Can 20 Years of Twitter Be Preserved

As Twitter – now rebranded as X – celebrates its twentieth anniversary in 2026, the platform has established itself as a long-standing component of many people’s everyday and professional lives. Through its lens, we have witnessed contemporary events unfold in real time: elections and political upheavals, natural disasters and humanitarian crises, social movements, and cultural moments. Twitter became a space where society’s perspectives on an extraordinary variety of topics were articulated, debated, and transformed. This sustained presence has rendered Twitter a valuable source of information, not only about our present but also as a potential resource for understanding the past. Two decades of accumulated discourse represent an unprecedented record of public conversation that future researchers – including for example historians and social scientists – might draw upon to understand early twentyfirst-century society (Burgess and Baym; Weller et al.). At the same time, information shared on Twitter was never intended to persist indefinitely. Tweets belong to the category of ephemera: byproducts of everyday life that nevertheless serve as witnesses to their time. Like playbills, pamphlets, and postcards of earlier eras, tweets capture the texture of daily existence and the immediate reactions to unfolding events in ways that more formal records often cannot (Papacharissi). The value of tweets as contemporary sources has been repeatedly recognised, prompting different initiatives over the past two decades aimed at preserving parts of Twitter, including the Library of Congress’s temporary effort to archive all public tweets (Library of Congress, “How Tweet It Is!”). Yet the very qualities that make Twitter valuable as a historical source, namely its immediacy, its informality, and its embeddedness in the flow of everyday life for an unprecedented number of people, also make it resistant to traditional preservation approaches. From an archival perspective, preserving Twitter in its entirety may be viewed as impossible, but also, perhaps, as unnecessary when accepting that preservation involves selection processes. On the other hand, archiving and documenting those subsets of Twitter that have been used as research data for various fields over the past decades would be highly desirable for reproducibility and secondary analysis. However, this is also challenged in various ways, for legal, ethical, and technical reasons. The factors driving data loss on Twitter operate at multiple levels. Some are inherent to the platform’s design: Twitter was built for immediacy, not permanence, and content has always been subject to deletion by users and removal by the platform. Others reflect broader shifts in the platform economy: operators continuously modify interfaces and features to optimise user experience or pivot to new markets, and restrictions on research data access have tightened considerably since the Cambridge Analytica controversy in 2018. And while there may have been some advancement on archiving textual parts of user-generated content on Twitter, several other dimensions of what constitutes Twitter, including its look and feel, remain largely unresolved. These challenges have only intensified following the transition from Twitter to X in late 2022, as changes to API access, verification systems, and content moderation have transformed the conditions of possibility for both real-time data collection and historical reconstruction. This article summarises which approaches have been applied to archive Twitter in different contexts. It differentiates between three levels of analysis: 1) the platform level, 2) the content and data level, and 3) the user level. It explores which aspects of Twitter are already difficult or impossible to reconstruct today and which information may be fully lost, reviewing past and ongoing challenges and the potential limitations of preservation efforts. The Platform Level: Preserving Affordances and Interfaces Perhaps the most significant preservation challenge concerns the platform itself, referring to its interface and affordances: what Twitter looked like, how it functioned, and what it felt like to use at different moments in its history. The look and feel of Twitter have undergone continuous transformation over twenty years, yet comprehensive documentation of these changes remains elusive. Interface designs, colour schemes, typography, the placement and appearance of buttons, the way timelines were organised and displayed: all these elements shaped the user experience profoundly but were rarely systematically recorded. We lack reliable timelines for when specific affordances were introduced, modified, or removed. The evolution of the platform’s visual identity from its early minimalist design through various iterations to its current form as X represents a significant aspect of digital culture that has largely escaped preservation. Figure 1: The first ever tweet retrieved over the years from the Internet Archive’s Wayback Machine. Consider the accumulation of small but meaningful changes over Twitter’s lifespan. The hashtag was not originally a platform feature but a user innovation, proposed by Chris Messina in August 2007 and only later adopted by Twitter itself (Messina). Messina’s original tweet proposing the hashtag is no longer accessible on X, but preserved on the Internet Archive. The “star” button transformed into a “heart” in November 2015, subtly shifting the semantics of appreciation from “favoriting” to “liking” (Twitter, “Hearts on Twitter”). The prompt in the compose field changed from “What are you doing?” to “What is happening?”, a small modification that reflected and reinforced a fundamental shift in how the platform positioned itself in relation to public discourse, moving from personal status updates to real-time news commentary (Rogers). The addition of Bookmarks in February 2018 similarly responded to observed user behaviour, offering a private means of saving tweets for later without the implicit public endorsement carried by a “like” (Twitter, “An Easier Way”). These changes shaped user behaviour and communication patterns. Yet reconstructing exactly when they occurred and how users responded requires detective work across scattered sources. Some traces of Twitter’s interface evolution do persist: the Internet Archive’s Wayback Machine offers partial access to Twitter’s historical interface, preserving snapshots of various Twitter pages at different points in time – and Time.com used this to retrace Twitter’s history or its tenth anniversary (Time). These captures can provide glimpses of how the platform looked in earlier eras, as we have done in fig. 1, though coverage is inconsistent and the interactive, dynamic nature of Twitter is necessarily flattened into static page captures. In addition to the Wayback Machine, researchers have attempted to document changes to Twitter’s user interface. For instance, Brigadir maintained a list documenting changes to the platform’s user interface, with source references and supporting evidence. However, the most recent entries date from 2019 and therefore do not cover the rebranding from Twitter to X. Perhaps unexpectedly, other online platforms might also be a source for retracing Twitter’s development: YouTube hosts numerous tutorial videos demonstrating how to use Twitter at various points in its history, offering unintentional documentation of interface designs and features that might otherwise be forgotten. These informal sources exist outside traditional archival frameworks yet may prove valuable for reconstructing aspects of Twitter’s evolution. Nevertheless, the real-time, interactive experience of using Twitter, including scrolling through a live timeline, watching conversations unfold, and experiencing the platform’s particular rhythm and temporality, remains fundamentally unrecoverable. Although there are difficulties in preserving and accessing changes to Twitter’s interface and affordances, some aspects of the platform have never been visible, or have been observable only indirectly: its algorithms and back-end infrastructure, which underpin and shape the platform’s interface and affordances. In the European Union, the Digital Services Act (DSA) introduces a new researcher access mechanism to these hidden structures: under Article 40(4), researchers may under certain conditions (including that the research relates to systemic risks and appropriate safeguards are in place) request access to non‐public data from very large online platforms such as Twitter. This could make some of these previously invisible structures at least partially accessible. The change in ownership of Twitter and its rebranding as X illustrates how rapidly platform algorithms and interfaces can be altered, highlighting that, despite the impression of autonomous technologies, human power relations and capitalist imperatives continue to shape digital infrastructures. Beyond visual changes such as the logo and layout, functional shifts have included modifications to feed behaviour and content visibility, increased character limits per tweet (or now post), a loosening of content moderation practices framed as protecting free expression, the repurposing of Twitter Blue as X Premium, and the introduction of AI conversational features such as Grok. The Content and Data Level: Tweets, Datasets, and Archives At the content level, the preservation landscape is more complex but also more recoverable than at the platform level. Tweet texts and tweet IDs persist in considerable quantities, scattered across various repositories, including the individual or institutional hard drives of researchers. Using Twitter or X as a historical source now fundamentally depends on the access to such stored copies of tweets. The free and low-cost API access that enabled a generation of social media research has largely vanished with the shift to monetised access for researchers under X’s new ownership – and with X’s pricing model not being affordable for most researchers. This makes existing collections of stored tweets more valuable than ever – but also more precarious. Many were assembled under data access conditions that were never strictly compliant with platform Terms of Service, and the current ownership environment has made sharing and reusing such datasets even more complicated. Moreover, collections stored as tweet IDs depend on rehydration through API calls, which is now financially prohibitive and increasingly futile as content disappears from the live platform. These archives are thus not simply more significant because access has narrowed; they are also at risk of becoming stranded. The Library of Congress’s (LoC) Twitter Archive represents the most ambitious institutional attempt at content preservation. Announced in 2010 as a partnership between Twitter and one of the world’s premier archival institutions, the project aimed to preserve all public tweets. The reality proved more complicated. Technical challenges of processing the massive volume of tweets meant that by 2013, a single search of the 2006-2010 archive could take 24 hours (Zimmer). The LoC eventually scaled back its ambitions in December 2017, shifting from comprehensive collection to selective preservation. Crucially, the LoC only received text, with no images, videos, or linked content, a constraint that grew more consequential as tweets became increasingly multimedia (Library of Congress, “Update” 1). The archive exists, containing billions of tweets from 2006 to 2017, yet remains largely inaccessible to researchers, representing a unique case of preservation without access (Bruns). Beyond the Library of Congress, Twitter content persists through academic publishing. Thousands of peer-reviewed papers have analysed Twitter data, and while these publications rarely include the underlying datasets, they do document findings, methodologies, and often substantial quotations from tweets that served as evidence. This scholarly literature constitutes a distributed record of Twitter’s role in public life, even if it represents a highly selective and mediated form of preservation. Twitter data also exists in a fragmented state across the research community. Sharing research datasets created from social media content is not common practice for researchers, also due to data sensitivity concerns and potential data privacy constraints. The discussion has recently begun shifting from whether it is possible to share social media datasets to how to share them responsibly (Schwalbach and Maurer). Challenges include ethical concerns around online privacy and contextual integrity: the principle that information should flow according to the norms of the context in which it was originally shared (Nissenbaum). Legal aspects also constrain the archiving and sharing of tweet datasets. In addition to data protection regulations (and challenges for anonymisation, pseudonymisation, and data minimisation), additional challenges come from regulations in platform terms of service and obligations around data deletion (Golland et al.). Research data archives such as the one at GESIS have limited archived datasets to collections of tweet IDs, potentially enriched with metadata, in reaction to legal obligations that affect what can be shared and with whom. Specialised initiatives like CrisisLex have compiled datasets related to natural disasters, while Documenting the Now has focussed on preserving tweets related to social justice movements, developing ethical frameworks that respect users’ rights to delete their content while maintaining historically significant records (Jules and Summers). A different model is the Australian Twittersphere project, which took a nationally bounded rather than topic-driven approach by systematically capturing tweets from Australian users over an extended period (Bruns, Burgess, and Highfield), later continued through the QUT Digital Observatory. Data repositories like Zenodo and Harvard Dataverse host numerous Twitter-related datasets, for example deposited by individual researchers. There are also full-text datasets on platforms like Kaggle that do not exercise quality checks on user uploaded data and leave it up to users to verify that they are permitted to upload this kind of content. However, un-curated collections lack standardisation and often fail to meet FAIR principles: being Findable, Accessible, Interoperable, and Reusable. Finally, the open-source software nitter and hosted versions of it (like xcancel.com) are radical versions of the idea to mirror the entire platform by circumventing anti-scraping measures and creating a mirror of currently accessible content on x.com. An alternative source for historically relevant Twitter content emerges through traditional media. Tweets, especially in fields like politics, became a frequent source in newspapers, television news, and online journalism. Screenshots of tweets embedded in news articles preserve specific content that achieved public significance, albeit divorced from its original platform context. This represents an intriguing inversion whereby new media content is preserved through old media forms, creating a mediated but persistent record of Twitter’s most publicly visible moments. There is no comprehensive overview of what Twitter datasets exist; data are everywhere, including on individual researchers’ hard drives and institutional servers, but not in any systematically documented form. The question of reproducibility looms large over this fragmented landscape, reducing the research community’s ability to verify, replicate, and build upon previous work. Twitter datasets are inherently unstable: tweets are deleted, accounts are suspended, and the rehydration of tweet IDs through API calls, once considered a compliant method of data sharing, has become increasingly impractical as free API access has been restricted and more content has disappeared from the live platform. But even where content is preserved in the form of stored collections of tweets or other content datasets, additional challenges remain. The contextual knowledge that made Twitter intelligible to its users at particular moments has also eroded considerably. For example, the first tweets from the Egyptian uprising in January 2011 did not primarily use the hashtags #egypt or #arabspring that researchers and journalists might now search for. Instead, early participants used #jan25, a reference to the date of planned protests that was meaningful at the time but may be opaque to those encountering these archives without contemporaneous context (Lotan et al.). The hashtags that would later become historically significant were often not the ones used in the heat of the moment; the vernacular of crisis communication evolved rapidly, and much of that evolution is difficult to reconstruct – so is the question of which users were relevant for a particular event in Twitter content’s history. The User Level: Profiles, Networks, and Privacy Constraints The user level presents perhaps the most difficult preservation challenges, caught between historical significance and legitimate privacy concerns. User profiles, including the biographical information people provided, the avatars they chose, and the networks of following and followers they constructed, provided essential context for understanding individual tweets and broader patterns of communication. When accounts are deleted or suspended, this contextual infrastructure typically vanishes entirely, making remaining tweets difficult to interpret without their original context. Almost all images and videos from tweets, as well as user profile information, have been lost from systematic preservation efforts. General statistics about Twitter’s user base over time can be partially reconstructed through company reports, academic studies, and industry analyses. We can trace the platform’s growth from niche service to mass phenomenon, document shifts in user demographics and geographic distribution, and identify patterns in adoption and abandonment. However, granular information about individual users raises significant legal and ethical concerns that appropriately constrain preservation efforts. The case of prominent users who have been removed from the platform illustrates these tensions acutely. The suspension of Donald Trump’s account in January 2021, following the Capitol riot, removed a historically significant corpus of presidential communication from the live platform. While his tweets had been extensively documented by journalists, researchers, and the National Archives, which has jurisdiction over presidential records, the removal highlights how platform governance decisions can suddenly render significant content inaccessible. It also speaks to the growing importance of platform ownership (Leerssen). For example, Donald Trump founded the microblogging service Truth Social to avoid dependence on Twitter’s infrastructure and its political‐normative orientation. His subsequent reinstatement on Twitter/X after Elon Musk’s takeover constitutes a dramatic reversal in the platform’s political culture. Other high-profile suspensions and departures, whether of politicians, celebrities, journalists, or influential accounts, similarly create gaps in the historical record that are difficult to fill retrospectively. These examples underscore the significance of digital platforms as sources for historical and social research in the future. Beyond high-profile bans, a platform’s user base can also shift: following Elon Musk’s takeover, alternative microblogging services such as Bluesky, Mastodon, and Threads saw significant gains from an exodus of disaffected users (Peters and Thimm). These high-profile cases, however, represent only one side of the user-level preservation challenge: while the removal of prominent accounts creates historical gaps, the preservation of ordinary users’ data raises a different and equally pressing set of concerns. The preservation of user-level data raises legitimate privacy concerns that must be taken seriously. Most Twitter users did not post with posterity in mind; they shared thoughts, opinions, and personal information in what felt like an ephemeral, conversational context. The principle of contextual integrity suggests that information flows should respect the norms of the context in which content was originally shared (Nissenbaum). Archiving and making accessible the tweets of ordinary users, potentially revealing past opinions, associations, or activities, conflicts with reasonable expectations about how that information would flow. Data protection regulations like European Union’s General Data Protection Regulation (GDPR) protect users’ rights to deletion and provide legal backing for these expectations. The tension between historical preservation and individual privacy has no easy resolution, and different jurisdictions and institutions have reached different conclusions about where to draw the line. But reflecting on user expectations and potential consequences of exposing user-generated content beyond its original context should be part of any broader preservation initiative. Alternatively, the users could themselves decide on what gets preserved. The individual data download feature that Twitter, now X, provides offers a user-centric approach to preservation, allowing people to export their own tweet histories. This places the decision about preservation with individual users rather than institutions, enabling personal archives for those motivated to maintain them. Data donation initiatives have explored whether users might voluntarily contribute their personal archives to research, though uptake remains limited (Kmetty et al. 2025). The extent to which users utilise these features, and what happens to personal archives over time, whether they are maintained, shared, or eventually lost, remains largely unknown. Conclusion Twenty years of Twitter cannot be fully preserved, but neither will they be entirely lost. By differentiating between platform, content, and user levels, we can see that preservation challenges and possibilities vary significantly across these dimensions. The platform level is most thoroughly lost: the experience of using Twitter at different moments, its evolving interface and affordances, can only be partially glimpsed through scattered screenshots, Wayback Machine captures, and YouTube tutorials. The content level exists in a fragmented but partially recoverable state, distributed across the Library of Congress archive, institutional repositories, researcher hard drives, data sharing platforms, traditional media – or potentially paid access to X’s API – though questions of accessibility, standardisation, and reproducibility remain unresolved. The user level is constrained by appropriate legal and ethical considerations that limit what can or should be preserved about individuals, even as the disappearance of prominent accounts creates gaps in the historical record. The lessons from Twitter’s archival challenges should shape how the research community approaches preservation for current and emerging platforms. Most importantly, researchers need to remain aware of their reliance on platform-provided infrastructure for data access. Twitter research was built almost entirely on API access – granted and controlled by the platform itself. When that access was removed, much of the field’s research infrastructure collapsed with it. For platforms like Bluesky, which offers a more open and decentralised architecture, or Mastodon, which operates on federated protocols, researchers have the opportunity to build collection and archiving practices that do not depend on a single company’s continued cooperation. At the same time, newer platforms that integrate AI-driven content curation or operate under closed proprietary models may offer even less room for independent archiving than Twitter once did, and the research community should be cautious about building on access conditions that could be revoked just as easily. Beyond the question of access, the question of what to preserve also deserves more deliberate attention. Total capture of any major platform is neither realistic nor necessary, but the alternative need not be the scattered, opportunity-driven collection that characterises the landscape of remaining Twitter datasets. Nationally or regionally bounded archives represent one promising model: curated, longitudinal collections that act based on clear collection policies and selection strategies, for example placing content within a defined population rather than capturing isolated events or random samples. Systematic documentation of platform changes, including interface design, algorithmic behaviour, and terms of service, would also help preserve dimensions of platforms that are otherwise lost without notice. Preserving metadata and contextual information alongside content would further improve the long-term usefulness of archived datasets. Concrete steps are also needed to address what already exists. Much Twitter data survive on individual hard drives, institutional servers, and scattered repositories, but no comprehensive overview of these holdings exists. Scoping reviews to map what datasets are available, in what form, and on which legal basis would be a valuable starting point. Clearer frameworks for responsible data collection and sharing, building on the work of initiatives like the Australian Internet Observatory or Documenting the Now, could allow broader reuse while respecting ethical and legal boundaries. Institutional partnerships between research data archives, libraries, and platform studies communities would provide more sustainable long-term custodianship than researchers working in isolation. Twenty years of Twitter cannot be fully reconstructed, but the research and archiving community can learn from the losses we have experienced. References Brigadir, Igor. “igorbrigadir/twitter-history.” Zenodo release, version v14. Zenodo, 2021. 27 Jan. 2026 . Bruns, Axel. “The Library of Congress Twitter Archive: A Failure of Historic Proportions.” Medium: DMRC at Large, 10 Jan. 2018. 27 Jan. 2026 . Bruns, Axel, Jean Burgess, and Tim Highfield. “A ‘Big Data’ Approach to Mapping the Australian Twittersphere.” Advancing Digital Humanities: Research, Methods, Theories. Eds. Paul L. Arthur and Katherine Bode. London: Palgrave Macmillan, 2014. 113–29. Burgess, Jean, and Nancy K. Baym. Twitter: A Biography. New York: NYU Press, 2020. Golland, Luisa, et al. “From (Almost) Open to Heavily Restricted Data Access – the Development of the Twitter/X Developer Policies.” Big Data & Society, 13.1 (2026). . Jules, Bergis, and Ed Summers. “Documenting the Now.” Documenting the Now. N.d. 27 Jan. 2026. . Kmetty, Zoltán, et al. “Determinants of Willingness to Donate Data from Social Media Platforms.” Information, Communication & Society 28.7 (2025): 1324–49. Leerssen, Paddy. “From Murdoch to Musk: Platform Ownership and the Political Economy of Online Content Governance.” Platforms & Society 2 (2025). DOI: 10.1177/29768624251386260. Library of Congress. “How Tweet It Is! Library Acquires Entire Twitter Archive.” Library of Congress Blog, 14 Apr. 2010. 27 Jan. 2026 . ———. “Update on the Twitter Archive at the Library of Congress.” Library of Congress White Paper, Dec. 2017. 27 Jan. 2026 . Lotan, Gilad, et al. “The Revolutions Were Tweeted: Information Flows during the 2011 Tunisian and Egyptian Revolutions.” International Journal of Communication 5 (2011): 1375–1405. Messina, Chris. “Twitter Hashtags for Emergency Coordination and Disaster Relief.” Factory Joe, 22 Oct. 2007. 27 Jan. 2026 . Nissenbaum, Helen. Privacy in Context: Technology, Policy, and the Integrity of Social Life. Stanford: Stanford UP, 2010. Papacharissi, Zizi. “Without You, I’m Nothing: Performances of the Self on Twitter.” International Journal of Communication 6 (2012): 1989–2006. Peters, Yannik, and Caja Thimm. “Microblogging.” Elgar Encyclopedia of Political Communication. Eds. Alessandro Nai, Max Grömping, and Dominique Wirz. Edward Elgar Publishing, 2025. 517–520. . Rogers, Richard. “Debanalising Twitter: The Transformation of an Object of Study.” Twitter and Society. Eds. Katrin Weller et al. New York: Peter Lang, 2014. ix-xxvi. Schwalbach, Jan, and Rainer Mauer. “Sharing Digital Trace Data: Researchers’ Challenges and Needs.” Big Data & Society 12.1 (2025). . Shah, Jesah. “An Easier Way to Save and Share Tweets.” Twitter Blog, 28 Feb. 2018. 20 Feb. 2026 . Time. “See the Evolution of the Twitter Profile over the Past 10 Years”. 20 Feb. 2026 . Twitter. “Hearts on Twitter.” Twitter Blog, 3 Nov. 2015. 27 Jan. 2026 . Weller, Katrin, et al., eds. Twitter and Society. New York: Peter Lang, 2014. Zimmer, Michael. “The Twitter Archive at the Library of Congress: Challenges for Information Practice and Information Policy.” First Monday 20.7 (2015). 27 Jan. 2026

Conversation · Everyday life · Politics · Social media · Oral History, Memory, Narrative Analysis · Social Media in Health Education · Web and Library Services