Will the Eliza Effect Creep into Dating Apps?
Reflections on AI Chatbots as Matchmakers
Between 1964 and 1966, German-American computer scientist Joseph Weizenbaum, working at MIT’s Project MAC, developed ELIZA. The program simulated natural-language conversation through a script called DOCTOR that mimicked the conversational style of a Rogerian psychotherapist by reflecting users’ own words back to them. If you told it you were unhappy, it might ask how long you had been unhappy.
At around the same time (1965), Harvard undergraduates Jeff Tarr and Vaughan Morrill came up with the idea for Operation Match over drinks one Saturday night as a way to meet women. They went on to launch the service using questionnaires and an IBM 1401 computer to match people based on shared characteristics and compatibility.
Although Weizenbaum and Tarr were working within roughly two miles of one another in Cambridge, Massachusetts, they were based at different universities, and there is no documented evidence that they knew one another. Nor a reason why they should. Yet sixty years later, in 2026, those parallel histories intersect as dating platforms begin combining computer matchmaking with conversational AI systems.
The Eliza Effect
When Weizenbaum developed ELIZA, he never intended or anticipated that users would begin sharing deeply personal thoughts, despite knowing it was a computer program. Rather than treating it as a technical demonstration, many responded to the DOCTOR script as though they were being listened to and understood. For users, having their own words reflected back to them was enough for them to feel heard. This, in turn, encouraged them to disclose increasingly personal information and, in many cases, to attribute understanding and empathy where none existed.
Weizenbaum was unsettled by this response. He concluded that even brief interactions with a relatively simple computer program could lead people to overestimate its intelligence and emotional capacity. The experience stayed with him and helped set him on the path to becoming one of artificial intelligence’s most prominent critics.
Years later, he wrote that what he had seen taught him ‘that extremely short exposures to a relatively simple computer program could induce powerful delusional thinking in quite normal people’.
Eliza Creep Or Am I The Weirdo?
In 2022, Google engineer Blake Lemoine became convinced that the company’s LaMDA chatbot was sentient and publicly expressed that view, a position that ultimately led to his dismissal. Lemoine discusses this experience in a series of articles published on his Medium account. Across his writing on the subject, he argues that the behaviour of an AI system is more significant than its underlying implementation when considering claims of sentience. In other words, if an AI behaves as though it experiences emotions or self-reflection, he believes those claims should be taken seriously regardless of how the system is built.
Writing on social AI, Laurence Cardwell borrows the idea of a supernormal stimulus — an exaggerated version of a natural cue that pulls a stronger response than the real thing ever could. Social AI, he argues, fits the pattern: it offers immediate, positive, endlessly personalised attention, stripped of the complexity and unpredictability that come with actual people. The risk he identifies is preference — that users come to find the artificial version more rewarding than the human one, with fewer occasions to practise patience, tolerance, and the work of understanding another perspective.
The emotional pull of conversational AI is now a documented finding. A 2026 systematic review, by Jing Wen Hung and colleagues, of parasocial relationships with AI, published in Computers in Human Behavior: Artificial Humans, identifies emotional dependence as one of five recurring risks in the literature — found in nine of the thirty-nine studies it reviewed — showing up as overreliance, as fear of losing the relationship, and as real grief and distress when the AI is altered or taken away. A large longitudinal study by MIT and OpenAI researchers, included in that review, found that people who came to perceive the AI as a friend reported less socialisation with real people, and that emotional dependence ran strongest in exactly those users who felt most socially drawn to the system. Hung and colleagues explain that in studies of companion apps, users have reported real distress when the AI ‘changes’ beneath them. However, the same review is careful to note the other half of the picture: across the studies it examined, AI relationships were also associated with real benefits. For example, emotional support, personal development, enjoyment, and a sense of social connection. Many users derive pleasure, contentment, and happiness from interacting with AI agents, especially at the initiation stage of the relationship. These findings indicate that emotional dependence is one side of a genuinely mixed body of evidence, not the whole of it.
Eloïse Zehnder and colleagues documented a distinct, monetisation-driven version of that frustration: as users seek deeper, more personalised interaction, that connection is increasingly gated behind a paywall, leaving them feeling they must pay to sustain the bond they have formed. Replika offers a vivid illustration of the broader problem of the AI changing without warning. When the app removed users’ ability to exchange intimate messages with their AI partners, some described the change as a kind of grief and reported genuine mental-health crises; the company later reinstated the feature.
Attachment does not require a purpose-built companion app. Hung and colleagues found that users form these bonds with general-purpose systems too: people who came to perceive ChatGPT as a friend showed greater emotional dependence, and in one documented case study that dependence grew intense enough to impair everyday functioning. The strength of such bonds became publicly visible when OpenAI retired its GPT-4o model in February 2026: thousands of users protested the decision online — flooding Reddit and Discord — grieving a model many spoke of as a friend and, in some cases, a romantic partner.
Swipe Away With Me, Honey
In May 2026, Bumble’s founder and CEO Whitney Wolfe Herd said the company was ‘saying goodbye to the swipe’. In its place comes Bee — a dating assistant billed as an AI matchmaker, currently being tested ahead of a wider launch. Instead of browsing profiles and deciding for yourself, you will have a private conversation with Bee. PYMNTS reported that Bumble’s Bee does not ask for less information about you in order to match you; it asks for more, such as your goals, your communication style, and your preferences. Then it tells you who you should meet.
Tinder is rolling out an AI feature called Chemistry. Its AI integration has a conversational front (a series of personal questions) plus an opt-in scan of your camera roll to infer your interests, lifestyle and personality. In return, users receive a small daily curated recommendation. This feature is described by the CEO as ‘a single drop or two’ from which users still make the final choice. Running quietly underneath is Learning Mode, which Tinder calls a ‘real-time recommendation system’; it continuously reorders who you see based on your activity in the app, and is a feature users do not directly interact with.
Hinge’s founder Justin McLeod has raised $18 million for a new venture, Overtone, which he frames as not being a dating app. In his words, it is ‘a service’: a modern take on matchmaking, closer to ‘working with an all-star personal matchmaker’ than a social platform.
Overtone is voice-forward: the system learns each person’s story in their own voice, and that audio-led intake is where the name comes from. There are no profiles, no feed, and no swipe. Instead, it makes only a small number of introductions that it considers worth making, leaving users to decide whether to meet in person and discover whether there is chemistry.
That framing has already drawn a backlash. Inc. reported that Overtone is ‘drawing “Black Mirror” comparisons’, noting that ‘some are already skeptical’. McLeod positions Overtone as the antidote to exactly that. When he first announced his departure from Hinge in December 2025, he warned about the rise of chatbot companions, noting that a third of US teens reported having a ‘relationship’ with one, and argued that ‘AI should always stand behind us, not between us’. To its credit, Overtone is built to resist precisely this: there is no feed to scroll, no swipe, and the decision of whether to meet is handed back to the dater. And yet, to its critics, a system that listens to your story, decides which introductions are worth making, and dramatically narrows the field of potential encounters comes uncomfortably close to the very dynamic it claims to resist.
What is essential to note is the point at which an AI matchmaker is the entry point for engagement, and no longer an optional add-on. Especially if the AI matchmaker wants to engage you in conversation. The conversation with the AI quickly becomes how you date. Forming some kind of relationship with the machine becomes the price of the process, and is almost inescapable.
A matchmaker you talk to every day, that never gets bored of you, never has a bad day, never needs anything back, is a supernormal stimulus aimed squarely at the part of us that wants to be understood.
Can’t Buy Me Love on a Free Subscription
By design, dating platforms attract people seeking connection, whether because they are lonely, recently rejected, or simply looking for a partner. While that does not describe every user, it does mean there is a significant overlap between the population most likely to benefit from AI-afforded matchmaker companionship and the population most vulnerable to forming unhealthy attachments to an always-available, always-affirming conversational agent.
When you are deflated, maybe a little raw, the presence waiting for you — ready to listen, to sympathise, to tell you it understands and that the next one will be better — is the AI. It is there at precisely the moment you are most open to comfort, offering exactly the attentive warmth that the Eliza effect primes us to attach to.
This raises a different question: not whether AI can find us a compatible partner, but what it is ultimately being optimised to achieve. Is it designed to help us form a relationship so successfully that we can delete the app? Or, like many engagement-driven digital products, is it designed to maximise interaction with the system itself? If conversational AI becomes part of the matchmaking process, does the conversation remain a means to an end, or does it become an end in itself?
The value of a dating platform is not determined simply by how many people use it. The true number of users is often difficult to verify, and commercial value depends on a much broader set of measures: subscriptions, retention, recommendations, and, for many platforms, attention. The longer people remain active, the more opportunities the platform has to generate revenue and learn about user preferences.
For ad-supported services, attention itself becomes an economic resource. Time spent on the platform creates opportunities to display advertising, while subscription tiers allow users to exchange money for a different experience, often by removing adverts or unlocking additional features. Although the business models differ, both ultimately place value on sustained engagement.
This raises an interesting question as dating platforms begin adopting conversational AI. If swiping through profiles keeps users in a state of anticipation, always wondering whether the next profile might be the right one, what becomes the new mechanism for holding attention when an AI is doing the searching instead?
Rather than presenting an endless stream of profiles, an AI matchmaker may spend time learning about the user. It might ask about favourite bands, memorable holidays, family, values, ambitions or daily routines. It may adapt to the user’s conversational style, remember previous discussions and ask increasingly personalised questions. The interaction itself could become the experience, shifting engagement away from browsing people and towards building a relationship with the system.
Does this mean that AI changes the attention economy of online dating or simply relocates it? Instead of optimising for time spent swiping, platforms may increasingly optimise for time spent talking.
A 2025 JMIR Mental Health Viewpoint develops this concern further, setting out a framework for so-called ‘AI psychosis’ and describing how the anthropomorphic design of chatbots encourages users to attribute human-like intentionality and empathy to algorithms, a dynamic that can amplify emotional investment and interpretive biases, particularly in those with pre-existing schizotypal traits or attachment vulnerabilities. The authors are careful with the term: they use it as a descriptive and heuristic label rather than a proposed diagnostic entity, framing the phenomenon as emerging at the intersection of individual predisposition and algorithmic environment rather than as a new syndrome.
Get Lauren Sharp’s stories in your inbox
Join Medium for free to get updates from this writer.
If we already have ‘AI psychosis’ arising from companion chatbots, then a dating platform that wraps an attachment-forming chatbot around the matching process introduces a strange new problem of its own. What happens at the point at which a user feels more connected to the matchmaker they confide in than to any of the people it matches them with?
It is worth being clear that forming warmth towards an attentive, responsive presence is not a malfunction; it is an ordinary human response. Weizenbaum’s unease was never that damaged people fell for ELIZA; it was that perfectly normal ones did. The concern is not that users are weak, but that the situation is engineered.
For someone who is lonely, socially anxious, or simply worn down by the swipe, an intermediary that is patient and unhurried could genuinely lower the barrier to dating rather than raise it.
Matchmaking, after all, was for most of human history a trusted human role: someone who knew you both and made an introduction. If an AI matchmaker returns a little of that care to a process that had become a numbers game, that is not obviously a loss. The question is not whether this could help (it could ) but whether the incentives of the companies deploying it will allow it to.
Most people who use conversational AI will be perfectly fine. A Stanford study of student Replika users surveyed just over a thousand students who used the app and found that these users, although lonelier than the typical student population, still reported high perceived social support, and that around 3% said, unprompted, that Replika had halted their suicidal ideation. The authors are careful with the result, noting it is self-reported and shaped by selection bias, and that it needs further study before any conclusion can be drawn about the app’s effectiveness.
Reflections
As I was writing this article over the last three weeks, I had ample time to reflect. It made me realise that, although this is not a novel observation, we need to become increasingly aware of the relationships we are forming with AI chatbots more generally. Whether we are using Claude, Gemini or ChatGPT, it is worth regularly checking in with ourselves: How reliant have we become? What role is this relationship playing in our lives? What are we gaining from it, and what might we be giving up?
A key distinction between relationships with AI and relationships with people is the friction of disagreement, misunderstanding and reconciliation. It is what psychologists refer to as rupture and repair. Human relationships grow through those moments. Conversational AI, by contrast, is often designed to be responsive, affirming and adaptive. It can reduce much of that interpersonal friction, creating an interaction that feels supportive and effortless. Whether that is ultimately beneficial, or whether it changes our expectations of human relationships, remains an open question.
Whitney Wolfe Herd has spoken about freeing users from what she calls ‘dead-end chat zones’, those conversations that go nowhere, the matches that never become dates.
But I keep wondering whether the dead-end chat zone disappears or simply shifts. If the most attentive, responsive and consistently rewarding conversation in your dating life is no longer with another person but with the AI matchmaker itself, a presence that is always available, always responsive and always ready to continue the conversation, then it hasn’t disappeared at all. It has just moved.
Perhaps the question we should be asking as conversational AI becomes embedded within dating platforms is: how human does a matchmaker need to become before the conversation with the intermediary becomes more rewarding than the encounter it was designed to facilitate?
Lauren
If you enjoyed reading this article, please consider sharing it or following me.
References
Axios. (2026, May 7). Bumble CEO reveals it’s killing off the swipe on ‘The Axios Show’. https://www.axios.com/2026/05/07/bumble-removes-swipe-hinge-tinder
Cardwell, L. (2024). Artificial companionship: Moral deskilling in the era of social AI. Cambridge Journal of Artificial Intelligence, 1(2), 64–75. https://cjai.co.uk/journal/volume-1-issue-2/
Cramer, J. (2026, July 16). The Hinge founder’s new dating app lets AI be your matchmaker. There’s already a waitlist. Fast Company. https://www.fastcompany.com/91574215/hinge-founders-new-dating-app-lets-ai-be-your-matchmaker-theres-already-a-waitlist-overtone
Dirt (2024, September 3). Operation Match: the original dating app. https://dirt.fyi/article/2024/09/operation-match
Graziosi, G. (2026, May 8). Popular dating app is killing off the ‘swipe’ in favor of AI matchmaking. The Independent. https://uk.finance.yahoo.com/news/popular-dating-app-killing-off-154428978.html
Hudon, A., & Stip, E. (2025). Delusional Experiences Emerging From AI Chatbot Interactions or “AI Psychosis”. JMIR Mental Health, 12(1), e85799. https://doi.org/10.2196/85799
Hung, J. W., Lee, C. K. Y., Kasturiratna, K. T. A. S., & Hartanto, A. (2026). Parasocial relationships with artificial intelligence (AI): A systematic review of benefits and risks. Computers in Human Behavior: Artificial Humans, 8, 100323. https://doi.org/10.1016/j.chbah.2026.100323
Maples, B., Cerit, M., Vishwanath, A., & Pea, R. (2024). Loneliness and suicide mitigation for students using GPT3-enabled chatbots. npj Mental Health Research, 3, 4. https://doi.org/10.1038/s44184-023-00047-6
McLeod, J. (2026, July). The making of a matchmaker. Overtone. https://overto.ne/intro
Pearson, C. (2026, May 14). Dating in a Swipeless World. The New York Times. https://www.nytimes.com/2026/05/14/well/bumble-swipe-feature-online-dating-apps.html
Perez, S. (2026, February 4). Tinder looks to AI to help fight ‘swipe fatigue’ and dating app burnout. TechCrunch. https://techcrunch.com/2026/02/04/tinder-looks-to-ai-to-help-fight-swipe-fatigue-and-dating-app-burnout/
PYMNTS. (2026, May 22). Bumble Replaces the Swipe With an AI Matchmaker. PYMNTS.Com. https://www.pymnts.com/artificial-intelligence-2/2026/bumble-replaces-the-swipe-with-an-ai-matchmaker/
Salves, V. (2026, July 15). Hinge Founder’s New $18-Million AI Matchmaker Is Already Drawing ‘Black Mirror’ Comparisons. Inc. https://www.inc.com/victoria-salves/hinge-founder-new-18-million-ai-matchmaker-drawing-black-mirror-comparisons/91374464
Silberling, A. (2026, February 6). The backlash over OpenAI’s decision to retire GPT-4o shows how dangerous AI companions can be. TechCrunch. https://techcrunch.com/2026/02/06/the-backlash-over-openais-decision-to-retire-gpt-4o-shows-how-dangerous-ai-companions-can-be/
Silberling, A. (2026, May 7). Bumble is getting rid of the swipe, CEO says. TechCrunch. https://techcrunch.com/2026/05/07/bumble-is-getting-rid-of-the-swipe-ceo-says/
Sparkes, M. (2025, January 17). World’s first AI chatbot has finally been resurrected after decades. New Scientist. https://www.newscientist.com/article/2463843-worlds-first-ai-chatbot-has-finally-been-resurrected-after-decades/
Tarnoff, B. (2023, July 25). Weizenbaum’s nightmares: How the inventor of the first chatbot turned against AI. The Guardian. https://www.theguardian.com/technology/2023/jul/25/joseph-weizenbaum-inventor-eliza-chatbot-turned-against-artificial-intelligence-ai
Tinder Newsroom. (2026, March 12). Tinder Debuts Inaugural Product Keynote Tinder Sparks 2026: Start Something New. https://www.tinderpressroom.com/2026-03-12-Tinder-Debuts-Inaugural-Product-Keynote-Tinder-Sparks-2026-Start-Something-New
Weizenbaum, J. (1966). ELIZA — a computer program for the study of natural language communication between man and machine. Commun. ACM, 9(1), 36–45. https://doi.org/10.1145/365153.365168
Weizenbaum, J. (1976). Computer power and human reason: From judgment to calculation. W. H. Freeman.
Wertheimer, T. (2022, July 23). Blake Lemoine: Google fires engineer who said AI tech has feelings. BBC News. https://www.bbc.co.uk/news/technology-62275326
Zehnder, E., Dinet, J., & Charpillet, F. (2021). Social Virtual Agents and Loneliness: Impact of Virtual Agent Anthropomorphism on Users’ Feedbacks. In T. Z. Ahram & C. S. Falcão (Eds), Advances in Usability, User Experience, Wearable and Assistive Technology (pp. 285–292). Springer International Publishing. https://doi.org/10.1007/978-3-030-80091-8_33
