‘My boyfriend is AI’. 

Detected Manipulations & Suggested Changes

Unbalanced reporting

Presenting one side of an issue more extensively or favorably than others, without proportionate representation of alternative perspectives.

The article relies almost entirely on one psychologist (Dr Ho) and one study that concludes AI companion systems pose “unacceptable risks” for users under 18. It gives detailed space to potential harms (warped sense of self, social isolation, poor emotional regulation, stunted conflict resolution skills, echo chambers, unhealthy dependency) but offers only brief, general acknowledgment that AI companions can be used in a ‘viable and healthy way’ and that many teens use them pragmatically and prefer human interaction. Examples: - “The study concluded AI companion systems pose ‘unacceptable risks’ for users under 18, exposing them to responses containing sexual material, offensive stereotypes and even dangerous, life-threatening ‘advice’.” - Long sections on loneliness, echo chambers, distorted relationships, social isolation, and stunted skills, contrasted with only short mentions of possible benefits (e.g., learning social skills, safe space, viable and healthy use).

5 suggestions

Include perspectives from additional experts (e.g., AI ethicists, child psychologists, researchers who study therapeutic or supportive uses of AI companions) who may highlight potential benefits, safeguards, or more nuanced risk assessments.

Provide more detail on the positive findings from the cited study, such as how many teens reported using AI companions to practice social skills, manage anxiety, or for harmless entertainment, and whether any measured benefits were observed.

Add context about existing or proposed safety measures (e.g., content filters, parental controls, age-gating, platform policies) and discuss how these might mitigate some of the highlighted risks.

Explicitly acknowledge uncertainties and gaps in the research (e.g., limited longitudinal data, cultural differences, variation between platforms) so that the risk framing is not one-sided.

Briefly compare the risks of AI companions with those of other common digital tools (social media, online games, messaging apps) to help readers understand relative, not just absolute, risk.

Appeal to authority

Relying heavily on the opinion of an expert or authority figure as proof, without sufficiently presenting underlying evidence or alternative expert views.

The article centers on the views of a single psychologist, Dr Djavila Ho, and uses her authority (licensed associate clinical psychologist, immediate past president of the Jamaican Psychological Society) to support broad claims about potential harms. Examples: - “Ho shared that long-term use of AI companions could possibly contribute to a warped sense of self, among other issues.” - “Ho added, ‘I think it can distort our relationship with others… So the conflicts or the differing views and opinions of others may come to us as a surprise, depending on how we’re engaging.’” - “She noted that there is a possibility of social isolation from real-life family and friends as relationships with AI companions deepen, warped perceptions of the world in general, poor emotional regulation and possibly stunted conflict resolution skills.” These are plausible concerns but are presented largely as expert assertions without reference to specific empirical studies (beyond the one teen study, which focuses on content risks rather than long-term psychological outcomes).

4 suggestions

Cite specific empirical research (beyond the Common Sense Media/Stanford report) that supports claims about long-term psychological effects such as warped sense of self, social isolation, or stunted conflict resolution skills, or clearly label these as hypotheses or clinical concerns rather than established outcomes.

Include at least one other expert with a different or more moderate view on AI companions, or one who emphasizes potential therapeutic or supportive uses, to avoid over-reliance on a single authority.

Clarify the evidentiary status of statements by adding phrases like ‘some clinicians worry that…’ or ‘there is concern, though long-term data are limited, that…’ instead of presenting them as likely outcomes.

Where possible, provide data (e.g., prevalence rates, study sample sizes, effect sizes) rather than relying primarily on expert opinion to convey the magnitude of risks.

Appeal to emotion

Using emotionally charged language or scenarios to provoke fear, concern, or sympathy, potentially overshadowing a balanced evaluation of evidence.

The article’s focus on youth, loneliness, and psychological harm, combined with phrases like ‘unacceptable risks’ and ‘dangerous, life-threatening “advice”’, can evoke strong emotional reactions and concern among caregivers. Examples: - “The study concluded AI companion systems pose ‘unacceptable risks’ for users under 18, exposing them to responses containing sexual material, offensive stereotypes and even dangerous, life-threatening ‘advice’.” - Emphasis on ‘loneliness’, ‘warped sense of self’, ‘social isolation from real-life family and friends’, ‘poor emotional regulation’, and ‘stunted conflict resolution skills’ without quantifying how common or severe these outcomes are. While these concerns may be valid, the emotional framing is not consistently balanced with data on prevalence, severity, or comparison to other risks youth face online.

4 suggestions

Provide quantitative context for the risks described (e.g., what proportion of users encountered sexual or dangerous content, how often, and under what conditions) to help readers gauge the actual likelihood and severity of harm.

Balance emotionally charged descriptions with neutral, data-driven language, for example: ‘The study found that X% of teen users encountered sexual content at least once, and Y% reported receiving advice that researchers classified as potentially harmful.’

Include examples of neutral or positive outcomes (e.g., practicing social skills, feeling less lonely in the short term) with similar specificity, so that emotional emphasis is not solely on negative scenarios.

Explicitly distinguish between potential risks (what could happen) and documented harms (what has been observed), using careful wording such as ‘may’, ‘could’, and ‘there is concern that’, and clarifying when evidence is preliminary.

Cherry-picking data

Selecting specific data points or findings that support a particular narrative while omitting other relevant data that might complicate or balance the picture.

The article cites one 2025 report (‘Talk, Trust, and Trade-Offs’) and highlights its most alarming conclusion (‘unacceptable risks’) and the percentage of teens who find AI conversations as satisfying as or more satisfying than those with real friends (31%). It briefly notes that ‘the majority of teen respondents approach the bots pragmatically and use them mainly for entertainment, overwhelmingly preferring human interaction,’ but does not provide detailed statistics or discuss any measured benefits or neutral outcomes. Examples: - Emphasis on: “31 per cent of respondents found conversations with AI companions as satisfying or more satisfying than those with real-life friends.” - Emphasis on: “The study concluded AI companion systems pose ‘unacceptable risks’ for users under 18…” - Lack of detail on: exact percentages for teens who prefer human interaction, how many reported positive or neutral experiences, or whether the study found any constructive uses (e.g., practicing language, coping with stress).

4 suggestions

Summarize a broader range of findings from the cited study, including specific percentages for teens who prefer human interaction, those who use AI companions mainly for entertainment, and any reported positive or neutral outcomes.

If available, reference additional studies with different or more nuanced findings about AI companions and youth, and briefly compare their conclusions to avoid over-reliance on a single, strongly worded report.

Clarify the methodology and limitations of the cited study (sample size, demographics, platforms studied, time frame) so readers can better understand how generalizable the ‘unacceptable risks’ conclusion is.

Avoid highlighting only the most alarming statistics; present both risk-related and non-risk-related data in similar detail to give a more complete picture.

Oversimplification

Reducing a complex issue to a relatively simple cause-effect narrative, potentially glossing over nuance, variability, and contextual factors.

The article tends to frame AI companions primarily as a risk factor for youth development and social functioning, with limited exploration of individual differences, platform design variations, or contextual factors (e.g., parental guidance, mental health status, cultural norms). It also suggests a relatively straightforward link between extended use and a range of negative outcomes. Examples: - “Ho shared that long-term use of AI companions could possibly contribute to a warped sense of self, among other issues.” - “She noted that there is a possibility of social isolation from real-life family and friends as relationships with AI companions deepen, warped perceptions of the world in general, poor emotional regulation and possibly stunted conflict resolution skills.” These statements group multiple complex psychological outcomes under the broad umbrella of ‘long-term use’ and ‘over-dependence’ without discussing moderating factors (e.g., intensity of use, existing mental health conditions, quality of offline relationships, safeguards in the app).

4 suggestions

Clarify that potential harms are likely influenced by multiple factors, such as frequency and intensity of use, the specific design and safeguards of the AI platform, the user’s existing mental health and social support, and parental or caregiver involvement.

Include examples of different usage patterns (e.g., occasional use for entertainment vs. heavy use as a primary source of emotional support) and discuss how risks may differ between these scenarios.

Add nuance by noting that some youth who are already socially isolated or anxious may be more vulnerable to negative effects, while others may use AI companions without significant impact on their offline relationships.

Explicitly state that current evidence on long-term psychological outcomes is limited and that more research is needed, rather than implying a simple, direct causal pathway from ‘extended use’ to multiple negative outcomes.

Framing effect

Presenting information in a way that emphasizes certain aspects (e.g., risks, harms) over others, influencing how readers interpret the issue.

The headline ‘My boyfriend is AI’ and the repeated emphasis on youth, loneliness, and psychological harm frame AI companions primarily as a potentially dangerous substitute for human relationships. While the article does mention that AI can be used in a ‘viable and healthy way’ and that many teens prefer human interaction, these points are brief and less prominent than the risk-focused content. Examples: - Headline: ‘My boyfriend is AI’ suggests intimate, possibly troubling relationships with AI, priming readers to view the topic as inherently problematic. - The bulk of the article is devoted to warnings, potential harms, and calls for caution, with only short, less detailed mentions of pragmatic or positive uses. This framing can lead readers to interpret AI companions mainly as a threat, even though the evidence presented is mixed and still emerging.

4 suggestions

Adjust the headline or subheading to reflect both the risks and the nuanced findings (e.g., ‘Teens, AI Companions, and Mental Health: Psychologist Urges Caution Amid Mixed Findings’), so readers are not primed to see the issue as purely alarming.

Rebalance the article structure by giving more space and detail to the study’s findings that teens overwhelmingly prefer human interaction and often use AI companions pragmatically for entertainment.

Include a short section explicitly titled or framed as ‘Potential Benefits and Neutral Uses’ that summarizes constructive or harmless ways people use AI companions, alongside the ‘Risks and Concerns’ section.

Conclude with a balanced summary that reiterates both the potential risks and the conditions under which AI companions might be used safely (e.g., with supervision, clear boundaries, and awareness of limitations).