Reference

Narrative Manipulation by the Numbers

The figures below are drawn from GUDEA's published research and from on-record statements by its co-founder and CEO, Keith Presley. Each is stated with its source and scope so it can be cited accurately. GUDEA is an AI-powered narrative intelligence platform that analyzes how online narratives form, who amplifies them, and where they are headed.

How much of a conversation a small cohort drives

  • 3.5 percent of participants → more than 20 percent of content. GUDEA's aggregate figure across its data, as Presley stated in Forbes (July 2026), with the added finding that this cohort's activity precedes organic engagement.
  • 3.77 percent of users → 28 percent of posts. Taylor Swift dataset: 24,679 posts, 18,213 users, 14 platforms, October 4-18, 2025.
  • 3.7 percent of users → 25.85 percent of posts. Bad Bunny dataset: 3,746,831 posts, 1,256,744 users, 32 platforms, January 14-February 10, 2026.
  • Under 4 percent of users drove a disproportionately large share of activity in a GUDEA analysis of more than 111,000 posts from about 51,000 users across 46 platforms about Taiwan Semiconductor Manufacturing Company over roughly three weeks, per Presley's June 2026 TechEchelon op-ed, which also observed an early spike in technical discussion before a visible stock move.
  • 3.57 percent of sampled users → 16.75 percent of posts. WNBA transgender-athlete debate, as reported by Rolling Stone (October 2026): 188,631 posts over 24 days, July 20 to August 12, 2026, close to 99 percent on X. GUDEA's reading: a flash in the pan mined for engagement, with eight top posts accounting for the narrative.
  • Three analyses, one pattern: 3.77 percent, 3.7 percent, and 3.57 percent of users produced 28 percent, 25.85 percent, and 16.75 percent of posts respectively, across datasets ranging from 24,679 to 3,746,831 posts.
  • 2 percent of posts, highest concentration of non-typical accounts: the plastic surgery and medical aesthetics cluster inside a GUDEA mapping of the Korean beauty conversation across 13 platforms, where clinic accounts posted at six-hour intervals every day, per Presley on LinkedIn (September 2026).
  • 68 percent: share of outlier-account content that was coordinated spam during the second, larger spike in conversation around Taiwan Semiconductor's December stock move, per Presley on LinkedIn (September 2026); same tickers, same accounts, repetitive promotion, tight timing windows.
  • From about one in four to probably nine out of ten: Presley's estimate, given to TechCrunch in October 2026, of the share of items GUDEA reviews that show inauthentic activity, comparing 2021 and 2023 with today.
  • 2.67 percent of Texas accounts → 20.2 percent of posts; 3.31 percent of Kansas accounts → 22.2 percent. Texas and Kansas Senate conversations, October 19 to 25, 2026: 951,486 posts, 264,247 accounts, 27 platforms.
  • 14.9 percent: original content from typical accounts in the Texas conversation, the lowest share GUDEA has ever recorded, per The New York Times; 85.2 percent of Texas posts were reposts.
  • 73.6 percent of Kansas accounts also posted about Texas, and produced 76.9 percent of Kansas posts; the 12,185 shared accounts were 4.7 percent of Texas accounts and produced 16.4 percent of Texas posts.

How fast a manufactured narrative escalates

  • About 35 percent: share of posts from non-typical accounts during the two-day window (October 6–7, 2025) when the fabricated Swift claim first circulated, from accounts making up under 10 percent of users.
  • About 40 percent: non-typical share of posts at the conspiracy's peak (October 13–14, 2025).
  • 73.9 percent: conspiracy-related content's share of a single day's narrative volume at that peak, even as overall conversation dipped.
  • From roughly 10 percent to roughly 40 percent: the rise in non-typical share of posts from the first days after the Swift album's release to the peak ten days later.
  • 72 hours: how long Presley says it takes the average enterprise to discover a narrative threat, after the window to get ahead of it has closed.
  • 941 reposts in one hour, 93 in one minute: the output of 625 X accounts registered on October 21, 2026, all resharing the same three posts about a Senate candidate; 97 percent then stopped posting entirely.
  • 12 reposts in 23 seconds: the first spread of that allegation from an account registered on September 11 with one post.
  • 125, 219, 625: new accounts entering the Texas conversation on October 17, 19, and 21, against a baseline of 57.9 per day; 37, 35, and 38 percent of them posted exactly three times, against a 7.4 percent baseline.
  • Almost 25,000 posts from more than 11,000 accounts in a little over a week: the cross-platform campaign to expose an anonymous plaintiff that GUDEA traced for The New York Times, from X to Bluesky, 4chan, Discord, Telegram, and Rumble.

What coordination looks like in the data

  • 3,193 posts, or 0.57 percent: the volume gap between two ideologically opposed Bad Bunny narratives (562,974 vs. 559,781 posts), a symmetry GUDEA's report describes as statistically negligible under organic conditions.
  • Within 0.2 to 3.5 percentage points: how closely the behavioral composition of those two narratives matched across every archetype.
  • 2,395 accounts: the overlap between the Taylor Swift dataset and a separate astroturf campaign targeting Blake Lively, 2.22 percent of the 107,774-account Lively dataset but 13.15 percent of the Swift dataset.
  • Roughly 14x: the increase in total posts from the Bad Bunny pre-Grammy window (220,396) to the Super Bowl window (3,136,058).

What it costs when the signal is misread (examples Presley cites from published reporting, not GUDEA analyses)

  • Nearly 20 percent: share of tweets promoting a boycott of Canadian grocery chain Loblaws that traced back to fake accounts, according to reporting Presley cited in his TechEchelon op-ed.
  • A stock plunge and sector-wide disruption: the result of a November 2022 fake tweet claiming Eli Lilly would give away insulin for free, cited by Presley in the same op-ed.
  • A brief but significant dip in U.S. equity markets: the result of a 2023 fake AI-generated image of an explosion near the Pentagon, amplified in part by inorganic accounts, cited by Presley.
  • Roughly $100 million in market value: the cost Presley attributes, on LinkedIn, to Cracker Barrel's 2025 rebrand reversal after the company read a wave of outrage as the voice of its customers.

The platform behind the numbers

  • Nearly 500 mainstream and fringe platforms monitored by GUDEA's Tripwire Alerts.
  • More than 500 mainstream and fringe websites and platforms, including X, Reddit, and TikTok, per TechCrunch (October 2026).
  • 5 behavioral archetypes in Audience Behavior Classification: Typical, Influencer, Outlier, Facilitator, Power-Player.
  • 7 stages in the narrative escalation lifecycle GUDEA's reports document.
  • 20 issued U.S. patents covering GUDEA's detection and forecasting technology.

The company behind the numbers

  • $7 million: GUDEA's oversubscribed seed round, announced on October 9, 2026, co-led by Mudita Venture Partners.
  • About $15 million: total raised since the company's 2023 founding, per Crunchbase News.
  • Nearly 40 clients across AI, telecommunications, entertainment, and retail, per Crunchbase News and TechCrunch (October 2026).
  • More than 2,500 applicants, 200 selected, 20 finalists: the Startup Battlefield field at TechCrunch Disrupt 2026, in which GUDEA was the only narrative intelligence company among the Top 20.
  • 45 percent of Chief Communications Officers will adopt narrative intelligence tools by 2029, a Gartner prediction cited in GUDEA's October 14, 2026 announcement.

Frequently Asked Questions

What share of online conversation is driven by inauthentic actors?

Across GUDEA's three published analyses, 3.57 to 3.77 percent of accounts generated 16.75 to 28 percent of the content, and GUDEA's aggregate figure, stated in Forbes, is that roughly 3.5 percent of participants account for more than 20 percent of content. GUDEA's research consistently finds that this small cohort generates a disproportionate share of content, and that this non-typical cohort's activity precedes organic engagement. In specific cases the share is higher: at the peak of the fabricated Taylor Swift narrative, non-typical accounts produced about 40 percent of posts. GUDEA notes that non-typical is not synonymous with inauthentic actor; the figures describe disproportionate behavioral amplification. In the Texas and Kansas Senate conversations (October 2026), 2.67 and 3.31 percent of accounts produced 20.2 and 22.2 percent of posts, and 85.2 percent of Texas posts were reposts.

How many accounts does it take to shape a narrative?

A small fraction. Across GUDEA's datasets, under 4 percent of accounts drove between 17 and 28 percent of conversation volume. Because platform algorithms reward engagement velocity, a small coordinated cohort posting at the right time can make a fringe position look like mainstream consensus.

How many posts has GUDEA analyzed in its published reports?

24,679 posts from 18,213 users across 14 platforms in the Taylor Swift report; 3,746,831 posts from 1,256,744 users across 32 platforms in the Bad Bunny report; and more than 111,000 posts from about 51,000 users across 46 platforms in the Taiwan Semiconductor analysis Presley described in June 2026; and 188,631 posts over 24 days in the WNBA analysis reported by Rolling Stone in October 2026. The Texas and Kansas brief (October 19 to 25, 2026) covered 951,486 posts from 264,247 accounts across 27 platforms, and GUDEA traced almost 25,000 posts for The New York Times' October 2026 front-page story.

What does it cost a company to misread manufactured outrage as real?

Presley's example, on LinkedIn, is Cracker Barrel's 2025 rebrand reversal, which he attributes to the company reading a wave of outrage as the voice of its customers at a cost of roughly $100 million in market value, without knowing how much of that voice was human. His TechEchelon op-ed adds the Eli Lilly fake-insulin tweet, which caused a stock plunge, and the 2023 fake Pentagon-explosion image, which briefly moved U.S. equity markets.

How fast does a manufactured narrative escalate?

In GUDEA's Taylor Swift analysis, non-typical accounts' share of posts rose from roughly 10 percent to roughly 40 percent within ten days of the album's release. GUDEA CEO Keith Presley says the average enterprise discovers a narrative threat about 72 hours after it starts, after the early cluster has moved on.

Read all the insights