Body Doubling in ADHD: Technique & Framework

• Body doubling is a self-management technique that uses the presence of another person or virtual agent to provide accountability, structure, and reduce isolation.
• It extends traditional practices by incorporating virtual avatars, prerecorded videos, AI, and robots to enhance social facilitation and ambient companionship.
• A twelve-dimensional framework categorizes setups by individual motivation, agent attributes, interaction style, context, and efficacy, guiding experimental validation and design.

Body doubling is a self-management technique used by adults with ADHD in which the individual performs a target task—such as studying, working, or chores—“in the presence” of one or more other agents, termed body doubles. In the current literature, those agents may be real people co-present in a café or library, live or prerecorded videos, virtual avatars, socially assistive robots, or even voice-only AI. The technique is theorized to work through social pressure and accountability, ambient companionship and reduced isolation, external structure, and anxiety reduction when facing difficult or unstimulating tasks, thereby supporting task initiation, sustained focus, and perseverance to completion in domains affected by inattention, impulsivity, and executive-function deficits (Tan et al., 8 May 2026, Ara et al., 15 Sep 2025).

1. Definition and conceptual scope

A narrower formulation defines body doubling as the practice of working alongside another person so that the mere presence of that person provides passive social support, accountability, and a visual reminder to stay on task. A broader formulation, used in recent HCI and mixed-reality work, explicitly includes non-human and non-co-present agents. This broader framing is technically important because it treats body doubling not as a single interpersonal arrangement but as a family of setups that vary by identity, embodiment, interaction, and context (Tan et al., 8 May 2026, Ara et al., 15 Sep 2025).

Theoretical accounts in this literature converge on several mechanisms. One account emphasizes social facilitation, including Yerkes–Dodson type effects, accountability cues, and external scaffolding that reduces self-monitoring load. Another emphasizes ambient companionship, reduced isolation, external structure such as timers or reminders, and emotional regulation support. Within ADHD communities, anecdotal reports further suggest that body doubling reduces overwhelm, provides momentum, and mitigates dysregulated attention lapses. At the same time, the literature does not treat social presence as uniformly beneficial: neurodivergent individuals may experience social anxiety, fear of judgment, and stigma when observed by strangers, and this tension directly shapes the design of AI or virtual companions as “safe” presence.

A recurrent misconception is that body doubling necessarily requires a physically co-present human. The reviewed work rejects that restriction. A second misconception is that body doubling already has conclusive behavioral evidence. The literature instead reports mixed signals: large surveys report subjective benefits, whereas two small controlled trials cited in the VR study found no behavioral effect. The current research trajectory therefore combines conceptual expansion with experimental validation rather than presuming settled efficacy.

2. Twelve-dimensional framework

Recent work proposes a work-in-progress, twelve-dimension framework to describe any body-doubling setup and to guide both research and design. The framework accounts for individual motivation, agent-related dimensions, interaction related dimensions, contextual dimensions, and efficacy. It was built by iteratively reviewing existing HCI and ADHD literature, neurodivergent community discussions, Johansen’s time-space taxonomy, and Gutwin and Greenberg’s awareness framework; it also expands Eagle et al.’s two-axis model into finer-grained levels and adds dimensions that were observed but not formalized (Tan et al., 8 May 2026).

Group	Dimensions	Representative categories
Individual Motivation	Expected self-management effects that motivate adoption	Accountability; task initiation; sustained motivation; distraction reduction; emotional regulation support; external structure & habit cues
Agent-Related	Number of Agents; Identity of Agents; Embodiment of Agents	Familiar vs. unfamiliar; human vs. non-human; living vs. non-living; low/medium/high embodiment
Interaction-Related	Level of Awareness; Level of Interaction; Shared Activity Type; Degree of Explicit Participation	Peripheral/attentive/continuous awareness; high/medium/low interaction; same/different/mixed task; reciprocal/unilateral/mixed participation
Contextual	Time & Space; Situational Emergence; Type of Task	Co-located/remote; synchronous/asynchronous; spontaneous/informally co-created/formally organized; ICATUS 2016 task categories
Efficacy	Measures of effectiveness	Qualitative feedback; usability scales; task performance metrics; clinical/self-report scales

The framework’s formal significance lies in how it decomposes a body-doubling setup into orthogonal or partially orthogonal parameters. Identity of agents is specified through three distinctions—familiar versus unfamiliar, human versus non-human, and living versus non-living—while embodiment ranges from low, such as ambient lights, sounds, or AI voice only, to medium, such as 2D videos or avatars, to high, such as co-present people, 3D AR/VR agents, or socially assistive robots. Interaction is likewise decomposed into awareness, interaction level, shared activity type, and explicit participation.

The framework also hypothesizes directed influences among subgroups, for example Agent-Related → Interaction-Related, Contextual → Interaction-Related, and Interaction-Related → Contextual, but deliberately stops short of modeling every possible pairwise link, which would yield a complete digraph of 12 nodes and $12 \times 11 = 132$ possible directed edges. This suggests a design logic in which body doubling is treated as a constrained multidimensional space rather than an exhaustive relational graph.

3. Agent forms and interaction modalities

By combining Identity and Embodiment, the literature derives a taxonomy of body-doubling agents. The principal categories are real co-present humans, live video streams and VR avatars, prerecorded “study with me” videos, ambient AI assistants, and socially assistive robots in mixed reality. These categories are not merely descriptive; they are linked to different awareness and interaction profiles. Real co-present humans are characterized as high embodiment, high interaction, and continuous awareness, whereas prerecorded videos are medium embodiment, low interaction, and peripheral or attentive awareness, and ambient AI assistants are low embodiment, low interaction, and peripheral awareness (Tan et al., 8 May 2026).

This taxonomy is closely coupled to motivation and task demands. Human doubles may offer the strongest accountability and competitiveness, but they can also elicit social anxiety or fear of judgment. AI doubles may provide companionship without evaluation and reduce social threat, but they may be easier to ignore or may feel less real. The VR study reports this trade-off directly: body doubling was clearly preferred to working alone, yet opinions diverged between human and AI conditions, with some participants valuing “safe accountability” and others preferring stronger interpersonal pressure (Ara et al., 15 Sep 2025).

Interaction design in this literature is not limited to conversation. Awareness may be peripheral, attentive, or continuous; interaction may be high, medium, or low; shared activity may involve the same task, different tasks, or a mixed/hybrid arrangement; and explicit participation may be reciprocal, unilateral, or mixed. A unilateral setup, such as using strangers in a café as unwitting body doubles, differs materially from a reciprocal setup in which all agents knowingly agree to body-double. The framework therefore distinguishes between social presence as mere witness, social presence as occasional acknowledgment, and social presence as mutual feedback.

4. Contexts, task classes, and immersive implementations

Context is formalized through a $2 \times 2$ time-space matrix with co-located versus remote and synchronous versus asynchronous arrangements. The examples given are co-study in a library, a live streamed “study with me,” prerecorded videos, and an in-room VR agent. Situational emergence further differentiates spontaneous, informally co-created, and formally organized body doubling, while task type is classified using ICATUS 2016 categories such as employment, domestic services, caregiving, volunteer work, learning, socializing, leisure, and self-care (Tan et al., 8 May 2026).

Task type and setting directly shape the choice of setup. High-precision or high-cognitive-load tasks may benefit from co-located synchronous setups with continuous awareness, such as a live MR agent that can detect distraction and intervene. Routine or low-stakes tasks can leverage asynchronous prerecorded videos or ambient AI presence. Formal coaching contexts may layer structured reminders and metrics on top of body doubling. Environmental factors including noise, privacy, and technology access also modulate Embodiment and Interaction; one explicit example is that wearing a VR headset while doing household chores may not be feasible.

A concrete immersive instantiation appears in the VR construction study. Study 2 implemented a virtual bricklaying task in a simulated outdoor construction site in Unity for Meta Quest Pro. Participants replicated a 50-brick reference pattern with three colors over four wall bases; each brick had to be picked and placed and could not be removed once placed, and wrong placements were logged as mistakes. Three moving objects—a truck, a cherry picker with sound, and a silent crane—appeared unpredictably, and participants pressed a button upon detection to measure situational awareness. The interface included progress bars, a secondary camera feed showing the double’s workspace, and, in the AI condition, self-talk captions above the agent for idle, milestone, and error events (Ara et al., 15 Sep 2025).

The same study operationalized three conditions: C1 alone, C2 with a human body double, and C3 with an AI body double implemented through Wizard of Oz control. In C2, both players saw each other’s avatars, shared the reference pattern, and viewed dual progress bars and the secondary camera feed. In C3, the participant was told the partner was an autonomous AI named “Remy,” while the motions were secretly human-operated with the same pacing protocol. These implementation details matter because they instantiate several framework dimensions simultaneously: medium embodiment, synchronous remote co-presence, attentive or continuous awareness, and configurable feedback cues.

5. Efficacy, measures, and empirical status

Efficacy is treated as a distinct dimension and may be measured through qualitative feedback, usability scales such as SUS, task performance metrics such as time to completion and errors, and clinical or self-report scales such as ASRS Inattention reduction. The roadmap notes that formal formulae for effect size can be adopted, for example Cohen’s $d$,

$$d = \frac{\mu_{pre} - \mu_{post}}{\sigma_{pooled}}.$$

This framing places body doubling within a standard evaluative repertoire rather than treating it as an informal coping strategy alone (Tan et al., 8 May 2026).

The reviewed empirical work is heterogeneous. Ara et al. (2025) VR bricklaying reported that C2/C3, corresponding to human or AI double, versus C1, corresponding to alone, improved speed, perceived accuracy, and sustained attention, with $t$-tests at $p < .01$. O’Connell et al. (2024) reported 91% voluntary continued use of the Blossom robot and that 73% wanted attention monitoring features. Lalwani et al. (2025) reported that 12/15 participants expressed interest in ongoing use of the Alex robot. The smartphone–coach intervention cited in the roadmap reported that ASRS-Inattention dropped from 28.1 (SD 4.5) to 22.9 (SD 4.3), $p < .001$, with 33% clinically improved versus 0% in control.

The VR comparison study provides the most explicit controlled evidence in the supplied literature. It used a within-subjects design with 3 conditions, counterbalanced by Latin-square, with $N = 12$ adults aged 18–31 $(M = 22.75,\ 5F/7M)$ who had clinically diagnosed or self-reported ADHD, and screening by ASRS v1.1. Each session comprised 20 minutes of task, a post-session 5-item Likert survey, a 5 min break, and a final semi-structured interview of approximately 30 min. Dependent variables were Task Efficiency, Task Accuracy, Object Detection Accuracy, and survey perceptions of Efficiency, Accuracy, Detection, Sustained Attention, and Task Continuity (Ara et al., 15 Sep 2025).

Quantitatively, Task Efficiency showed a repeated-measures ANOVA of $F(2,22)=6.51,\ p=0.006$, with means of $2 \times 2$0, $2 \times 2$1, and $2 \times 2$2 bricks per minute. Bonferroni-corrected post-hocs yielded $2 \times 2$3 versus $2 \times 2$4; $2 \times 2$5 versus $2 \times 2$6; and $2 \times 2$7 versus $2 \times 2$8. Behavioral Task Accuracy did not differ significantly, with Friedman $2 \times 2$9 and means of $d$0, $d$1, and $d$2 for $d$3, $d$4, and $d$5, respectively. Behavioral Object Detection likewise showed no significant difference, with Friedman $d$6. Perceived Task Accuracy showed Friedman $d$7, with post-hocs $d$8 versus $d$9 and $$d = \frac{\mu_{pre} - \mu_{post}}{\sigma_{pooled}}.$$0 versus $$d = \frac{\mu_{pre} - \mu_{post}}{\sigma_{pooled}}.$$1. Sustained Attention and Task Continuity each showed Friedman $$d = \frac{\mu_{pre} - \mu_{post}}{\sigma_{pooled}}.$$2, with both $$d = \frac{\mu_{pre} - \mu_{post}}{\sigma_{pooled}}.$$3 and $$d = \frac{\mu_{pre} - \mu_{post}}{\sigma_{pooled}}.$$4 greater than $$d = \frac{\mu_{pre} - \mu_{post}}{\sigma_{pooled}}.$$5 at $$d = \frac{\mu_{pre} - \mu_{post}}{\sigma_{pooled}}.$$6.

Qualitatively, many participants reported that working alone felt relaxed but also more distractible, slower, and more error-prone. With doubles, participants reported increased focus, energy, and “motivational pressure,” although a minority found human presence distracting. Human doubles were associated with accountability and competitiveness; AI doubles were associated with companionship without evaluation. Progress bars were highly valued for pacing and social comparison, the secondary camera was appreciated but sometimes under-used, and AI self-talk captions were described as adding playfulness, companionship, and light reminders. The empirical state of the field is therefore not one of uniform or conclusive efficacy, but rather one in which measurable gains coexist with preference heterogeneity and context sensitivity.

6. Research gaps and design implications

Several gaps are explicitly identified. The roadmap points to limited mixed reality prototypes that span high embodiment, customizable environments, and seamless real/virtual transitions; to opportunities for more interactive body doubles that sense attention or emotional state and deliver timely nudges; to the need for empirical studies that systematically vary framework dimensions to chart their individual and combined effect on efficacy; and to extensions across other neurodivergent groups and lifespan stages (Tan et al., 8 May 2026).

The design implications are correspondingly structured. Mixed reality testbeds are proposed as a way to embed 3D avatars or robots into the user’s physical workspace, enabling high embodiment and continuous awareness without full immersion overhead. On-the-fly reconfiguration of setting, such as café versus library scene, of body-double appearance, and of interaction modality is recommended. Adaptive interaction is framed around integrating sensors such as eye-tracking, posture, and voice to detect mind wandering or frustration, then providing graduated interventions ranging from subtle color shifts at early distraction to verbal encouragement if off task for more than a threshold interval. Dimension-driven design further recommends explicitly declaring which dimensions a prototype targets and evaluating it against those dimensions.

The VR study sharpens these recommendations by identifying specific design levers: framing collaboration versus competition, conversation style as passive versus proactive, autonomy cues, pacing algorithms as fixed versus adaptive, avatar aesthetics, progress visualizations, and platform-specific feedback. It also recommends personalizing social presence to user comfort, including the choice of human or AI and known or unknown partner, and adapting across VR, desktop, mobile, and wearables, including subtle vibrotactile cues on wearables. AI doubles are presented as potentially more scalable than human facilitation and as offering configurable levels of presence that may reduce social threat (Ara et al., 15 Sep 2025).

Taken together, the literature characterizes body doubling as an informal coping strategy that is being translated into a rigorously parameterized, technology-mediated intervention space. The central unresolved question is not whether presence matters in the abstract, but which combinations of motivation, agent type, embodiment, awareness, interaction level, task type, and context produce measurable benefit for which individuals. This suggests that future progress will depend less on a single canonical body-double design than on systematic navigation of the twelve-dimension design space.