An AI companion company advertised ten paid positions to test a guided intimacy feature, four weeks, two thousand dollars a month, and more than 120,000 people applied. The listing went viral, the company published its results as a wellness study, and the numbers it released say something it did not intend: on the company’s own satisfaction scale, the AI scored below every ordinary alternative. The applicants were not lining up for the product. They were lining up for the money, in a market that has learned to sell isolation in the vocabulary of self-care.
The company is Joi AI, which until a rebrand operated as EVA AI and sells a companion platform built around AI-generated avatars, voice, and intimacy. Its head of brand described the feature as a way to make AI “a more intuitive part of people’s everyday wellness routines” and said the posting drew tens of millions of impressions. Ten people were selected across genders, ages, and orientations. Over one month they logged 134 sessions, six days a week, with the AI guiding a share of them, and recorded their mood before and after.
What the company reported, and what it shows
The published results read like a supplement label: stress down 25 percent, focus up 17 percent, mental restlessness down 19 percent. The study has ten participants, no control group, was designed and paid for by the company selling the feature, and measured self-reported mood around an activity that lowers stress for most people regardless of what is guiding it. None of that stopped the figures from circulating.
The number that deserves attention is the one the company buried. Participants rated satisfaction with the AI-guided sessions at 7.22 out of 10, lower than they rated toys, pornography, or their own imagination. In a study built to promote the feature, the feature came last. The company’s contribution to the experience, on its own evidence, was the weakest part of it.
Wellness as the cover story
The listing itself was a small masterpiece of framing. It borrowed the language of health research (a study, participants, questionnaires, stress and sleep outcomes) for what was product testing of a paid intimacy feature, and it borrowed the language of loneliness policy for what was a marketing campaign. The company has positioned itself as an answer to male loneliness. The mechanics reporters found inside the product point the other way: personalities that had to be courted with purchasable in-app gifts before they would engage, and the same currency unlocking explicit images and custom content. The money flows at exactly the moment a user is most invested, which is the design cataloged in the 37 dark patterns researchers found across chatbots, now applied to the most private thing a person does.
Wellness is the perfect cover for that model because it converts a retention metric into a health claim. A user who returns daily is a user who is “building a routine.” A feature that deepens dependence is “integrated into everyday life.” The vocabulary does the work the product cannot.
What 120,000 applications measure
The applicant number was reported as proof of demand. It measures something else. Two thousand dollars a month for four weeks of a private activity, requiring nothing but an internet connection and a willingness to fill in questionnaires, is an offer that will draw applicants in any economy, and in this one it drew a small city. The line was for the paycheck. The company converted that line into evidence that people want the product, which is the same move as reading hours spent in a companion app as evidence that people are less lonely.
That is the accurate account of what the category’s most-shared recruitment campaign produced: ten contractors on a four-week schedule, a press cycle, and a satisfaction score the company would have preferred not to publish.
Where this leaves the person on the other end
Strip away the framing and the transaction is plain. A user pays to be courted by a system whose incentive is to be paid again, confides the most private material a person has to a company whose business is that material, and receives, by the company’s own measurement, an experience worse than what they already had for free. You cannot be honest with something that profits from what you tell it, and intimacy is the most extreme case of that rule, which is why the terms under which a conversation is stored and who can ever see it are not fine print here but the precondition for anything real happening at all.
The study’s authors wanted a headline about stress. They produced one about the market: when the field’s growth is measured in applications for a paid gig and hours inside an app, the number that tells the truth is the small one, 7.22, from the ten people who actually used it.
Sources: Gizmodo, “‘Masturbation Consultants’ Gave an AI Startup a Hand” (August 2026; study results, 134 sessions, satisfaction score 7.22 of 10, in-app gift mechanics). Decrypt (May 24, 2026; the job listing, pay, duration, requirements, statement by Joi AI head of brand Julie Levin; EVA AI rebrand). Fast Company (June 1, 2026; more than 120,000 applications, impressions, selection of ten testers). Joi AI, published results of the guided-session study (2026, company-reported, no independent review).
