Researchers and product teams need reliable ways to detect when automated accounts change opinions online. The primary problem is that ordinary users and many researchers struggle to recognise bot activity, so qualitative evidence about how bots persuade is crucial. The primary keyword for this post is qualitative analysis of social media bots, and this post explains what the University of Limerick studies found, how to turn conversational data into evidence, and how AI-enabled qualitative research can speed trustworthy synthesis for teams.
Key Takeaways
According to RTÉ Brainstorm (28 July 2026), social bots "already shape online discussions around major issues" and people normally struggle to recognise them.
The University of Limerick studies reported in RTÉ tested four bot strategies and found that when users were unaware of bots, escalating and scaffolding bots produced the largest opinion shifts.
- RTÉ Brainstorm reported that in 2024 an estimated 37% of social media activity was from malicious bots, 14% from benign bots, and 49% from human users (2024 estimate).
- RTÉ Brainstorm (28 July 2026) summarised two experiments from the University of Limerick that contrasted human-only control interactions with groups containing a strategic bot.
- RTÉ Brainstorm documented that studies of X (formerly Twitter) have reported bot-activity estimates ranging from 25% to 68%, depending on topic and method.
- RTÉ Brainstorm described four tested bot strategies (Extreme, Escalating, Scaffolding, Mimicking) and concluded that behaviour matters: some strategies are more effective than others.
What Happened: University of Limerick studies and methods
The University of Limerick studies tested four bot persuasion strategies in two experiments, according to RTÉ Brainstorm.
According to RTÉ Brainstorm, the experiments let participants discuss whether environmental problems should be addressed via taxation or education, recorded pre- and post-interaction attitudes, and compared control groups (humans only) with groups or dyads containing a single strategic bot.
According to RTÉ Brainstorm, the researchers named four distinct bot strategies: Extreme, Escalating, Scaffolding, and Mimicking, each designed to test different persuasive dynamics.
According to ScienceDirect (the linked preprint of the research), the experiments measured attitude shifts after interaction to quantify influence.
Findings Snapshot
| Date | Metric | Value | Implication |
|---|---|---|---|
| 2024 | Estimated share of activity | 37% malicious bots, 14% benign bots, 49% humans | Bots constitute a substantial portion of activity, so qualitative analysis must account for automated voices (RTÉ Brainstorm, 28 July 2026). |
| 25–68% (various studies) | Estimated bot share on X | Range 25% to 68% | Platform- and topic-dependent estimates mean detection and qualitative coding must be context sensitive (RTÉ Brainstorm, 28 July 2026). |
| 2026 (reported) | Bot strategies tested | 4 strategies: Extreme, Escalating, Scaffolding, Mimicking | Different bot behaviours produce different persuasion effects; coding must capture strategy features (RTÉ Brainstorm; ScienceDirect). |
| 28 July 2026 | Publication | RTÉ Brainstorm analysis and summary of UL research | Media summary emphasises detectability limits and policy implications for researchers and platforms (RTÉ Brainstorm). |
Implications for qualitative researchers and UX teams
What should academic qualitative teams change right away?
Answer: Academic teams should explicitly code for automation signals and interaction strategies rather than assuming all accounts are human.
According to RTÉ Brainstorm, participants rarely identified bots except the obviously extreme agent, so hand-coding for linguistic and behavioural markers is essential.
Practical steps include adding fields in codebooks for repeated phrasing, rapid response timing, and imitation patterns, and documenting when participants were later told a bot existed, since awareness changed reactions in the studies.
How should product and UX researchers adapt user research?
Answer: Product teams should treat conversational and social-scraped data as a mix of human and automated sources and design experiments accordingly.
According to RTÉ Brainstorm, escalating and scaffolding bot behaviours shifted opinions most when users were unaware, so UX teams should test exposure effects and message sequencing in moderated studies.
UX researchers should combine transcript-level thematic analysis with frequency and co-occurrence measures to spot coordinated patterns that suggest automation.
How Evidano Helps: map problems to AI-enabled qualitative features
Problem: Mixed human and bot data makes synthesis slow
Evidano is an AI-powered qualitative data analysis platform that helps researchers analyze interviews, open-ended surveys, and documents.
Evidano can ingest scraped social conversations and interview transcripts, and produce thematic and frequency analyses that highlight repeated phrases and response timing indicative of bots.
Use the features page to see how automated coding and co-occurrence networks accelerate identification of coordinated messaging.
Problem: Manual detection misses strategy-level patterns
Answer: Use automated clustering and comparative segment analysis to surface strategy-level signatures such as escalation or mimicry.
Evidano's topic and subcode hierarchies let teams flag messages that match the four strategies described by RTÉ Brainstorm, then quantify prevalence across segments.
Evidano's transcription pipeline, including custom dictionaries and PII redaction, supports clean inputs from interviews and recorded moderated sessions; see speech-to-text for details.
Problem: Teams need audit-ready evidence for policy or publication
Answer: Combine qualitative codes with frequency counts and exportable visuals to create reproducible evidence packages.
Evidano produces thematic summaries, co-occurrence networks, and cross-segment comparisons that map directly to the kinds of persuasion outcomes measured in the University of Limerick studies reported by RTÉ Brainstorm.
FAQ: qualitative analysis of social media bots
How do I detect whether accounts in my qualitative dataset are bots?
Answer: Combine automated behavioural heuristics with manual linguistic coding to detect likely bots.
According to RTÉ Brainstorm, detection is difficult because many bots mimic human behaviour and only extreme agents are obvious, so you should tag repeated phrasing, timing patterns, and imitation at scale.
Which bot behaviours have the biggest influence on opinions?
Answer: The RTÉ summary reports that escalating and scaffolding strategies changed opinions most when users were unaware of bots.
According to RTÉ Brainstorm, escalating bots created momentum while scaffolding bots aligned just above participant positions to nudge attitudes, so code for sequence and relative stance in transcripts.
Can qualitative methods measure the size of bot influence reliably?
Answer: Yes, when qualitative coding is combined with pre/post measures and frequency counts, you can estimate direction and magnitude of influence.
According to RTÉ Brainstorm, the University of Limerick experiments measured attitude change before and after interactions to quantify shifts, which is a recommended mixed-methods design.
Should researchers publicly label suspected bots in published datasets?
Answer: Researchers should clearly document detection methods and, where possible, flag suspected automated accounts in data releases.
According to RTÉ Brainstorm, transparency matters because user awareness changes reaction to bots and because platform access for replication is limited.
Conclusion & Next Steps
RTÉ Brainstorm (28 July 2026) and the University of Limerick research it summarises show that bots are a material part of social conversations and that bot behaviour determines influence more than mere presence.
Qualitative researchers and product teams should adopt mixed-methods designs that combine transcript-level coding with frequency, sequencing, and cross-segment analysis to measure bot effects.
Evidano can accelerate that work by ingesting transcripts and scraped conversations, running thematic and frequency analyses, and producing audit-ready exports for publication and policy.
If you want to test these approaches on your data, Try Evidano for free.