Caregiver Queries: The Overlooked Half of Patient-Side AI Search

When a pharma brand maps “patient-side” AI search, it typically pictures one person: the individual living with the condition, typing symptoms into an AI assistant late at night. That picture is incomplete. A large and growing share of the questions AI systems answer about a therapeutic area are not asked by the patient at all. They are asked on the patient’s behalf, by someone managing medications, coordinating appointments, or trying to make sense of a diagnosis for a parent, spouse, or child. Generative Engine Optimization (GEO) strategies built around a single, first-person patient voice are, by construction, missing this audience.

A bigger population than most content plans assume

Family caregiving is not a niche audience. The 2025 edition of Caregiving in the US, produced by AARP and the National Alliance for Caregiving, found that nearly one in four American adults — an estimated 63 million people — provide ongoing care to an adult or child with a medical condition or disability, a 45% increase from a decade earlier.¹ Of these, 59 million care for an adult with a complex condition, and nearly a third report caring simultaneously for both a child and an adult, the so-called sandwich generation.¹ Caregivers are also getting younger and taking on heavier loads: 44% now describe their role as high intensity, and nearly one in five rate their own health as fair or poor as a direct result of caregiving.¹

Dementia care illustrates the point at scale. The Alzheimer’s Association’s 2026 Facts and Figures report estimates that more than 12 million family members and other unpaid caregivers provided upward of 19 billion hours of dementia care in 2025 alone, valued at more than $446 billion in unpaid labor.⁵ Every one of those caregivers is, functionally, a health information consumer navigating a condition that is not their own.

Caregivers don’t search like patients

The behavioral evidence is consistent: caregivers are not simply patients with a different diagnosis. A national analysis using the Health Information National Trends Survey found that caregivers were significantly more likely than non-caregivers to use a computer, smartphone, or other electronic device to look up health information for someone else, and to report that their most recent search served both themselves and another person at the same time.³ Caregivers were more than twice as likely as non-caregivers to report that their most recent search served both themselves and someone else at the same time, roughly 43% versus 21%.³

This proxy-seeking behavior has its own research literature. A 2022 mixed-methods review synthesizing 28 studies found that informal caregivers report more frequent and more sustained proxy information-seeking than non-caregivers, driven by a felt sense of responsibility for someone else’s health outcome.⁴ That responsibility changes the shape of the query itself. A patient asks what a side effect means for them; a caregiver asks whether something is normal for their mother, or whether a change in behavior is serious enough to call a doctor. The phrasing is third person, the stakes involve a proxy decision rather than self-management, and the underlying anxiety is often about recognizing when a situation has become urgent enough to escalate.

Why AI answer engines are still built around the wrong voice

Large language models retrieve and synthesize what the web has already produced at scale, and the disease-education content corpus has been written overwhelmingly in the first person, framed around living with a condition or managing your own treatment. Caregiver-authored or caregiver-directed content exists, but it is a smaller and less structured slice of the total corpus, which makes it less likely to surface as a matched, citable source when an AI system parses a proxy-phrased query. This is consistent with the mechanics behind how AI systems select sources more broadly: citation depends on a structural and topical match between the query and the content, not on subject-matter accuracy alone. A page written entirely in second-person patient voice is a weaker structural match for a third-person caregiver query, even when the clinical content underneath is identical.

The result is a quiet visibility gap. It is not that AI systems refuse to answer caregiver questions. Pew Research Center’s October 2025 survey found that 22% of US adults now get health information from AI chatbots at least sometimes, and that only 18% of AI-chatbot users who have ever received health information from a chatbot describe that information as very or extremely accurate.² The gap is that pharma-produced content is largely absent from the source pool those answers draw from, ceding the caregiver moment to forums, general consumer health sites, and unmoderated AI synthesis with no clinical governance behind it.

A privacy and ethics layer that patient content doesn’t have

Caregiver-facing content also carries a proxy-decision dimension that pure patient content does not. A 2025 nationally representative US survey on digital proxy behavior found that while most caregivers with formal account access used it appropriately, roughly one in three caregivers who help manage a care recipient’s medical or financial accounts do so informally, without authorized proxy permissions, representing an estimated 18 million US adults; among these informal proxies, more than half reported knowing the account holder’s login credentials, and close to a third reported using the account without the account holder present.⁶ That single finding should shape how pharma content addresses caregivers. Informational content can support a caregiver’s understanding, but it should not be built or worded in ways that assume the caregiver holds full decision-making authority, since a meaningful share of caregivers navigating a loved one’s health information do not hold that authority formally at all. Content aimed at caregivers needs the same MLR discipline applied to any unbranded disease-education page, plus explicit language that respects the care recipient’s own autonomy and privacy, consistent with the principles behind the EFPIA Code of Practice governing patient-directed communication.⁷

Where the gap matters most

The caregiver visibility gap is not evenly distributed across therapeutic areas. It is widest wherever a condition combines cognitive impairment, a pediatric patient, or a rare and complex diagnosis, because in each of these cases the caregiver is not simply supporting the patient’s search — they are frequently the only person capable of describing symptoms, tracking medication response, and making time-sensitive escalation decisions. Dementia and other cognitive conditions sit at one end of this spectrum, with caregiving responsibilities that grow heavier as the disease progresses and a documented association between caregiving intensity and the caregiver’s own risk of emotional distress and declining health.⁵ Pediatric chronic disease sits at the other end, where, by definition, the parent or guardian is the primary information seeker for the duration of the child’s care, not an occasional proxy. Rare disease caregiving compounds both problems at once: caregivers of rare disease patients are often searching for information about a condition that is itself thinly represented across the web, layering the same structural mismatch between proxy-phrased queries and first-person patient content onto an already sparse citable corpus. A GEO program that treats caregiver content as a footnote to patient content will therefore underperform most severely in exactly the therapeutic areas where caregiver information needs are highest.

Building caregiver-inclusive GEO without duplicating effort

None of this requires a parallel content operation. It requires deliberately widening the audience a disease-education asset is written for. Four adjustments do most of the work.

First, map the caregiving journey stage a piece of content is meant to serve — early diagnosis shock, day-to-day management, or crisis escalation — since caregiver information needs shift sharply across that arc, just as patient needs do across a treatment journey.

Second, build FAQ sections using proxy-phrased questions, such as what a symptom means when it appears in a parent rather than in the reader, alongside first-person ones, since these are structurally different queries even when the underlying clinical answer is the same, and both need to be independently retrievable.

Third, attribute caregiver-relevant sections to credible caregiver-facing voices — a nurse educator, a patient advocacy organization, or a caregiving-specific program — reinforcing the Experience dimension of E–E–A–T for an audience whose lived experience of the condition differs from the patient’s own.

Fourth, track caregiver-phrased query citations as a distinct KPI segment, the same way global and local citation performance are already tracked separately, since a brand can be well cited for patient queries and functionally invisible for caregiver ones.

The bottom line

Caregivers are not a subset of the patient audience. In query behavior, information needs, and even legal standing around account access, they are a distinct audience that current GEO strategies largely write around rather than for. As AI systems become a larger share of how both patients and caregivers first encounter a condition, the brands that build structurally distinct, ethically grounded caregiver content now will be the ones AI systems can actually find and cite when the next wave of caregivers starts typing.

References

1. National Alliance for Caregiving, AARP. Caregiving in the US 2025. Washington, DC: AARP; 2025. https://www.aarp.org/pri/topics/ltss/family-caregiving/caregiving-in-the-us-2025/
2. Pew Research Center. Health information from social media and AI rated more convenient than accurate. Pew Research Center. Published April 7, 2026. Accessed July 13, 2026. https://www.pewresearch.org/science/2026/04/07/users-of-social-media-and-ai-chatbots-for-health-information-are-more-likely-to-say-they-are-convenient-than-accurate/
3. Bangerter LR, Griffin J, Harden K, Rutten LJ. Health information-seeking behaviors of family caregivers: analysis of the Health Information National Trends Survey. JMIR Aging. 2019;2(1):e11237. doi:10.2196/11237
4. El Sherif R, Pluye P, Ibekwe F. Contexts and Outcomes of Proxy Online Health Information Seeking: Mixed Studies Review With Framework Synthesis. J Med Internet Res. 2022 Jun 24;24(6):e34345. https://pubmed.ncbi.nlm.nih.gov/35749210/
5. Alzheimer’s Association. 2026 Alzheimer’s disease facts and figures. Alzheimers Dement. 2026;22:e71345. doi:10.1002/alz.71345 https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/alz.71345
6. Foong PS, Zakaria C, Pakianathan P, Phua AIH, Koh GC. The prevalence and predictors of digital proxy behavior in the United States: cross-sectional survey study. J Med Internet Res. 2025;27:e69806. doi:10.2196/69806 https://www.jmir.org/2025/1/e69806
7. European Federation of Pharmaceutical Industries and Associations. EFPIA Code of Practice. Brussels, Belgium: EFPIA; updated February 26, 2025. Accessed July 13, 2026. https://www.efpia.eu/relationships-code/the-efpia-code/

Olivier Gryson, PharmD, MSc
25 years of experience in digital marketing in the pharmaceutical industry
Special focus on AI Search in Pharma Marketing