Skip to main content

Influence Tactics Analysis Results

28
Influence Tactics Score
out of 100
68% confidence
Moderate manipulation indicators. Some persuasion patterns present.
Optimized for English content.
Analyzed Content
X (Twitter)

Robert Youssef on X

Penn State just published research testing 5 politeness levels on ChatGPT-4o with 50 questions: Very Polite: 80.8% accuracy Polite: 81.4% Neutral: 82.2% Rude: 82.8% Very Rude: 84.8% Prompts like "Hey gofer, figure this out" beat "Would you be so kind?" by 4 percentage points. pic.twitter.com/5j4qwID

Posted by Robert Youssef
View original →

Perspectives

Blue Team's perspective is stronger due to verifiable academic source, precise falsifiable data, and visual evidence, outweighing Red Team's concerns about simplified framing and omissions typical of social media summaries. The content shares a counterintuitive but empirically grounded finding with mild sensationalism, aligning more with legitimate science communication than manipulation.

Key Points

  • Both teams agree the content is based on a real Penn State study with specific, testable metrics (e.g., 50 questions, ChatGPT-4o, accuracy percentages), reducing fabrication risk.
  • Red Team identifies simplification and competitive language ('beat') as mild manipulation patterns, while Blue Team views this as neutral, concise sharing without emotional or urgent appeals.
  • Omissions like statistical significance or limitations are noted by Red but contextualized by Blue as standard for social media, not deceptive intent.
  • No evidence of overgeneralization beyond the study's scope (one model), supporting Blue's transparency assessment.
  • Areas of agreement: Counterintuitive narrative drives engagement without outrage or division.

Further Investigation

  • Access full Penn State study for p-values, confidence intervals, error bars, methodology details, and limitations to assess if 4-point difference is statistically significant.
  • Verify visual evidence (pic.twitter.com/5j4qwIDyD6) matches study charts and check for any alterations.
  • Review study replications or independent tests on other models (e.g., GPT-4, Claude) to evaluate generalizability beyond ChatGPT-4o.
  • Examine post author's history for patterns in science sharing vs. sensationalism.
  • Compare raw data averages across politeness levels for trend confirmation and sample adequacy (n=50 per level?).

Analysis Factors

Confidence
False Dilemmas 2/5
No binary extremes forced; lists five levels without presenting only two options.
Us vs. Them Dynamic 3/5
Subtle us-vs-them in contrasting polite ('Would you be so kind?') vs rude ('Hey gofer') prompts, implying rudeness superior for AI users.
Simplistic Narratives 3/5
Frames as straightforward good-vs-bad: polite prompts lower accuracy (80.8%), rude higher (84.8%), oversimplifying study nuances.
Timing Coincidence 1/5
Study published October 2025, tweet January 9, 2026; no suspicious correlation with January 10-13 events like protests or AI policy news, appearing as organic AI tip resurfacing.
Historical Parallels 1/5
No resemblance to propaganda techniques or campaigns; factual summary lacks emotional appeals, divisions, or patterns seen in state-sponsored AI disinformation.
Financial/Political Gain 2/5
Academic Penn State study via seed grants benefits researchers broadly; poster @rryssf_ (prompt expert) gains minor visibility, but no clear companies, politicians, or funding ties pushing narrative.
Bandwagon Effect 1/5
No claims of widespread agreement or 'everyone knows'; presents isolated study results without implying consensus.
Rapid Behavior Shifts 2/5
Mild viral pickup in early January 2026 posts with moderate likes/views; no urgent pressure, bots, or manufactured trends forcing opinion shifts.
Phrase Repetition 3/5
Verbatim repetition of exact phrasing, numbers, and example 'Hey gofer...' across X posts by @rryssf_ and others, suggesting shared viral sourcing rather than diverse framing.
Logical Fallacies 3/5
Implies causation from correlation ('beat ... by 4 percentage points') without explaining why (e.g., brevity vs tone), potential hasty generalization from one model/task.
Authority Overload 1/5
Cites 'Penn State just published research' as sole authority without questionable experts or overload.
Cherry-Picked Data 4/5
Highlights accuracy rise from 'Very Polite: 80.8%' to 'Very Rude: 84.8%' and specific example, ignoring broader methodology, p-values, or non-significant aspects.
Framing Techniques 4/5
Biased language like 'beat' personifies competition between prompts; 'just published' sensationalizes old October 2025 study.
Suppression of Dissent 1/5
No mention of critics or alternative views; silent on study's limitations or counter-findings.
Context Omission 4/5
Omits full context like 250 total prompts (50 questions x5 levels x10 runs), statistical caveats, warnings on uncivil language harming norms, and older models' opposite trends.
Novelty Overuse 2/5
Highlights counterintuitive result 'Prompts like "Hey gofer, figure this out" beat "Would you be so kind?" by 4 percentage points' as mildly shocking, but avoids excessive 'unprecedented' claims.
Emotional Repetition 1/5
No repetition of emotional words or phrases; straightforward list of accuracy percentages without emphatic looping.
Manufactured Outrage 2/5
No outrage expressed or evoked; factual presentation of data like 'Rude: 82.8% Very Rude: 84.8%' connected directly to study without exaggeration.
Urgent Action Demands 1/5
No demands for immediate action, sharing, or behavior change; simply summarizes study results like 'Very Polite: 80.8% accuracy' without pressure.
Emotional Triggers 2/5
Content reports findings neutrally without fear, outrage, or guilt language; mild surprise implied in rude prompts outperforming polite ones, but no strong emotional triggers.

Identified Techniques

Loaded Language Doubt Name Calling, Labeling Black-and-White Fallacy Whataboutism, Straw Men, Red Herring

What to Watch For

This messaging appears coordinated. Look for independent sources with different framing.
This content frames an 'us vs. them' narrative. Consider perspectives from 'the other side'.
Key context may be missing. What questions does this content NOT answer?

This content shows some manipulation indicators. Consider the source and verify key claims.

Was this analysis helpful?
Share this analysis
Analyze Something Else