The exact text we saved

This is the version exactly as it was saved, character for character. The SHA-256 hash below is a 64-character code worked out from that text. Run the same text through SHA-256 and you get the same code; change a single letter and the code comes out completely different. Each version also stores the previous version’s hash, which chains them together so no version can be swapped out quietly.

TRUVACE RECORD VERSION
record: TRV-2026-2795
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-11-12T17:24:58.029011Z
status: published
lens: g_space
sector: health
headline: Efficacy of relational agents for loneliness across age groups: a systematic review and meta-analysis
dek: Abstract Background Loneliness is a serious public health concern. Although previous interventions have had some success in mitigating loneliness, the field is in search of novel, more effective, and more scalable solutions. Here, we focus on “relational agents”, a form of software agents that are increasingly powered by artificial intelligence and large language models (LLMs). We report on a systematic review and meta-analysis to investigate the impact of relational agents on loneliness across age groups. Metho…
gain_title: A meta-analysis of 14 studies with 286 participants aged 19 to 100 found that relational agents (companion robots and app-based conversational agents) reduced self-reported loneliness by a moderate amount (Hedge's g = -0.552).
problem_title: (none)
trace_subject: (none)
gain_reading: A meta-analysis of 14 studies with 286 participants aged 19 to 100 found that relational agents (companion robots and app-based conversational agents) reduced self-reported loneliness by a moderate amount (Hedge's g = -0.552).
gain_evidence: Relational agents reduced loneliness significantly at an adjusted α Bonferroni ( g = -0.552; 95% Knapp-Hartung CI, -0.877 to -0.226; P = 0.003), which corresponds to a moderate reduction in loneliness.
problem_reading: (none)
problem_evidence: (none)
quick_read: A systematic review and meta-analysis in BMC Public Health pooled 14 studies of relational agents, software or robots that build rapport through conversation and play and increasingly use AI. Together the studies covered 286 participants aged 19 to 100, and the pooled effect on self-reported loneliness was moderate (g = -0.552, Knapp-Hartung 95% CI -0.877 to -0.226).

Social robots such as PARO, AIBO, NAO and Pepper showed a larger pooled effect (g = -0.774) than app-based agents such as Alexa and Elena+ (g = -0.286). The authors suggest the agents could complement existing loneliness services such as the NHS's social prescribing, and that a regional pilot could be a next step.

The evidence is fragile. Nearly all studies were small and at high risk of bias, and a publication-bias adjustment cut the estimate to g = -0.198, a drop of roughly 64%, with a confidence interval that included zero. The authors judge the true average effect to be small to moderate.
limitation: Most of the 14 studies were small, high risk of bias and short in follow-up (mean 5.92 weeks), and a publication-bias adjustment shrank the effect by roughly 64% to a non-significant g = -0.198.
tag: Evidence-backed gain
key_points: The search of 11 databases up to 16 September 2022 found 3,935 records; 14 studies met the criteria, nine uncontrolled non-randomised trials and five randomised trials. | Study sizes ranged from 4 to 42 participants, 286 in total; participants' ages ran from 19 to 100, and only two studies included anyone under 50. | Nine studies used social robots (Sony's AIBO, PARO, NAO, Pepper, Joy for All pets) and five used app-based agents (Laura, Elena+, Alexa, Bella, PACO). | Pooled effect: g = -0.552; the five randomised trials alone gave g = -0.437, significant at the conventional 0.05 level but not after Knapp-Hartung adjustment. | Social robots gave g = -0.774 (Z = -2.909, P = 0.004) versus g = -0.286 for app-based agents, which was not significant after adjustment. | Heterogeneity was substantial (I2 = 71%), and the 95% prediction interval ran from -1.519 to 0.415. | Eighty-six percent of studies were classed as feasibility studies, and attrition ranged from 0 to 94%. | The mean gap between pre-test and final post-test was 5.92 weeks.
rundown: The authors preregistered the review with PROSPERO and screened titles, abstracts and full texts independently in duplicate. Studies needed a quantitative self-reported loneliness outcome with follow-up of at least one week, and all agents were autonomous rather than human-operated. Risk of bias was rated with RoB 2 for randomised trials and ROBINS-I for non-randomised studies.

After removing 1,910 duplicates, reviewers screened 2,063 records and read 155 full texts. All randomised trials except one, and all non-randomised studies, were rated high risk of bias. Egger's test (P = 0.004) and funnel plots suggested a small-study effect. The trim-and-fill adjustment gave g = -0.198 (95% CI -0.505 to 0.109), though the authors argue it overcorrects under heterogeneity.

Subgroup results were mixed. Eight robot studies and six app studies were too few to give strong power, and the Bonferroni correction (α = 0.0125) was conservative. The authors point to the NHS's resource constraints and suggest agents as one option within social prescribing alongside human-centred schemes, and note that digital literacy, access and data privacy would need resolving in any pilot.

The authors call for more high-quality randomised trials with larger samples and longer follow-up. Young people were largely absent from the studies, and the agents' use of behaviour-change theory was rarely documented.
sources:
- peer_reviewed | BMC Public Health | https://doi.org/10.1186/s12889-024-19153-x | 2024-07-06
prev: 0000000000000000000000000000000000000000000000000000000000000000

SHA-256 hash

7863697b3267f630c01c1109ebd671fad507601847d2664c66a84f9f0fdd109d

Previous version’s hash

0000000000000000000000000000000000000000000000000000000000000000