Research DIM tracks. Updated weekly.
Published research the Dead Internet Monitor tracks on AI-generated content, automated traffic, and the dead internet thesis. Reverse-chronological. Each entry includes a methodology note describing what DIM does with the work — adopt, contradict, extend, or watch.
Inclusion criteria. Top-tier academic research (peer-reviewed journals, ACL / EMNLP / NeurIPS, university research centres) and high-level industrial independent research where the research output is editorially independent of the publisher's customer base (Pew, Edelman, Reuters Institute, Imperva, NewsGuard, Stanford HAI). Vendor self-benchmarks, marketing whitepapers without published methodology, and press coverage of research are excluded. Borderline industrial sources carry an explicit vendor-framing note. Every entry's publication date is verified from the linked source on the day it's added.
Recency cutoff. Only research from the last 24 months is shown. Older entries are filtered out at render time — measurement of AI-generated content is a fast-moving field and stale references misinform.
Curated by an autonomous weekly Friday trawl across arXiv (cs.CL / cs.SI / cs.CY), ACL / EMNLP / NeurIPS proceedings, Pew, Reuters Institute, Edelman, NewsGuard, Imperva, and Stanford HAI, with editorial review before publication. Suggest an entry: research@deadinternetmonitor.com.
Imperva (Thales) Threat Research — Tim Chang et al. · published 29 Apr 2026 · added 12 May 2026
Imperva's annual industry report on automated traffic, focused this year on the implications of agentic AI for the bot-vs-human distinction. Verified 2026-05-12 against the Imperva resource-library listing and Tim Chang's announcement blog post on imperva.com dated 2026-04-29.
Methodology note
Vendor framing flagged: Imperva sells bot-protection products, so the audience is buyers of those products. The longitudinal data is still the de-facto industry baseline cited by Stanford HAI, Reuters Institute, and academic work — no comparable independent alternative exists at industry scale. The 2026 edition's focus on agentic AI is methodologically relevant to DIM: the same problem (which behaviours classify as bot vs. human) is the one DIM faces on the bot-scoring side. Direct adoption candidate for the Autopsy Matrix bot-consumption axis if the 2026 figures revise the cross-industry blend that Rev 03 currently uses (45%).
bot traffic · agentic AI · industry report
Stanford Institute for Human-Centered AI · published 13 Apr 2026 · added 12 May 2026
Stanford HAI's annual omnibus report on the state of AI — capability benchmarks, deployment trends, training-cost trajectories, regulatory developments, and societal-impact chapters. The 2026 edition runs over 400 pages and is the most-cited single reference work for AI-trend reporting in mainstream press. Date verified via IEEE Spectrum coverage at spectrum.ieee.org/state-of-ai-index-2026 published 2026-04-13, which references the report as already live — Stanford's own page does not display a release date.
Methodology note
DIM's positioning opportunity: the AI Index has no continuous-measurement source for discourse-level AI prevalence — the report cites point-in-time studies but no nightly dataset. A formal data-sharing approach (post Public API #18 ship) is the most plausible academic-citation pathway. Watch for the next edition cadence and any DIM-adjacent measurement projects appearing in the cited-work section.
academic · omnibus · citation pathway
NewsGuard · published 17 Mar 2026 · added 12 May 2026
Running tracker of AI-generated news and information sites identified by NewsGuard's analyst team. As of the verified 2026-03-17 update, NewsGuard had identified 3,006 AI Content Farm news and information websites spanning 16 languages — sites that publish dozens of articles daily, frequently originating false claims about brands, public health, political figures, and celebrities. Date and figure verified 2026-05-12 from the live page header.
Methodology note
Vendor framing flagged: NewsGuard sells news-source ratings to advertisers and platforms, so the AI tracking centre is a vendor product. The analyst methodology is editorially independent of the customer base (the rated sites are not the customers), publicly documented, and cited in Reuters Institute and Pew work. NewsGuard tracks the supply side (sites generating AI content); DIM tracks the consumption side (what proportion of comment-and-discourse content is AI). Cross-reference candidate: when a NewsGuard-flagged site appears in DIM's Guardian-comments or aggregator collectors, it qualifies for an upgraded confidence weighting.
news sites · supply-side · misinformation
Michelle Faverio and Emma Kikuchi (Pew Research Center) · published 12 Mar 2026 · added 12 May 2026
Pew synthesis of five years of survey work on U.S. adult attitudes to AI. Half of U.S. adults are more concerned than excited about increased AI use (up from 37% in 2021); only 10% are more excited. 47% report having heard a lot about AI (June 2025, up 21pp since 2022). Workplace AI use rose from 16% (2024) to 21% (Sep 2025). 64% of teens 13–17 use AI chatbots. Just 9% of U.S. adults get news at least sometimes from AI chatbots. All figures verified 2026-05-12 from the article itself.
Methodology note
The 9% "get news from AI chatbots" figure is the cleanest single audience-side number to pair with DIM's measured AI rates: chatbot-mediated news consumption is still small, but the supply-side AI content prevalence DIM measures is the upstream condition. The concerned-not-excited trend (37% → 50% in four years) is the canonical audience-trust argument for why an independent measurement layer matters. Watch for the next Pew AI tracker cycle.
audience perception · survey · trust
Edelman / Edelman Trust Institute · published 18 Jan 2026 · added 12 May 2026
Annual cross-country survey of trust in institutions. The 2026 edition reports that seven in ten respondents are unwilling to trust those with differing values, approaches to social issues, or information sources. Methodology: 30-minute online interviews conducted 23 October – 18 November 2025, launched at Davos 18 January 2026. Date and key finding verified 2026-05-12 via the PR Newswire release announcing the launch.
Methodology note
Sits alongside Reuters Institute Digital News Report as the canonical audience-trust signal — Edelman covers institutions broadly, Reuters covers media specifically. The unwillingness-to-trust-different-information-sources finding is directly relevant to DIM's editorial framing: if audiences won't trust sources they disagree with, the marginal value of a neutral measurement layer (DIM's positioning) increases. Cross-reference DIM's per-platform AI rates against trust trajectories in subsequent editions.
trust · cross-country survey · institutions
Nic Newman, Richard Fletcher, Craig T. Robertson, Amy Ross Arguedas, Rasmus Kleis Nielsen · published 17 Jun 2025 · added 12 May 2026
Annual cross-country survey of how audiences access, trust, and pay for news, conducted by the Reuters Institute for the Study of Journalism at Oxford. The 2025 edition is the 14th in the series. Title, lead author, and 17 June 2025 publication date verified 2026-05-12 from the report landing page.
Methodology note
Provides the audience-side trust signal DIM's methodology lacks — DIM measures what's on the platforms; Reuters Institute measures what audiences think about what's on the platforms. The trust-decline data is the strongest single argument for why DIM's measurement matters. Future angle: cross-reference DIM's per-platform AI rates against Reuters' per-platform trust rates — does the trust collapse correlate with measured AI saturation? The 2026 edition typically lands in June; watch for it.
trust · audience · cross-country survey