Deep storytelling: Collective sensemaking and layers of meaning in U.s. elections

Summary

This paper argues that false and misleading election narratives on Twitter during the 2020 and 2022 U.S. elections were sustained less by discrete factual claims than by collaboratively performed “deep stories” — evolving narratives whose meaning depends as much on how they are told as on what is said. Drawing on Election Integrity Partnership data, the authors qualitatively coded thousands of tweets across ten major incidents using paired tweet-level and incident-level codebooks. They show a stylistic shift between cycles: 2020 discourse explicitly constructed mechanics of alleged fraud, while 2022 discourse increasingly relied on allusion and cueing, trusting audiences to supply meaning from the matured 2020 deep story. The paper offers theoretical, empirical, and methodological contributions to studying participatory disinformation whose force lives in narrative context invisible to single-post analysis.

Key Contributions

• Theoretical: Extends Hochschild’s “deep story” concept by arguing that storytelling style — allusiveness, cueing, platform-mediated performance — is constitutive of meaning, and integrates this with participatory disinformation and collective sensemaking traditions (Weick, Klein, Shibutani).
• Empirical: A cross-cycle (2020→2022) longitudinal account of how election rumoring on Twitter shifted from explanation-heavy to allusion-heavy as a shared narrative matured.
• Methodological: A dual-codebook design pairing tweet-level and incident-level qualitative coding with temporal visualizations, providing a template for measuring context-dependent misinformation that resists single-post coding.

Methods

Interpretive, grounded analysis of EIP Twitter datasets covering the 2020 and 2022 U.S. elections (over a billion tweets total). The authors purposively sampled the five largest incidents per cycle (10 total), refining 2020 incidents to event-specific subsets comparable to 2022. Each incident received two samples — a ~100-tweet close-reading set and a ~350-tweet broader set — drawn from top retweets, random tweets, and quote tweets. Two iterative codebooks were developed: a tweet-level one (e.g., “Explains story specifics,” “Presents artifact/event without describing meaning”) and an incident-level one capturing story specifics and meta-frames. Multi-coder consensus with outside arbitration yielded Cohen’s κ of .78 and .73 on key codes. Thematic analysis was paired with quantitative tweet-volume and code-distribution visualizations and case-study interpretation.

Findings

• In 2020 incidents, “Explains story specifics” dominated, reflecting active narrative construction of how alleged fraud occurred.
• In conservative-leaning 2022 incidents, “Presents artifact/event without describing meaning” dominated, suggesting reliance on audience-held prior narratives.
• In Maricopa County 2022, influencers like Kari Lake circulated ambiguous footage and testimonials with minimal explicit framing; audiences supplied the fraud reading themselves, consistent with the 2020 deep story.
• In the USPS/DeJoy case, 2020 discourse built an intersubjective sabotage narrative; 2022 discourse invoked it as common knowledge to mobilize “Fire DeJoy” calls.
• Allusive cues — “Box 3,” “Sharpie,” “Dominion,” “DeJoy” — functioned as triggers activating the deep story without explicit claim-making.
• Deep-story dynamics were largely invisible at the single-tweet level and emerged only at the incident level, validating the dual-codebook approach.

Connections

This paper extends the research program on participatory disinformation and elite-influencer-audience collaboration developed by the same group, most directly Starbird2025-jj on rumoring infrastructures and Marwick2025-ov on networked audience interpretation. Its argument that meaning resides in narrative context rather than discrete claims complements debates over misinformation prevalence and impact in Budak2024-ef and Gonzalez-Bailon2024-rq, and resonates with work on election rumor ecosystems such as DeVerna2025-dl and influencer-driven dynamics in Bollenbacher2026-vz. The deep-story framing also connects to studies of identity- and worldview-anchored belief in Frischlich2025-vn and Renault2025-uh.