From Crowd to Collection: Mapping the Integration Gap in GLAM Crowdsourcing

Heritage crowdsourcing and the integration gap

Heritage crowdsourcing emerged from converging forces: efforts to challenge traditional museum authority, practical needs for enhanced collection documentation, resource constraints, and expanding possibilities for public engagement (Simon 2010; Oomen and Aroyo 2011; Ridge 2014; Terras 2016). Early scholarship was largely celebratory—social tags surfacing community perspectives (Light and Hyry 2002; Van Hooland 2006; Anderson and Allen 2009; Yakel 2011), transcriptions enabling text access, initiatives like Steve.museum (formed 2005) and Your Paintings Tagger (launched 2011) demonstrating practical viability at considerable scale. Even so, as these projects proliferated, fundamental questions about how public contributions translate from platforms to collections remained stubbornly unresolved.

Citizen science quality assurance: lessons for heritage practice

Although integration challenges persist across participatory-science contexts, citizen science has developed sophisticated quality-assurance frameworks in response to scientific-rigour requirements, including consensus-based validation, expert review protocols, and tiered verification systems (Wiggins et al. 2011; Allahbakhsh et al. 2013; Kosmala et al. 2016; Kaldeli et al. 2021; Sharma et al. 2022), that could potentially inform heritage institutions grappling with integration questions. Projects like Galaxy Zoo and Snapshot Serengeti demonstrate that consensus-based vote aggregation, where non-expert classifications are statistically combined, can achieve accuracy levels rivalling expert assessment, while delivering educational benefits (Lintott et al. 2008; Swanson et al. 2016). These successes, then, raise a compelling question: Can such approaches effectively transfer to heritage contexts?

Not easily. Oomen and Aroyo (2011) highlight the difficulties heritage-volunteer collaboration faces in locating skilled participants and maintaining quality control, with project success contingent on careful design and implementation (Noordegraaf, Bartholomew, and Eveleigh 2014). More fundamentally, as discussed earlier, heritage validation operates through provenance assessment, contextual interpretation, and cultural appropriateness—criteria resisting the standardisation characterising scientific-observation tasks, and requiring attention to community sovereignty alongside accuracy (Tuhiwai Smith 1999; Christen 2015).

Platform-adaptation research demonstrates these incompatibilities concretely, while also documenting sustained efforts to bridge disciplinary divides. Blickhan et al. (2019) compared Zooniverse’s standard independent classification method—developed for STEM projects—against collaborative transcription where volunteers could view others’ work. Collaborative approaches produced significantly richer data, prompting platform redesign specifically for heritage projects (Van Hyning 2019). This finding carries important theoretical weight: Transcription tasks involving full-text processing and contextual interpretation operate under markedly different constraints than discrete classification tasks such as galaxy morphology or species identification, and STEM methodologies require considerable modification accordingly.

Even purpose-built solutions encounter such barriers. Zooniverse’s heritage-specific development—beginning around 2014—demonstrates both cross-domain adaptation complexity and the sustained technical investment it demands, with sophisticated aggregation algorithms (Krawczyk 2018) and ALICE (Aggregate Line Inspector and Collaborative Editor), which addressed heritage institutions’ limited capacity to process technical data formats by providing simplified exports for teams without specialist expertise. Yet, these tools still require substantial institutional capacity. Data cleaning, staff labour, ongoing system maintenance—none of these yielded to better software alone (Van Hyning and Jones 2021; Thomer and Rayburn 2024). Collectively, that even well-resourced platforms with dedicated technical teams required significant infrastructural adaptation is, in itself, telling—and points to integration mechanisms as critical yet underexplored research terrain.

Sampling strategy

Sampling prioritised theoretical saturation—the point where additional data yields no new theoretical insights—over statistical representativeness, following established principles for theory-building research in organisational studies (Eisenhardt 1989; Yin 2018). The multi-stage approach enabled systematic comparison across institutional contexts while respecting domain-specific vocabularies and operational constraints.

The survey component focuses on United Kingdom (UK) museums, examining crowdsourcing practices across diverse institutional types, ranging from national and art museums to local-authority museums and historic houses. This sector-specific focus reflects museums’ distinctive collections-management frameworks. Interview data provide international and cross-GLAM breadth, capturing integration mechanisms across museums, libraries, and archives in multiple national contexts. Given significant differences in descriptive paradigms across institutional types, and variations in national funding models and regulatory frameworks, comprehensive international comparative analysis would require separate investigation beyond this study’s scope. This sampling architecture enables in-depth museum-sector analysis while capturing broader integration patterns through comparative interview data.

Survey methodology

Three targeted surveys gathered perspectives from distinct professional communities, each designed to facilitate cross-domain comparison while respecting community-specific vocabularies. Streamlined instruments are provided in Supplemental files 1–3: Appendices A–C.

Interview methodology

Semi-structured interviews (Supplemental file 4: Appendix D) elicited practitioner perspectives on institutional-integration approaches, drawing on organisational ethnography (Spradley 1979) and institutional analysis (Scott 2014).

Data analysis

Heritage-professional survey: integration-preference patterns

Survey responses from heritage professionals (n = 49, 22.6% response rate) exhibited limited engagement with integration planning. Only 40.8% (n = 20) addressed integration-methodology questions, with fewer still (n = 11) providing specific implementation details. Analysis of these responses revealed three distinct patterns: recording into main database (46.1%, n = 6), validation from professional staff (30.7%, n = 3), and incorporation in places other than collections management databases (15.3%, n = 2).

Technical infrastructure: implementation-feasibility analysis

Among seven responding SPECTRUM-compliant CMS vendors (38.8% response rate), three reported current system support for UGC integration. These systems distinguish user contributions from professional documentation through separate data fields, temporal metadata (contributor name, date, time), and labelled public-comment sections.

All three require manual professional validation before integrating content into existing catalogue records. Of these vendors, two capture contribution provenance, while only one prominently displays contributor attribution alongside submission details.

Citizen science survey: quality-assurance approaches

Citizen science coordinators (n = 8, 6.1% response rate) reported rigorous quality-assurance approaches. Among initiatives implementing quality control (n = 4 of 5 responding to Q5; Supplemental file 3: Appendix C), multiple validation methods were employed simultaneously: staff moderation (n = 3), peer-review tools (n = 3), automatic checks (n = 2), and controlled vocabularies (n = 2). Seven of eight projects reported contributions meeting project objectives, with 80% of total respondents characterising UGC as “very valuable” for scientific research.

Interview findings: heritage-professional perspectives

The sample comprised three museums, two libraries, and one combined library/archive, spanning five larger institutions and one small organisation across the UK (n = 3), US (n = 2), and Spain (n = 1). Practitioners predominantly relied on existing external platforms for crowdsourcing activities, including Zooniverse, FromThePage, and proprietary software, with only one institution developing a bespoke in-house solution.

Interview findings: citizen science coordinator perspectives

Interviews with citizen science project managers (n = 6) revealed validation architectures and processing capabilities across diverse scientific domains. In contrast to heritage practitioners, most coordinators (5 of 6) built custom tools in-house, integrating purpose-built applications with existing recording infrastructure.

Epistemic authority and power dynamics

The three integration approaches listed above differ in how institutions exercise epistemic authority, not whether they retain it (Figure 1). While Haklay’s (2013) widely cited participation scale maps increasing citizen involvement in problem definition and data interpretation, heritage crowdsourcing integration mechanisms operate along an orthogonal dimension entirely: the locus of validation authority. Peripheral contribution externalises validation through spatial separation; supervised collaboration centralises it through hierarchical gatekeeping; workflow integration embeds contributions within institutional systems while retaining gatekeeping functions. Importantly, movement along this operational spectrum does not correlate with epistemic-authority redistribution. Even workflow integration, which would achieve full incorporation into authoritative records, preserves institutional control over what knowledge enters collections, how it is contextualised, and whose claims receive institutional legitimation.

Figure 1

It is this decoupling of technical integration from epistemic parity—wherein contributor and institutional-knowledge claims would carry equivalent authoritative weight within collections—that resonates most strongly with Esguerra and van der Hel’s (2021) finding that participatory designs simultaneously open knowledge-creation processes while reinforcing institutional-power structures. All three approaches cluster at the same point on the epistemic spectrum: institutional retention of validation authority. True epistemic parity would require community involvement in validation-process design, shared publication decisions, and infrastructure enabling diverse epistemological frameworks (as exemplified by CARE standards for Indigenous data governance; Christen 2015)—none of which were observed in the institutions studied here.

Peripheral contribution operationalises what might be termed “institutional epistemic compartmentalisation”—building on existing scholarship (Srinivasan et al. 2009) to describe how heritage institutions separate public- and professional-knowledge contributions through technological design, extracting participatory content while preserving traditional-authority structures. One survey respondent’s reflection exemplifies this limitation:

It is this very architectural separation that enacts epistemic hierarchy: physical distance within database structures becomes the mechanism through which institutions assert authority over knowledge legitimacy.

Supervised collaboration increases participatory breadth while maintaining professional gatekeeping, integrating UGC into authority records but subsuming contributions within institutional voice, without crediting their source. As Jansson (2017) and Ridge et al. (2021b) demonstrate, this institutional oversight is reinforced by CMS design: Lacking dedicated fields for diverse user contributions, systems force professionals to selectively incorporate only information that aligns with existing metadata schemas and institutional-credibility requirements. Interview evidence uncovered the scalability costs—progressing from personal review to volunteers to paid contractors (Interviewee 1.2)—laying bare validation as a resource-intensive bottleneck wherever universal professional mediation is mandated.

Workflow integration distributes gatekeeping across institutional processes without eliminating it. The implementation barriers interview participants identified—attribution infrastructure limitations, legal frameworks, professional-training requirements—indicate that genuine workflow integration requires substantial process redesign: validation embedded throughout institutional systems and contributor attribution built-in from the outset. As Eveleigh (2014) argues, participation often functions pedagogically, with volunteers learning professional norms and terminology through engagement; the challenge, then, lies in designing scalable validation processes that acknowledge educational value while building the attribution infrastructure that would enable genuine integration.

Cross-disciplinary methodological transfer

Comparative analysis demonstrates both opportunities and constraints for methodological transfer. Although citizen science has developed sophisticated quality-assurance frameworks, recent scholarship reveals considerable variation, with projects “still in need of norms” around data-quality assessment (Borgman 2016; Bowser et al. 2020). Survey evidence highlights marked differences in validation conceptualisation: Citizen science coordinators expressed confidence in public contributions, rating crowdsourced content as highly valuable for scientific research. Heritage survey responses, however, struck a different note. One respondent, for instance, cautioned that “without moderation, the ramifications for a museum could be disastrous,” reflecting precisely those institutional anxieties around legal compliance, cultural protocols and provenance tracking that Eveleigh (2014) identified.

Nevertheless, certain citizen science approaches prove adaptable to heritage contexts. Distributed validation systems combining automated screening with trained volunteer networks demonstrate particular promise. Interviewee 2.1’s county-recorder systems—where trained biological recorders validate species observations across geographical regions—exemplify scalable quality control under institutional oversight. Heritage institutions show parallel evolution, if less systematically: Interviewee 1.2 described scaling from individual review to “a team of volunteers” and eventually “paid contractors” as volume became “unmanageable”; Interviewee 1.4 implemented collaborative validation where “everything is keyed, reviewed, edited, and approved collaboratively.” Whether consensus aggregation (Kosmala et al. 2016; Swanson et al. 2016) where multiple independent inputs are algorithmically combined, or the collaborative approaches described above, these models share a structural logic: They distribute validation labour while retaining institutional authority over final publication decisions, with raw volunteer data rarely published directly. Such approaches offer GLAMs pathways to maintain verification standards at scales where individual professional review proves untenable.

Interview evidence revealed universal digital-tool adoption among citizen science coordinators (R, Python, and machine learning), while heritage professionals showed more variable computational capability. Half employed sophisticated automated approaches (machine-learning methodologies, Python scripts, and natural-language-processing libraries), while others lacked such infrastructure, instead relying on manual, digitally mediated methods, such as page-by-page review of individual contributions (Interviewee 1.1) or manual back-end verification of submissions against source metadata (Interviewee 1.2). This variation reflects institutional resource constraints rather than inherent domain limitations, underscoring that successful integration requires sustained investment in technical capacity. But the gap is not only one of resourcing. Heritage practitioners’ reliance on external crowdsourcing platforms, which sit entirely outside institutional CMS architectures, points to an integration problem that is fundamentally infrastructural. Citizen science coordinators, by contrast, built ingestion pathways into their tools from the outset. Where that design work happens upstream, then, incorporation tends to follow naturally; where it is retrofitted, the friction risks compounding.

Implementation capacity: resources and professional development

Interview data unpack the confidence gap survey findings identified. Practitioners describe persistent resource barriers including “temporal and budgetary demands” that force institutions to rely on “trusted, pre-trained volunteers” rather than broader public engagement (Interviewee 1.5). Among surveyed heritage professionals who provided budget information, 47% operate on budgets of £30,000 or less, while roughly a quarter exceed £250,000. While these findings cannot be generalised beyond the study sample, they align with broader sector evidence of financial constraints, with recent research indicating that heritage organisations face significant resource pressures (Heritage Alliance 2024).

Beyond financial limitations, interview evidence exposes fundamental shortfalls in professional preparation, particularly around technical infrastructure and cross-disciplinary collaboration. Interviewee 1.3 argued that successful crowdsourcing requires methodical institutional integration rather than isolated initiatives, emphasising both the need for automated data-processing capabilities and the essential human expertise required to extract, process, and incorporate crowdsourced content into institutional catalogues. This challenge is sharpened by capacity constraints: As Interviewee 1.5 observed, “institutions face time constraints and may lack the necessary skills within the cataloguing community,” explaining that “setting up a routine workflow that does not just rely on the goodwill of staff within institutions is harder than just doing this ad-hoc.”

Interviewee 1.6, in turn, identified enhanced interdisciplinary collaboration as essential, arguing that more fluid partnerships between cultural-heritage and technical professionals could enable co-creative approaches leveraging methodologies not traditionally used within heritage contexts. These converging constraints—temporal, financial, infrastructural, and skill-based—indicate that successful workflow integration requires ongoing institutional support well beyond technical capability: inter-professional collaboration, structured staff development, technical infrastructure, and transformation of workflows from ad-hoc initiatives into embedded institutional practice.

Attribution and infrastructure barriers

Attribution surfaced as a concern for five out of six interviewees (83%), spanning both technical and ethical dimensions, and cutting across all three integration approaches. As Interviewee 1.4 poignantly noted, archival systems lack designated fields for contributor attribution—an infrastructural absence the CMS-specialist survey confirmed across the sector. Among systems supporting user-generated content, only one prominently displays contributor attribution alongside submission details, while all require manual integration of audience-created content with existing catalogue records, creating bottlenecks constraining routine workflow implementation. This infrastructural gap exposes a critical dimension of the operational-systemic divide: Even workflow integration, which achieves technical incorporation into authoritative records, risks architecturally obscuring contributor identity—rendering labour visible as knowledge while contributors remain effaced as knowledge producers.

These technical limitations compound across institutional processes. Interviewee 1.2 described uncertainty beginning with fundamental copyright questions, extending through inconsistent contributor-recognition practices, and culminating in ambiguous protocols for citing crowdsourced research in scholarly work. Interviewed practitioners responded through workarounds—content-sharing agreements, archive-location tracking, data-protection compliance—yet, such ad-hoc strategies expose rather than resolve systemic infrastructure gaps. Attribution infrastructure, seen in this light, exemplifies how capacity barriers (technical systems, institutional protocols) intersect with epistemic structures (whose knowledge receives formal recognition). Addressing it requires not merely database fields but a reconceptualisation of contributors as knowledge producers warranting epistemic standing within institutional records.