Where did the conversation go?
Posted on: Thursday 24 September 2026
Where did the conversation go? Platform migration in professional discourse, and why pharmaceutical medicine has no data on itself
It has become received wisdom that scientific and clinical conversation has migrated from X to Bluesky. The peer-reviewed evidence for that claim is thinner than its confidence suggests, and it points in two directions at once: what people say they have done and what the platforms record them doing are not the same thing. This article sets out the sampling defect at the centre of the most-cited survey, places it alongside the bibliometric and network data, and argues that nobody has yet measured the population that pharmaceutical physicians occupy. Readers used to appraising observational evidence will recognise the problem on sight; it is exposure misclassification, wearing a new costume.
The evidence everyone cites
The source is a survey of 813 scientists, educators and science communicators published in Integrative and Comparative Biology [1]. More than three-quarters of respondents reported that Twitter had become much less useful for professional networking, with comparable figures for public engagement and for keeping abreast of new work; for every professional benefit once obtained from Twitter, respondents judged that Bluesky now served them better. It is a careful piece of work, candidly reported, and its headline has been quoted often enough to have acquired the standing of an established fact. One detail from it is worth holding on to: roughly half of respondents still held a Twitter account, using it little or not at all.
The sampling frame is the exposure
The survey was distributed to scientists already on Twitter and Bluesky, recruited by snowball methods seeded from the authors’ own accounts and amplified by reshares from high-follower accounts on both platforms. The authors say precisely this, and note that purposive online sampling of this kind reaches those already endogenous to the community being sampled.
The consequence is structural rather than a failure of execution: the sampling frame is constituted by the exposure under study. Anyone who left both platforms, retreated into closed channels, or never left X at all cannot appear in the denominator. It is the sick-quitter problem in fresh clothing, a reference group defined by the very behaviour whose effect is to be estimated. The migration may well have occurred; the point is only that an inference drawn from a sample of this construction is defeasible. The authors deserve credit for flagging the limitation; the secondary commentary that repeats their headline does not.
What behaviour records, as against what recollection reports
Two datasets describe revealed rather than stated preference, and both cut across the survey finding.
The first matched 300,000 academic users across Twitter/X and Bluesky between 2023 and early 2025. Eighteen per cent transitioned [2]. Heavier users were more likely to move, but not to leave; they hedged, maintaining a presence on both platforms.
The second examined 14,497 papers from seven multidisciplinary and information science journals between January 2024 and March 2025 [3]. Coverage, the proportion of papers mentioned at all, was broadly comparable between the platforms. Participation was not. In Nature, the average number of accounts mentioning a paper was 133.40 on X, against 6.86 by the authors’ own method or 27.63 by a commercial altmetric tool. Note that the two Bluesky estimates differ roughly four-fold: the instrument is doing as much work as the platform, which is itself worth a moment’s reflection before anyone reports a migration to two decimal places.
In fairness, the activity that has moved appears qualitatively different. A large-scale analysis of Bluesky posts referencing scholarly articles found higher interaction and greater textual originality than had previously been reported on X, which the authors read as more interpretive engagement [4]. A community may be small, dense and highly visible to itself. That configuration is precisely the one that inflates self-report.
Nobody has sampled us
Every study above samples academic scientists. A marine biologist’s professional network is not a clinical development network, and the differences are not cosmetic. Those of us in clinical development, regulatory affairs, pharmacovigilance and medical affairs operate under employer social media policies, the ABPI Code of Practice or its equivalents, and reporting obligations that attach to what we read as well as to what we write. A pharmaceutical physician who encounters a suspected adverse reaction in a public thread acquires a duty by reading it. That alone is sufficient to drive professional conversation into closed rooms; the trade literature calls the result dark social, meaning private messaging groups and invitation-only communities. By construction, neither a platform-recruited survey nor an altmetric instrument can see any of it.
The position, then, is this. Every published estimate of where the professional conversation went was measured on the platforms it may have left, among a population that is not ours.
What would settle it?
The remedy is not another platform survey, which would be otiose, but sampling by professional body. A frame defined by qualification rather than by platform use is not difficult to construct: membership registers enumerate a specialty more or less exhaustively, and they include the people who never posted at all. A short instrument asking where colleagues now read, where they post, what they have stopped doing and why, with non-users retained in the denominator, would produce the first estimate of professional discourse in this specialty, and would avoid the defect that the best available survey concedes in its own limitations section. Until something of that kind exists, our sense of where our own conversation has gone rests on evidence gathered from a different population by methods that cannot reach us.
The answer might be that we never left; it might be that we were never really there. Either would be worth knowing. At present the claim that the conversation has moved is an assertion sampled from one side of its own hypothesis.
References
- Shiffman DS, Wester J. Scientists no longer find Twitter professionally useful, and have switched to Bluesky. Integr Comp Biol 2025;65(3):538–45.
- Quelle D, Bovet A et al. Why academics are leaving Twitter for Bluesky. arXiv:2505.24801 [preprint]. www.arxiv.org/abs/2505.24801 [Accessed 25 August 2026]
- Arroyo-Machado W, Robinson-Garcia N, Torres-Salinas D. Are there stars in Bluesky? A comparative exploratory analysis of altmetric mentions between X and Bluesky. J Informetr 2025;19(3):101700.
- Zheng E-T, Thelwall M et al. How is science discussed on Bluesky? arXiv:2507.18840v2 [preprint]. www.arxiv.org/abs/2507.18840 [Accessed 25 August 2026]