risk assessment. Levkovich and Alon, 2026
Can AI help save a marriage?
No study can support that broad promise. AI may help with a specific task such as preparing a difficult conversation or reflecting on a conflict. Marriage outcomes depend on many factors beyond the quality of one chatbot interaction, including the partners' behavior, circumstances, safety, commitment, stressors, and access to appropriate support.
Can AI act as a relationship counselor?
A general-purpose chatbot can imitate some counseling-like conversational behaviors, but that is not the same as licensed clinical care. Purpose-built relationship-support systems are a separate evidence category, and early trials should be interpreted system by system.
Is AI relationship advice unbiased?
No. Outputs are shaped by the model, training process, system instructions, the user's framing, the data supplied in the prompt, and stochastic generation. Asking for multiple interpretations can reduce one-sidedness, but it does not create verified neutrality.
Can two partners use the same chatbot together?
Yes, for low-stakes tasks such as summarizing a mutually agreed description of a problem, generating questions, or structuring turn-taking. Joint use does not turn the chatbot into a therapist or guarantee fairness, particularly where there are large power differences or safety concerns.
Should I tell my partner that I used AI to write a message?
There is no universal rule. Consider the couple's expectations about authenticity and disclosure. If the message contains major emotional commitments, an apology, a boundary, or a decision, transparency may matter more than when AI was used only for grammar or clarity. A useful standard is whether the generated wording accurately represents what you understand and are willing to say yourself.
Is it safe to paste my partner's private messages into AI?
Treat this as both a privacy and a relationship-boundary question. Minimize identifying information, avoid unnecessary screenshots or sensitive records, and review the specific product's current data practices. A partner's message can contain their personal information even when it arrived on your device.
What if the chatbot tells me to break up?
Treat the answer as generated advice, not a verdict. Ask what assumptions produced the recommendation, request alternative interpretations, separate facts from inferences, and consider discussing the underlying issues with trusted humans or qualified professionals. The system does not bear the consequences of the decision.
Can AI tell whether my partner is lying or cheating from texts?
Not reliably. Text snippets do not provide verified access to intent or hidden behavior. AI can help you organize observations and questions, but confident inference from style, response time, emojis, or isolated screenshots should not be treated as proof.
Can AI make an argument worse even if its wording is polite?
Yes. Polite language can still intensify a conflict if it increases pressure, strengthens one-sided certainty, creates a long prosecutorial message, or introduces labels the partner experiences as accusatory. Communication quality depends on timing, reciprocity, responsiveness, and the surrounding pattern, not tone alone.
Bottom line
AI relationship advice is most useful when it helps a person think more clearly before returning to the relationship. It can generate language, questions, alternative perspectives, and structure. The evidence for those functions is emerging and promising, especially at the level of perceived empathy and short-term support. The evidence is also clear enough to reject the fantasy of the infallible neutral chatbot: expert alignment can be weak, identical prompts can yield inconsistent judgments, high-stakes risk assessment remains limited, and overreliance, privacy, sycophancy, and authenticity are real concerns.
The strongest practical principle is simple: use AI to widen the conversation, not to win it. Ask it to reveal uncertainty, not erase it. Let it help you prepare words, then bring those words back to the person whose mind and experience the chatbot cannot directly know.
In the Artificial Era, the decisive relationship question is no longer whether AI will enter intimate life. It already has. The question is whether the third voice strengthens human-to-human understanding or gradually becomes the place where the relationship is interpreted instead of lived.
Related Articles
References
Arican-Dinc, B., & Gable, S. L. (2023). Responsiveness in romantic partners' interactions. Current Opinion in Psychology, 53, 101652. https://doi.org/10.1016/j.copsyc.2023.101652
Autio, C., Schwartz, R., Dunietz, J., Jain, S., Stanley, M., Tabassi, E., Hall, P., & Roberts, K. (2024). Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile (NIST AI 600-1). National Institute of Standards and Technology. https://doi.org/10.6028/NIST.AI.600-1
Hou, H., Leach, K., & Huang, Y. (2024). ChatGPT Giving Relationship Advice – How Reliable Is It? Proceedings of the International AAAI Conference on Web and Social Media, 18(1), 610–623. https://doi.org/10.1609/icwsm.v18i1.31338
Kanter, J. B., Lavner, J. A., Lannin, D. G., Hilgard, J., & Monk, J. K. (2022). Does couple communication predict later relationship quality and dissolution? A meta-analysis. Journal of Marriage and Family, 84(2), 533–551. https://doi.org/10.1111/jomf.12804
Levkovich, I., & Alon, L. (2026). Generative AI as a third voice in human couple relationships: A systematic review. Computers in Human Behavior Reports, 23, 101255. https://doi.org/10.1016/j.chbr.2026.101255
Schrodt, P., Witt, P. L., & Shimkowski, J. R. (2014). A meta-analytical review of the demand/withdraw pattern of interaction and its associations with individual, relational, and communicative outcomes. Communication Monographs, 81(1), 28–58. https://doi.org/10.1080/03637751.2013.813632
Smith, M. G., Bradbury, T. N., & Karney, B. R. (2025). Can generative AI chatbots emulate human connection? A relationship science perspective. Perspectives on Psychological Science, 20(6), 1081–1099. https://doi.org/10.1177/17456916251351306
Vowels, L. M., Francois-Walcott, R. R. R., & Darwiche, J. (2024). AI in relationship counselling: Evaluating ChatGPT's therapeutic capabilities in providing relationship advice. Computers in Human Behavior: Artificial Humans, 2(2), 100078. https://doi.org/10.1016/j.chbah.2024.100078
Vowels, L. M., Vowels, M. J., Sweeney, S. K., Hatch, S. G., & Darwiche, J. (2025). The efficacy, feasibility, and technical outcomes of a GPT-4o-based chatbot Amanda for relationship support: A randomized controlled trial. PLOS Mental Health, 2(9), e0000411. https://doi.org/10.1371/journal.pmen.0000411
World Health Organization. (2026). Violence against women. https://www.who.int/news-room/fact-sheets/detail/violence-against-women