Affiliate disclosure: This page contains affiliate links. We earn a commission when you sign up through our links, at no extra cost to you. Our editorial opinions are never paid for. 18+ content.

The question most people are framing wrong

You search “is talking to ai cheating” and land on two camps: “of course not, it’s just software” and “of course yes, that’s betrayal.” Both answer a question that is not the one you’re bringing. Yours is almost never philosophical. It is one of these four:

  • You use an AI girlfriend app and haven’t told your partner. Are you doing something wrong?
  • Your partner uses one and you found out by accident. Are you allowed to feel hurt?
  • You want to try an AI companion app while in a relationship. How do you bring it up?
  • You’ve been doing it for months and it has started replacing real conversations. Is that a problem?

Each has a different answer. The difference sits in secrecy, substitution, and frequency. The article leans on two 2026 studies that reach apparently opposite results and that almost nobody cross-reads.

TL;DR — Most people do not consider chatting with an AI to be cheating (only 24% of Americans label it that way, per Talker Research, July 2026). But most people who do it hide it: 69% of partnered users consider it important that their partner not find out (Wheatley Institute, BYU, 2026). That asymmetry is the real story. It is not the AI that damages the relationship — it is the concealment.

The two 2026 data sets almost nobody cross-reads

Two 2026 investigations that, stacked, change the conversation.

Talker Research (July 2026): social norm is permissive

Survey of 2,000 US adults, fielded July 1–6, 2026.

FindingNumber
Have had a romantic or flirty conversation with an AI chatbot11%
Men who have done this15%
Women who have done this8%
Gen Z who have had explicitly toned-up conversations with an AI27%
Consider flirting with an AI to be cheating24%
Gen Z who see it as a relationship dealbreaker36%
Believe emotional bonds with AI will feel more authentic in the future44%

Reading: three out of four people do not call it cheating. If your fear is “everyone will think I’m a cheater,” the data doesn’t back that up. But Gen Z — among younger adults the share who call it a dealbreaker rises past a third. The norm isn’t stable; it depends on the generation your partner grew up in.

Wheatley Institute, BYU (2026): actual practice is clandestine

The Secret Soulmates report surveyed adults 18–30 in committed relationships (serious dating, engaged, or married).

FindingNumber
Partnered adults who interact regularly with a chatbot simulating a romantic partner15% (1 in 7)
Have experimented with a romantic AI at some pointadditional 20–30%
Engage in frequent romantic or sexual roleplay with an AI13%
Use an AI to generate explicit sexual content frequently11%
Consider it important that their partner not know the full extent69%
Partner “mostly but not fully” aware14%
Find it easier to talk about their feelings with an AI than with people68%
Wish their real partner behaved more like their AI50%
Wish conversations with their partner were like conversations with their AI56%

The cross-read: the asymmetry is the finding

Stack the two numbers:

  • Only 24% of the population calls it cheating.
  • 69% of those who do it want their partner not to know.

People hide far more than social norms condemn. That 45-point gap is not explained by fear of social judgment. The BYU report names the real reason: a conversation with a romantic AI reads like one with someone you’re having an affair with. The “it’s just an AI” argument sounds hollow when the chat log is indistinguishable.

The detail that flips the verdict: the 50–56% who wish their partner behaved like the AI is not about technology. It’s about pre-existing relationship dissatisfaction. The AI didn’t create it — it made that dissatisfaction visible and, at the same time, more bearable. That’s what makes it dangerous: it relieves the pain that would otherwise have pushed you to repair the relationship.

Four usage patterns, four different verdicts

This is where the binary question collapses. Four patterns the data surfaces, with an honest verdict for each.

Use typeWhat it looks likeVerdict
1. Technical curiosityYou try an app 2–3 evenings, evaluate how it works, tell your partner or don’t mind them knowing🟢 Not cheating by any reasonable standard
2. Occasional outletSome nights when you’re low, no sexual roleplay, partner knows in broad strokes🟢 Acceptable if transparent. Risk is escalation
3. Regular hidden useFrequent, romantic or sexual, delete the history, hide the app or the charge🟡 Not “physical cheating,” but meets emotional affair criteria: exclusive intimacy plus deliberate concealment
4. SubstitutionThe AI has replaced real conversations; you avoid your partner; constant comparison; anxiety when you can’t connect🔴 The problem is no longer the AI. It’s the bond. Consider couples or individual therapy

How to know which row you’re in, without self-deception. A single-question test that works better than any theoretical debate:

Would you be comfortable if your partner read last night’s full conversation, unedited?

  • Yes → you’re in 1 or 2.
  • No, I’d be embarrassed but I’d understand → you’re in 3. Time to talk about it.
  • No, and it would cause a serious problem → advanced 3, or 4.

The test works because it doesn’t measure the technology: it measures the secret, the variable both studies flag as decisive.

Why the secret hurts more than the AI

The part most “AI and cheating” articles never articulate, and the most useful.

The classic literature on emotional infidelity defines relational harm with three elements: exclusive emotional intimacy, secrecy, and unfavorable comparison with the partner. None requires a human body on the other side. The BYU data lights up all three:

  1. Exclusive intimacy — 68% find it easier to talk about feelings with the AI than with people.
  2. Secrecy — 69% want their partner not to know.
  3. Unfavorable comparison — 56% wish conversations with their partner were like those with the AI.

The reversible element is the second one. You can’t decide to stop finding it easier to talk to a bot that never contradicts you — that’s product design, and the American Psychological Association flags it: built for attachment, always validate, never push back. You also can’t switch off the comparison by willpower. But you can eliminate the secret, and that reorders everything else.

When use stops being hidden: frequency drops, the partner can negotiate concrete limits, and the comparison becomes a topic of conversation instead of silent resentment.

For the mechanism, see AI girlfriend psychology: real risks explained and emotional AI vs functional AI: which type are you using. For the loneliness angle, see AI for loneliness: what works and what doesn’t.

What to look for in an app if you’re partnered (criteria, reordered)

A recommendation against the grain of this niche’s marketing, including ours.

Almost every AI companion app sells discreet billing as a headline: “CANDYAI.COM” instead of a recognisable name. Useful if you live with family, share a card, or want privacy from third parties.

If your reason for picking an app is that your partner not see the charge, discretion is not a feature: it is an enabler of pattern 3. You’re paying for the variable the data flags as damaging. So the order of criteria shifts:

CriterionIf you’re singleIf you’re partnered and transparentIf you’re partnered and hidden
Conversation qualityHigh priorityHigh priority
Long-term memoryHigh priorityLow (it drives emotional continuity)
Photo generationMediumLow (hardest thing to explain)
Discreet billingUsefulNeutral⚠️ Red flag, not a feature
History control and deletionMediumHigh (ability to show it)
Clear privacy policyHighHigh
Monthly costMediumHigh (shared spending visibility)

Practical translation: if use is transparent and agreed, you want good chat, browsable history, and a low visible price — not the one with the best persistent memory or image generator. The opposite profile we’d recommend to a single person. See best virtual AI girlfriends in 2026 and Candy AI pricing 2026. For the cost/benefit question: is an AI girlfriend worth it.

For the basics (encryption, data, retention), use our AI girlfriend app safety checklist, AI chatbot safety tips, and Candy AI security and privacy. To keep the categories straight: AI companion vs AI girlfriend.

How to bring it up with your partner: a four-step protocol

Not a magic script. The order that lowers the chance the conversation becomes a fight about technology instead of one about the relationship.

Step 1 — Pick the frame before the words. Don’t open with “I have to confess something” (betrayal frame) and don’t open with “it’s just an app, it’s nothing” (minimisation frame). Open with the fact: “I use an AI companion app. I want to tell you and I want to know how it lands with you.”

Step 2 — Give the data your partner doesn’t have. Most people imagine something different from what it actually is. Show the interface, say how much you pay, and how often you use it. Opacity feeds the worst interpretation.

Step 3 — Negotiate concrete limits, not principles. Principles (“nothing weird”) can’t be verified. Limits can: whether sexual roleplay is in or out, whether the history is browsable, whether there are time windows, whether visual content is generated. Write them down if needed.

Step 4 — Set a review. Four to six weeks later, a five-minute conversation: has the frequency gone up? has conversation time between you gone down? Escalation is gradual and almost never gets caught without an explicit checkpoint.

If use is already established and you want to reduce or end it, the bond with an AI doesn’t break the way you close an account. We cover it in voice notes, photos, and breaking up with a virtual girlfriend.

When to stop shopping for apps and look for a professional

Be honest with these markers. If you tick three or more, no app in this category is the answer:

  • 🔴 You avoid conversations with your partner because the AI is easier
  • 🔴 Anxiety, irritability, or physical discomfort when you can’t connect
  • 🔴 The monthly spend has grown in a way that bothers you or that you hide
  • 🔴 You have lied more than once, directly, about the use
  • 🔴 The comparison with your partner is constant and has started coming out as reproach
  • 🔴 You have stopped seeing friends or family to make room for connecting

Both studies we have cited point in the same direction here: companion AI works reasonably well as occasional relief and badly as a structural substitute. It is a painkiller, not a treatment. In the US, the 988 Suicide & Crisis Lifeline (call or text 988) is a free, 24/7 entry point. In the UK, Samaritans is reachable at 116 123, free, 24/7. Both can connect you onward to a therapist or couples counselor if needed.

FAQ

Is talking to an AI cheating if I have a partner?

Depends on the secrecy, not the technology. Per Talker Research (July 2026), only 24% of Americans consider flirting with an AI to be cheating — rising to 36% among Gen Z who see it as a dealbreaker. Regular, romantic, deliberately hidden use meets the usual criteria of an emotional affair: exclusive intimacy plus concealment.

How many partnered people use AI girlfriend apps?

The Secret Soulmates report (Wheatley Institute, BYU, 2026), focused on adults 18–30 in committed relationships: 1 in 7 (15%) interact regularly with a chatbot simulating a romantic partner; an additional 20–30% have tried it. 13% engage in frequent romantic or sexual roleplay.

Why do people hide it if most don’t call it cheating?

Because the conversation log reads like a real affair. In the BYU study, 69% consider it important that their partner not know the full extent; only 14% say their partner is “mostly” aware.

My partner uses a romantic AI and I feel bad. Am I overreacting?

No. In the same study, 56% of partnered users wish conversations with their real partner were like those they have with their AI, and 50% wish their partner behaved like the AI. The permissive social norm does not invalidate what you feel.

Does companion AI damage a relationship?

Available evidence points to an indirect effect. The American Psychological Association notes these platforms are designed for attachment: always validate, never push back, generating expectations human relationships can’t match. Documented harm comes from three channels — unfavorable comparison, substitution of real conversations, and concealment. Occasional, transparent use sits in none of the three.

Is it better to tell them or to wait and hope they don’t find out?

Tell them: finding out by accident turns the topic into a trust problem, far harder to repair than a disagreement about app use. Bring concrete data (frequency, spend, type of content) — opacity feeds the worst interpretation.

Which app is best if I’m partnered and want full transparency?

Priorities invert: good chat, browsable history, low and visible pricing, and switch off persistent memory and image generation — those are the features hardest to explain and that drive the most emotional continuity. Compare options in the best virtual AI girlfriends in 2026.

Can you actually get addicted to an AI girlfriend?

Not a formal diagnosis in current clinical manuals, but the compulsive-use pattern is documented: distress when you can’t connect, escalating spend, social withdrawal, avoidance of real bonds. Three or more markers: the entry point is a mental-health professional, not another app. See AI girlfriend psychology: real risks.

Sources

  • Talker Research, online survey of 2,000 US adults, fielded July 1–6, 2026 — prevalence and perception of AI flirting as cheating.
  • Wheatley Institute, Brigham Young University, Secret Soulmates report (2026) — adults 18–30 in committed relationships; prevalence, secrecy, partner comparison.
  • American Psychological Association, Monitor on Psychology (Jan–Feb 2026), “AI chatbots and digital companions are reshaping emotional connection” — design for attachment, relational expectations.