The Influence Tactics Protocol (ITP) is an open standard published by Decipon that scores digital content across 20 categories of psychological manipulation on a 0–100 scale, with evidence per category.
The 20 Categories
Each category measures a specific influence technique on a 1–5 scale.
Categories are grouped into five composite factors. Click a category name
to read its definition.
Emotional Manipulation
(5 categories)
Emotional Triggers
(#1 · weight 0.35)
Emotional Triggers are language patterns that provoke fear, outrage, or guilt without offering proportionate evidence to justify the response. Decipon flags this category when content shows a high density of emotional words ("terrifying", "outrageous", "shameful"), an elevated intensity score across those words, and a heavy fear-or-anger ratio relative to neutral statements. The technique works by activating the limbic system before the reader has time to weigh the underlying claims; once a reader is emotionally engaged, supporting framing tends to be accepted and counter-arguments dismissed as further provocation. Distinguishing this from legitimate emotional reporting requires looking at proportionality: a story about a tragic event will use emotional language, but it will also anchor that language in verified facts and named sources.
Urgent Action Demands
(#2 · weight 0.20)
Urgent Action Demands compress decision time so the reader cannot research, consult, or reflect. The hallmark signals are urgency words ("now", "immediately", "before it's too late"), artificial deadlines (an event closing in hours rather than weeks), and explicit framing of delay as failure. Time pressure short-circuits the cognitive review that would otherwise surface contradictions in a claim or reveal alternative options. Decipon's scoring rises when urgency density combines with deadline pressure and severe time-constraint framing — a pattern common in political mobilization, fundraising appeals, and high-pressure sales copy. Legitimate urgency exists (genuine emergencies, time-bound regulatory windows) and is distinguished by independently verifiable timelines, transparent reasoning for the deadline, and absence of penalty framing for those who pause to verify.
Novelty Overuse
(#3 · weight 0.15)
Novelty Overuse frames events as shocking, unprecedented, or never-before-seen in order to bypass historical comparison. The signals are dense use of novelty words ("unprecedented", "first ever", "shocking"), superlative density ("biggest", "worst", "most extreme"), and the conspicuous absence of historical context that would otherwise contextualize the event. Calling something unprecedented closes off the reader's ability to ask "have we seen this before, and what happened then?" — the most powerful corrective question against panic. Decipon scores this category by weighing novelty-word density, superlative density, and the gap between expected and provided historical context. Authentic novelty (genuine first-of-kind events) is rare and distinguishable by its inclusion of explicit historical comparisons that confirm the lack of precedent.
Emotional Repetition
(#4 · weight 0.15)
Emotional Repetition reinforces a target emotion by returning to the same trigger language, imagery, or framing throughout a piece. The technique exploits a memory-and-priming effect: the more often a reader encounters a charged phrase, the more strongly it shapes their interpretation of subsequent claims. Decipon detects this via emotional-image frequency (count of repeated emotional images per length unit) and phrase-repetition rate (repeated emotional phrases divided by unique emotional phrases). At high score levels, the same outrage word appears more than ten times across a single article, often clustered to drown out neutral framing. Legitimate writing uses emotional language, but varies it — repetition that escalates without new evidence is the hallmark of priming rather than reporting.
Manufactured Outrage
(#5 · weight 0.15)
Manufactured Outrage is anger that appears suddenly, spreads in coordinated form, and is disproportionate to the underlying facts. The detection signals are an outrage-velocity spike (rapid social-signal growth per hour) combined with a high emotional-to-factual ratio inside the content itself. This pattern frequently accompanies astroturfed campaigns, where a small group amplifies anger to make it appear organically widespread. Decipon's scoring rises when both signals co-occur, and rises further when the outrage spreads through identical phrasing across many accounts in a short window. The category is deliberately conservative: legitimate outrage at genuine wrongdoing also spreads quickly, so the score depends not just on velocity but on whether the underlying factual basis grows in proportion to the emotional response.
Suspicious Timing
(3 categories)
Timing Coincidence
(#6 · weight 0.50)
Timing Coincidence flags content whose release aligns suspiciously with other newsworthy events — often at moments designed either to ride a wave of attention or to bury countervailing news. Decipon's scoring weighs the correlation coefficient between release time and known events, then weights more recent coincidences more heavily (a story released hours after a major event scores higher than one released a week later). Strong correlations include opposition research dropping during a political opponent's news cycle, or critical disclosures buried in the news cycle of a larger event. The category is informational, not accusatory: a high score signals that timing is unusual, not that intent is proven. Distinguishing genuine coincidence from strategic timing requires looking at the broader pattern across an outlet.
Financial/Political Gain
(#7 · weight 0.35)
Financial/Political Gain measures whether powerful actors stand to benefit disproportionately from a narrative being widely accepted. The detection method examines beneficiary networks: which corporations, parties, or individuals see capital flows, vote movement, or regulatory changes if the framing succeeds? Stories that consistently align with one group's commercial or political interest — and that omit competing perspectives that would complicate the framing — score high. Decipon does not declare such stories false; many true stories also benefit specific groups. Instead, the category surfaces a question for the reader: who is this story making more powerful, and is that interest disclosed transparently? Disclosure of conflicts (an outlet noting that a quoted expert is paid by a relevant industry) lowers the score.
Historical Parallels
(#8 · weight 0.15)
Historical Parallels flags stories that mirror documented manipulation campaigns from the past — the same playbook, the same targets, the same structural moves. Detection involves matching the content's structural features (emotional opening, named villain, urgent demand, simplistic resolution) against a library of known propaganda patterns. A high score doesn't mean the present story is false: similar patterns can recur in genuine events. It means the structural fingerprint is recognizable, and a reader benefits from asking whether the current claim has the same evidentiary support as the past pattern offered, or whether it relies on the same shortcuts. The weight is intentionally low (0.15 within Suspicious Timing) because pattern matching is suggestive rather than dispositive evidence of manipulation.
Uniform Messaging
(3 categories)
Phrase Repetition
(#9 · weight 0.50)
Phrase Repetition is the appearance of identical or near-identical phrasing across multiple media outlets, accounts, or articles within a short window. It is one of the strongest signals of message coordination — whether through wire-service syndication, talking-point distribution, or astroturfed amplification. Decipon's scoring measures the exact-match and near-match density of distinctive phrases (long n-grams that are unlikely to occur by chance) across the recent media corpus. High scores correspond to "every outlet is using the same words", a pattern that erodes the impression of independent verification even when each outlet may genuinely believe the claim. The category does not penalize legitimate quotation; it focuses on verbatim or near-verbatim framing repeated across sources without attribution to a common origin.
Bandwagon Effect
(#10 · weight 0.25)
Bandwagon Effect is rhetorical pressure to conform with what is portrayed as overwhelming consensus — phrasing like "everyone is doing it", "no serious person disagrees", or "the country has moved on". The detection signals are conformity-language density combined with absence of named dissent or substantive engagement with opposing positions. The technique exploits social-proof heuristics: humans calibrate their beliefs partly by what they perceive others to believe, so framing a position as universal makes individual disagreement feel costly. Decipon flags this category when consensus claims appear without supporting evidence (polls, named groups, vote tallies) or when the named consensus is narrower than the language implies. Legitimate consensus reporting names the source, sample size, and visible dissent.
Rapid Behavior Shifts
(#11 · weight 0.25)
Rapid Behavior Shifts captures the sudden, coordinated adoption of symbols, hashtags, profile pictures, or rhetorical positions across a population without a clear reasoning trigger. Detection weighs the velocity of adoption (how many accounts adopt within a fixed window), the homogeneity of expression (whether they use identical phrasing), and the absence of explanatory bridge content (posts that explain why the new position differs from the previous one). Genuine social shifts also happen quickly during major events, but they typically carry visible reasoning trails — debate, disagreement, partial adoption. A pattern of total adoption with no bridge content suggests external coordination rather than organic change. Decipon's scoring rises when velocity and homogeneity both peak in the same window.
Tribal Division
(3 categories)
Us vs. Them Dynamic
(#12 · weight 0.40)
The Us vs. Them Dynamic frames issues as moral conflicts between an in-group (virtuous, rational, suffering) and an out-group (dangerous, irrational, predatory). Decipon detects this by scanning for in-group and out-group pronouns, moral-charge density on group descriptors, and absence of mixed evaluations (any acknowledgement that the in-group has flaws or the out-group has legitimate concerns). The technique blocks compromise: once a reader has emotionally identified with one group, evidence supporting the other is felt as betrayal rather than information. The category is the highest-weighted within Tribal Division because it is the foundational move on which simplistic narratives and false dilemmas build. Legitimate conflict reporting names actors specifically rather than collapsing them into caricatured groups.
Simplistic Narratives
(#13 · weight 0.35)
Simplistic Narratives reduce complex situations to good-versus-evil framings, with one party fully innocent and the other fully culpable. Detection weighs character flatness (do actors have plural motives or are they single-attribute figures?), causal compression (is a multi-cause situation attributed to one decision?), and absence of contextual qualifiers ("amid economic strain", "after years of failed talks") that would complicate the moral reading. The technique is cognitively rewarding for readers because it resolves ambiguity, but it routinely produces inaccurate predictions and prevents productive policy responses. Decipon's score rises when stories about plural-cause situations (wars, market crashes, political shifts) are flattened to single-actor moral parables. Distinguishing legitimate reporting from this pattern is whether named complications appear and remain unresolved.
False Dilemmas
(#14 · weight 0.25)
False Dilemmas present only two options when more exist, often framed as catastrophic-or-acceptable. The detection method scans for either-or constructions, mutual-exclusivity claims, and the conspicuous absence of intermediate positions that would normally appear in a complex policy debate. The technique works by foreclosing the cognitive space where a reader could imagine a third path, forcing them to align with one of the two presented options. Decipon flags this when binary framings appear in domains where the actual policy space is multi-dimensional — economic policy, public-health responses, foreign-policy postures. Legitimate either-or framings exist (genuine binary votes, yes-or-no propositions), and these are distinguished by explicit acknowledgment of the foreclosed middle ground or by the procedural reason for the binary choice.
Missing Information
(6 categories)
Context Omission
(#15 · weight 0.25)
Context Omission is the systematic exclusion of perspectives, data points, or background facts that would change a reader's interpretation. Detection involves comparing the content's claim set against a baseline of contextual information that comparable reporting on the same event includes — date ranges, actor histories, prior agreements, contradictory data. Omission is harder to detect than commission because the absent material does not appear in the text itself; Decipon relies on signal triangulation across related sources to flag substantive gaps. The category is not about brevity (every article must omit something) but about omissions that systematically favor one reading. Reporting that explicitly names what it leaves out, and why, scores low even when the omissions are substantial.
Authority Overload
(#16 · weight 0.15)
Authority Overload uses dense reference to credentials, titles, and institutional affiliation as a substitute for evidence — quoting "experts" without disclosing methodology, naming institutions without summarizing their findings, or stacking authoritative-sounding sources whose actual claims contradict each other. Detection involves credential density relative to evidence density, and verification of whether named experts actually hold the credentials and positions implied. The technique exploits the ad-verecundiam shortcut: many readers defer to credentialed sources without checking the chain. Decipon's score rises when credentials appear without falsifiable claims, when affiliations imply expertise outside the source's actual domain, or when conflicts of interest are unstated. Legitimate expert citation includes the specific claim, the reasoning, and the source's verifiable scope.
Suppression of Dissent
(#17 · weight 0.15)
Suppression of Dissent occurs when critics or contrary positions are silenced, labeled, or characterized in ways that prevent readers from engaging with their actual arguments — "extremist", "fringe", "discredited" used without supporting reasoning, or critics named only via dismissive epithets. Detection examines whether named opposition appears, whether their substantive claims are summarized fairly enough that a reader could evaluate them, and whether labels are accompanied by definitions or simply applied. The technique short-circuits the dialectical process that produces calibrated belief: when one side's arguments are not even fairly stated, the reader cannot weigh them. Decipon flags this when opposition appears only in dismissed form, distinguishing from cases where opposition is named, summarized, and then refuted on substance.
Cherry-Picked Data
(#18 · weight 0.20)
Cherry-Picked Data is the selective use of statistics, quotes, or examples that support a thesis while omitting closely related evidence that complicates it. Detection methods include reverse-lookup on cited statistics (do they survive when the time window or population is shifted?), comparison against the full dataset behind a cited number, and identification of date or sample boundaries chosen suspiciously near a transition. The category is one of the most common in policy reporting because no story can include all data, but a high score reflects systematic bias rather than reasonable abridgment. Decipon flags this when cited evidence consistently supports one direction, when the boundaries of the cited data are unusual, or when a reader cannot reconstruct what was excluded from the analysis.
Logical Fallacies
(#19 · weight 0.15)
Logical Fallacies are structural errors in reasoning — ad hominem attacks substituting for engagement with arguments, post hoc reasoning treating sequence as causation, straw-manning a position into a weaker form before refuting it. Detection works by parsing the argument structure of the text and matching it against a library of known fallacy patterns, then weighting by frequency and centrality (a fallacy in the thesis sentence weighs more than one in a passing example). The category measures structural quality of reasoning, independent of whether the conclusion happens to be correct. Decipon's score rises when fallacies are dense, when they appear at decision points in the argument, and when alternative reasoning paths are not provided to the reader.
Framing Techniques
(#20 · weight 0.10)
Framing Techniques shape how a reader perceives an event before any evidence is presented — through metaphor selection, loaded vocabulary, scope choice (which actors are visible), and temporal anchoring (which moment the story starts at). Detection involves comparing the framing of a piece against a neutral baseline: would the same set of facts admit a substantively different framing, and would that framing change the implied evaluation? Frames are unavoidable; the category targets framing that constrains interpretation in ways readers don't notice. Decipon scores this lower than other categories (weight 0.10) because identifying a frame as manipulative requires more interpretation than the other signals. The score rises when frames are uniform across coverage and when neutral alternatives appear absent from comparable reporting.
Validation & limitations
Decipon is a working implementation of the ITP. Here is an honest account of what
has been measured, what has not yet been validated, and the formal study designed
to close that gap. We would rather state the gaps plainly than over-claim accuracy
on a tool that scores manipulation.
What has been measured
Deterministic consistency
Repeated analysis of the same content produces identical scores. The ML tier is deterministic, the rule-based heuristics are formula-driven, and the LLM tier runs at temperature 0. Verified across three independent runs on a Statistics Norway consumer-price-index release with caching disabled — zero variance. This is a reliability property (the system is consistent with itself), not a validity claim (that the scores match expert judgement).
ML scorer vs. the SemEval benchmark
Against the SemEval-2020 Task 11 propaganda-technique news corpus, the ML scorer reports a
mean absolute error of
0.368
on a 1–5 scale
(91.8% of predictions
within ±1 of the reference label).
Measures the ML-scored subset of ITP categories that map onto the SemEval propaganda-technique taxonomy — not the full 20-category protocol, and not the five LLM-scored categories.
Calibration spectrum
Selected case studies, ordered by overall ITP score — the ordering matches
editorial expectations (wire-style statistical reporting lowest, partisan
opinion highest):
-
Statistics Norway CPI release (wire-style statistical report)
4/100
-
Russian state media (RT) news article
33/100
-
Partisan opinion blog post
72/100
Disclosed variance
The 15 ML-scored categories are deterministic. The five LLM-scored categories show a coefficient of variation of roughly 10–14% on moderate-scoring content — a known characteristic of the hybrid architecture, disclosed here rather than hidden.
What has not yet been validated
- No peer-reviewed, independently published validation study has been completed — no paper submitted, accepted, or externally reviewed.
- No third party outside Decipon has run the tool against a ground-truth corpus and published the results.
- No inter-annotator agreement study comparing Decipon's scores against multiple human expert raters across a representative corpus.
- No Norwegian-language validation corpus. The ML models were trained primarily on English data; Norwegian scoring relies on a Norwegian spaCy model plus translated vocabulary lists that have not been validated against Norwegian-native expert annotation.
- No published test–retest reliability study beyond the single Statistics Norway example above.
The planned validation study
Empirical Validation of Multi-Dimensional Influence Technique Scoring in Nordic Digital Media
150–200 Norwegian articles stratified across six publisher types and five content types; three expert annotators per article scoring all 20 categories; inter-rater reliability measured with Krippendorff's α; plus a 30-sample Likert evaluation of the Perspectives qualitative output.
Pre-registered targets: Pearson/Spearman ρ ≥ 0.7 against expert consensus, mean absolute error ≤ 0.8, and binarized F1 ≥ 0.65 per category, with full 5×5 confusion matrices and a Norwegian-specific error analysis.
Target venues: NoDaLiDa 2027, CheckThat! Lab at CLEF, Norsk Medietidsskrift, PRIO / CMI policy brief.
Status: Designed in full; not yet funded or started. Decipon is seeking academic co-investigators and funding partners to execute it jointly rather than self-publishing numbers.
Researchers and institutions interested in collaborating can reach us at
hello@decipon.com.