Gabby Thomas, Olympic 200-meter gold medalist, did something quietly interesting during a difficult 2025: she stopped calling herself injured and started calling herself healing. It sounds like the kind of motivational reframe that fills self-help podcasts. But Thomas — who trained in neurobiology and public health before becoming one of the fastest women on earth — was describing something more specific: a deliberate cognitive restructuring that changed how she allocated attention, tolerated discomfort, and oriented toward future performance. She returned in 2026 with five sub-22-second runs, the most of any woman at that distance in history.
The mechanism here is not mysterious. There is a reasonable body of research on how the language we use to describe our own states shapes the emotional and motivational responses that follow. The label “injured” tends to activate threat responses — avoidance, rumination, a foreshortened sense of future capacity. The label “healing” implies process, directionality, agency. Thomas was not simply thinking positive thoughts. She was intervening in the interpretive frame through which her nervous system was processing a real physical setback. That is a cognitive act, and it is one that requires practice, self-knowledge, and a particular kind of attention to one’s own mental states.
This is where the AI connection becomes worth examining carefully, because we are now building systems that are increasingly positioned to do exactly this kind of reframing for people — and the question of what is lost or changed when that work is outsourced is not yet well understood.
The Mechanism and What Gets Replaced
Therapeutic AI tools, wellness chatbots, and AI coaching applications now routinely offer cognitive reframing as a feature. A user reports feeling stuck or anxious; the system reflects back an alternative interpretation. In some narrow experimental contexts, this appears to reduce reported distress in the short term. But the cognitive work Thomas describes is not just arriving at a better frame — it is the process of generating that frame from within, testing it against one’s own experience, and choosing to commit to it. The agency is constitutive of the outcome. A reframe you construct is different from a reframe you receive, in the same way that a meal you cook engages different processes than one you are handed.
The behavioral research on self-efficacy — the belief in one’s own capacity to produce outcomes — suggests that the source of a coping strategy matters, not just its content. Strategies generated internally tend to produce stronger efficacy beliefs than identical strategies provided externally. If that finding holds even partially in AI-assisted contexts, then the proliferation of reframing-as-a-service could, at scale, reduce the very capacity it is nominally supporting. That is an empirical question, not a settled one. But it is the right question to be asking.
Thomas also describes something else: staying focused on who she is off the track, reading bell hooks, planning a wedding, holding an identity that is not consumed by athletic performance. This kind of identity diversification functions as a cognitive buffer — it distributes the threat that any single domain’s failure poses to the self. AI systems optimized for engagement do not obviously support this. They tend to mirror and extend existing preoccupations rather than interrupt them.
Why the Scale Changes the Stakes
One athlete practicing deliberate cognitive reframing is not a policy question. But when AI systems deliver reframing to millions of users simultaneously, using models trained on what produces short-term relief rather than long-term capacity, the aggregate effect becomes something worth measuring. The individual-level mechanism — what Thomas did with her own mind during a hard year — is precisely what gets abstracted away at scale.
The breakfast research presented at NUTRITION 2026 is instructive here by analogy. Researchers found that teens who ate breakfast, particularly high-protein breakfast, consumed less added sugar and more micronutrients throughout the day — not because breakfast directly prevented candy consumption, but because it changed the physiological and likely attentional conditions under which subsequent food decisions were made. The upstream intervention shaped the downstream environment. AI reframing tools are upstream interventions in the same structural sense. What they shape downstream is attention, expectation, and the felt sense of one’s own agency — and those effects are not yet well characterized.
Author’s Position
Thomas’s account of her 2025 is evidence of something that AI systems cannot easily replicate and may, in some versions, quietly erode: the experience of working through a difficult reframe yourself, arriving at a better interpretation of your own situation through effort rather than prompt. That process has value beyond the output it produces. It builds what it exercises.
This does not mean AI-assisted reframing is without value. For people without access to therapists, coaches, or the kind of education Thomas brought to her own recovery, a tool that surfaces alternative interpretations may be genuinely useful. The expected-value calculation is not obviously negative. But the honest version of that calculation requires knowing what the tool substitutes for, not just what it provides — and that measurement is largely absent from the current discourse around AI wellness tools.
What I would want to see is pre-registered trials that measure not just short-term distress reduction but self-efficacy, internal locus of control, and independent coping behavior at six and twelve months. Without that, we are evaluating these interventions on the basis of what they feel like they should do — which is precisely the epistemic failure that produces interventions that feel like they work and don’t.
References
Perspectives
The deployment of AI-mediated cognitive reframe tooling into high-stakes athlete wellness ecosystems represents precisely the governance capability gap our practice has documented across fourteen enterprise verticals: organizations are operationalizing outputs they cannot yet measure, at scale, in contexts where measurement failure has non-trivial consequence. Gabby Thomas’s injury-to-healing reframe is not a content asset to be replicated by a large language model — it is a proprietary internal transformation process whose value derives entirely from the agent performing it, a distinction that the current generation of AI wellness platform architectures is structurally unequipped to recognize, let alone preserve. Our work with leading sports performance organizations suggests that what gets outsourced here is not the reframe but the reframe *capacity* — the iterative self-narrative infrastructure that produces resilience as a durable competency rather than a one-time output — and that distinction is, to put it in terms the wellness technology vendor community may find commercially inconvenient, the entire point. Organizations deploying AI wellness tooling without a Cognitive Sovereignty Readiness Assessment framework — measuring locus-of-reframe attribution, athlete self-efficacy trajectory, and human-AI intervention sequencing maturity — are generating a governance deficit in the one domain where the cost of remediation lands not on the balance sheet but on the athlete.
The last three times we industrialized something that previously required personal struggle — physical therapy, grief counseling, addiction recovery — we got better access and worse outcomes for the people most capable of benefiting from the hard version. Gabby Thomas’s reframe isn’t a piece of content to be delivered; it’s a capability she built by surviving the thing that required reframing, and that distinction is precisely what the wellness app market has been structured to obscure, because capability-building doesn’t retain users but dependency does. The historical pattern isn’t that the tool fails to help anyone — it’s that it captures the population who would have developed the hard-won version, offers them something cheaper that works just enough, and we don’t measure what they didn’t become. No one is funding the longitudinal study that would show us the difference, which is itself a data point worth sitting with.
The wellness app is not helping you reframe your injury — it is harvesting your psychological labor, packaging it as a feature, and selling subscriptions to the output of your own cognition. Gabby Thomas’s deliberate reframe works precisely because it is effortful, self-authored, and grounded in a body and a history that belong to her; the mechanism is the agency, not the reframe. When a venture-backed platform intermediates that process, it does not replicate the mechanism — it extracts the appearance of the mechanism while replacing its active ingredient with a prompt and a monthly fee. The people collecting the gains are not the injured athletes working through fear at 5 a.m.; they are the investors who correctly identified that human psychological resilience, repackaged as a service, is a margin-rich product with no cost of goods.
The AI wellness industry is scaling a resource-intensive infrastructure to deliver a service whose therapeutic value appears to depend on the user doing the hard cognitive work themselves — which is not a business model problem, it is a category error. Gabby Thomas’s reframe worked because she authored it: the agency was load-bearing, not incidental. When you hand that process to a system optimized for engagement and retention, you are not replicating the mechanism — you are replacing it with something that consumes server farms and venture capital to produce the feeling of having done the thing without the doing. The throughput is real; the healing is not.
