ConspiracyGraph Agent
What it does
ConspiracyGraph Agent is a narrative and claim relationship agent for fact checked misinformation.
It is not a normal fact check bot that only answers “true” or “false”. Instead, it explains how a user supplied claim connects to known fact checked claims, broader narrative frames, topics, entities, publishers, and evidence.
This lets the agent answer questions like:
Is this claim a variation of an older claim?
What narrative frame does it use?
Which entities connect it to other claims?
Which fact checks and publishers are connected to it?
Dataset and why a graph fits
The dataset is derived from the Google Data Commons Fact Check dataset, which aggregates structured fact check data from fact checking organizations worldwide.
Dataset source: Google Data Commons Fact Check dataset
https://datacommons.org
Dataset description:
https://datacommons.org/factcheck/download
The raw data is based on the schema.org/ClaimReview standard and contains structured fields such as:
claimReviewed
reviewRating
ratingExplanation
fact check URL
publisher
publication date
The data comes from publishers such as PolitiFact, FactCheck.org, BBC Reality Check, Washington Post, Logically Facts, FACTLY, Vera Files, Youturn, and The Healthy Indian Project.
The important point is that this is not a random conspiracy dataset. It is real world fact checked data. Each claim is linked to its source, verdict, publisher, and explanation.
A table can store fact checks.
A graph can explain how they are connected.
Data pipeline
I used a two step pipeline to turn the raw fact check data into a context rich graph.
First, the original fact check claims were cleaned, normalized, and scored. This included normalizing claim text, publishers, verdicts, dates, and stable IDs, then filtering for conspiracy related signals such as false flags, hidden elites, microchips, vaccine danger, climate hoax, election fraud, and global governance narratives.
Second, the selected candidate claims were embedded and semantically clustered. Each cluster was then enriched with structured information such as a canonical claim, specific topics, broad topics, narrative frames, entities, and a short explanation of why the claim is relevant to conspiracy narratives.
Graph schema
Node labels
| Label | Count | Meaning |
|---|---|---|
| ClaimVariant | 242 | Exact checked claim wording |
| Claim | 218 | Canonical claim cluster |
| FactCheck | 239 | Fact check article |
| Publisher | 28 | Fact checking organization |
| TheoryTopic | 205 | Specific topic labels |
| BroadTopic | 78 | Higher level topics |
| NarrativeFrame | 43 | Recurring narrative patterns |
| Entity | 229 | People, organizations, places, concepts, events |
Relationship types
| Relationship | Count | Meaning |
|---|---|---|
EXPRESSES |
268 | ClaimVariant to canonical Claim |
CHECKS |
242 | FactCheck to checked ClaimVariant |
PUBLISHED_BY |
239 | FactCheck to Publisher |
BELONGS_TO |
519 | Claim to TheoryTopic |
HAS_BROAD_TOPIC |
474 | Claim to BroadTopic |
USES_FRAME |
399 | Claim to NarrativeFrame |
MENTIONS |
295 | Claim to Entity |
Overall graph size:
1,282 nodes
2,436 relationships
Why this is more than semantic search
A semantic search system can find text that sounds similar.
ConspiracyGraph Agent goes further: it explains why claims are related.
For example, a claim like:
“COVID vaccines contain microchips”
can connect to:
Secret Technology
Population Control
Health and Vaccines
Bill Gates
Pfizer
COVID-19 vaccine
fact checks from multiple publishers
Agent tools
The Aura Agent uses similarity search plus Cypher templates.
1. Similar Claim Finder
Type: Similarity Search
Searches over ClaimVariant.embedding using the vector index claim_variant_embedding_idx.
Purpose:
User claim → closest known ClaimVariant
2. Claim Narrative Context
Type: Cypher Template
Input: variant_id
Returns the matched variant, canonical claim, frames, topics, entities, same-cluster variants, fact checks, publishers, verdicts, and URLs.
3. Related Narrative Claims
Type: Cypher Template
Input: variant_id
Finds broader narrative relatives through shared NarrativeFrame, BroadTopic, and Entity nodes.
4. Fact Check Evidence
Type: Cypher Template
Input: variant_id
Returns publisher, verdict, original verdict, explanation, date, fact check URL, and publisher URL.
5. Frame Explorer
Type: Cypher Template
Input: frame_name
Explores frames such as:
Secret Technology
Poisoned Medicine
Fake Event
Hidden Agenda
Election Theft
Population Control
Cover Up
Manipulated Science
6. Entity Bridge Explorer
Type: Cypher Template
Input: entity_name
Shows how entities such as Bill Gates, NASA, CDC, FDA, Pfizer, United Nations, World Economic Forum, or Donald Trump connect different claims and narratives.
7. Graph Statistics Fallback
Type: Text2Cypher
Used only for open ended statistics, counts, rankings, or aggregations not covered by the other tools.
Agent in action
Example 1: COVID vaccine danger claims
User question:
I keep seeing claims that COVID vaccines are dangerous. What patterns do you see?
The agent matched this to a claim variant about COVID vaccines being released as a depopulation measure.
It then connected the matched claim to:
Canonical claim:
The COVID-19 vaccine is part of a depopulation agenda.
Narrative frames:
Population Control
Poisoned Medicine
Topics:
depopulation
vaccine side effects
vaccine safety
Health and Medicine
Related claims:
COVID vaccines cause infertility
COVID vaccines are part of a depopulation agenda
COVID vaccines contain microchips for tracking and control
COVID vaccines alter human DNA
Fact check evidence:
PolitiFact
Reuters Fact Check
Lead Stories
The important part is that the agent does not only return one debunk. It shows a broader vaccine harm narrative and the evidence connected to it.
Example 2: Climate change and manipulated science
User question:
Someone claimed that climate change is a hoax and that NASA is misleading the public. What does the graph show?
The agent matched this to:
“NASA admits that man-made climate change is a hoax!”
It connected the claim to:
Narrative frame:
Manipulated Science
Broad topic:
Climate and Environment
Entity:
NASA
Fact check evidence:
Vishvas News
Verdict: False
Fact check URL
This shows how a specific claim connects to a broader pattern of institutional distrust and manipulated science narratives.
Why this agent is useful
Fact checks are usually published as isolated articles. That helps verify one claim, but it does not explain how misinformation narratives repeat and evolve.
ConspiracyGraph Agent turns isolated fact checks into a connected claim graph.
It helps answer:
- Have we seen this claim before?
- Is this a variation of a known fact checked claim?
- What broader narrative does it use?
- Which entities appear across related claims?
- Which publishers checked it?
- What evidence is available?
This can support e.g. journalists, researchers and fact checkers.
Final takeaway
ConspiracyGraph Agent is a context building agent for misinformation narratives.
It does not just answer:
- Is this claim false?
It answers:
- How does this claim fit into a broader narrative?
- Which known fact checked claims is it similar to?
- Which frames, topics, and entities connect it to other claims?
- Which fact checks, verdicts, and publishers are linked to the claim?
That is why this works well as a graph powered Aura Agent