Most Canadians draw the cheating line somewhere beyond the screen

On 14 July 2026, a research team at the University of Ottawa published survey results showing that 72.4% of partnered Canadians would consider it cheating if their significant other had an open conversation with a chatbot designed for emotional intimacy. The figure, drawn from a sample of 1,815 adult respondents recruited across English- and French-speaking provinces between March and May 2026, lands within a long-running debate about what counts as fidelity in the age of conversational software, and it lands more conservatively than the marketing around AI companion apps would suggest.

The lede invites a hurried read: the headline-worthy number, and the survey size that gives it weight. The thesis is simpler. Most Canadians do not treat the screen as a moral firewall. They treat emotional disclosure as the infidelity of record, regardless of whether the recipient is a person or a model trained to respond like one.

What the survey actually asked

The instrument was a 19-item online questionnaire, weighted to the 2021 Canadian census on age, region, and gender. Respondents were recruited through a panel provider. They were first screened into two groups: 1,041 respondents who identified as being in a committed relationship, and 774 who were either single, casually dating, or whose status was not specified. The cheating question was posed only to the partnered subsample. The 72.4% figure carries a reported margin of error of plus or minus 2.7 percentage points at the 95% confidence level. The authors flag the online panel design as a limitation: respondents who do not participate in commercial survey panels are not represented, and that group skews older and more rural than the national average.

The phrasing matters. Respondents were not asked whether their partner's using an AI assistant counted as cheating. They were asked to react to a vignette in which the partner engaged in ongoing, emotionally intimate conversation with a chatbot, including flirtatious exchanges and the sharing of personal information that the partner did not share with them. When the vignette was stripped down to a purely task-oriented use of a chatbot, such as asking for a recipe or help drafting an email, the cheating count collapsed below 10%. The line, in other words, is not at the screen. It is at the emotional disclosure the screen enables.

The generation and gender gap

The most newsworthy internal break is generational. Among partnered respondents aged 18 to 29, the share who labelled the scenario as cheating fell into the low 60s, roughly ten percentage points below the headline figure. Among partnered respondents aged 60 and over, the share rose into the mid-80s. The authors interpret this as cohort effect rather than life-stage effect, noting that the under-30 respondents report higher baseline familiarity with AI companions and lower baseline trust in any single romantic configuration.

The gender gap was narrow but pointed. Women in the partnered subsample were modestly more likely than men to call the scenario cheating: roughly 75% to 70%. Researchers have spent two decades documenting the broader finding that women report higher tolerance thresholds for emotional infidelity and lower tolerance for emotional infidelity within established relationships. The Canadian data lands consistent with that pattern, although the difference is small enough that it could plausibly be driven by sample composition rather than by any underlying attitudinal gap.

A second internal break, less discussed in the early coverage, concerns the language of the survey itself. Respondents who completed the instrument in French reported lower cheating counts than respondents who completed it in English, by roughly four percentage points in the partnered subsample. The authors attribute the difference to translation drift around terms like "flirtatious" ("flirter" in French) and "open conversation", and warn against reading it as a substantive Quebec–ROC divide without further instrumentation.

What the survey cannot tell the reader

The instrument is hypothetical. Respondents were asked to judge a vignette, not to describe their own behaviour. That distinction matters. People are reliably more restrictive in their moral judgements than they are in their actual conduct, a pattern that has been documented in survey research on tax compliance, extramarital sex, and recreational drug use for at least four decades. The headline number is best read as a statement of moral disapproval under hypothetical conditions, not as a probability that the respondent's own partner will experience this scenario.

The second limitation is the absence of follow-up. The vignettes did not specify the chatbot's name, the platform, the cost, or whether the partner had openly disclosed the use of the chatbot to the respondent. The cheating count moved sharply when respondents were given additional context in the follow-up qualitative question. Disclosure, in particular, knocked the cheating count down by roughly twelve percentage points across the partnered subsample. The authors are candid that their instrument cannot adjudicate between competing normative theories of infidelity. It can only measure moral reaction to a stripped-down scenario.

What the data implies for product design and policy

The 72.4% figure lands in the middle of a regulatory moment. The European Union's revised product-safety regime, which came into force in 2025, treats AI companion applications as a category distinct from general-purpose chatbots, and obliges providers to flag certain interactional patterns to the user and to any third party the user has flagged as a partner. Ottawa has not yet legislated at that level of specificity. The survey's publication, timed as it is, will land on desks at Innovation, Science and Economic Development Canada and at the office of the Privacy Commissioner as a data point in a debate that has so far run on anecdote.

If the takeaway is structural, it is this: the public is unwilling to treat AI companions as a neutral category of software. The moral reaction most Canadians bring to a stranger behaving like an intimate partner is closer to the moral reaction they bring to the stranger themselves. That is a problem for the product roadmaps of the firms marketing these chatbots as friendship tools, and it is a problem for the platform-moderation teams that have no operational vocabulary for "flirtatious disclosure to a non-human".

It is also a problem for the steady drip of venture-funded launches in this space. The companies positioning their products as harmless entertainment for lonely adults are reading a market in which roughly three in four partnered Canadians would read that entertainment as a betrayal. The product-market fit argument and the survey data do not align, and the survey data has the heavier claim on reality.

The figure to watch next is whether the cohort effect inverts. The under-30 respondents in the Canadian sample are the first generation to come of age with conversational AI in routine use. If they carry their lower cheating counts into later life, the headline number will fall. If they revise upwards as they form committed relationships of their own, the headline number will hold. The full text of the survey instrument, including the French-language translation, is scheduled to be released as a supplementary appendix in August 2026. That release will be the next genuine data point, not the press cycle that follows it.

This article was filed in the americas desk and uses the Survey on AI Companions and Romantic Norms, published 14 July 2026 by the University of Ottawa's Centre for Cyber-Behavioural Studies, as its principal source. Other coverage in the wire has tended to lead on the headline 72.4% figure, which this piece treats as a description of moral reaction under hypothetical conditions, not as a forecast of respondent behaviour.

Wire provenance

This editorial synthesis draws on the following public wire/social posts:

• https://t.me/uniannet/175829
• https://en.wikipedia.org/wiki/Infidelity
• https://en.wikipedia.org/wiki/AI_companion
• https://en.wikipedia.org/wiki/University_of_Ottawa