deep
trust
Menu
TECHNOLOGY
How we do it
STEP 1
Continuously gather relevant information
WHAT
An independent factual basis. By leveraging verified disinformation from sources like Correctiv or Politifact deeptrust builds on a reliable, traceable foundation.
HOW
The Harvester. We’ve developed a proprietary, highly-scalable, frequency-adaptive crawler that minimizes the time between publication and analysis.
IMPACT
Constant discovery. Because deeptrust doesn’t just monitor known fake-news sources, it uncovers new and previously hidden disinformation clusters.
STEP 1
Continuously gather relevant information
WHAT
An independent factual basis. By leveraging verified disinformation from sources like Correctiv or Politifact deeptrust builds on a reliable, traceable foundation.
HOW
The Harvester. We’ve developed a proprietary, highly-scalable, frequency-adaptive crawler that minimizes the time between publication and analysis.
IMPACT
Constant discovery. Because deeptrust doesn’t just monitor known fake-news sources, it uncovers new and previously hidden disinformation clusters.
STEP 1
Continuously gather relevant information
WHAT
An independent factual basis. By leveraging verified disinformation from sources like Correctiv or Politifact deeptrust builds on a reliable, traceable foundation.
HOW
The Harvester. We’ve developed a proprietary, highly-scalable, frequency-adaptive crawler that minimizes the time between publication and analysis.
IMPACT
Constant discovery. Because deeptrust doesn’t just monitor known fake-news sources, it uncovers new and previously hidden disinformation clusters.
STEP 1
Continuously gather relevant information
WHAT
An independent factual basis. By leveraging verified disinformation from sources like Correctiv or Politifact deeptrust builds on a reliable, traceable foundation.
HOW
The Harvester. We’ve developed a proprietary, highly-scalable, frequency-adaptive crawler that minimizes the time between publication and analysis.
IMPACT
Constant discovery. Because deeptrust doesn’t just monitor known fake-news sources, it uncovers new and previously hidden disinformation clusters.
STEP 2
Intelligently structure information
WHAT
GAP
Current technologies crumble. The volume of content relevant to disinformation is enormous — tens of millions of traditional news articles per year and hundreds of millions of social media posts per day.
Other search engines fail in terms of speed, precision, or language coverage. LLMs are too slow, costly, and energy-intensive to tackle this at scale.
HOW
A new hyper-fast, hyper-efficient AI. We created deepAI, a new proprietary NLP technology made in Germany. It generates a precise characteristic fingerprint for each document, that is used to precisely index, search, and cluster at scale.
Numbers matter. deepAI processes ~5 million characters or ~1,000 documents per second per CPU core. It forms thematic clusters in real-time (over 800 documents/second/CPU-core). It can manage over 250,000,000 documents with a single one of Apple’s new Mac Studios with 512GB of RAM.
IMPACT
deeptrust handles even the largest information tsunamis. Today, deeptrust analyses all available news in real-time. In the coming months, we will expand coverage to our first social network: Telegram.
Precisely fingerprints and indexes
~1,000 docs/CPU-core/s
thanks to throughput of ~5,000,000 characters per second
Intelligently clusters over
800 docs/CPU-core/s
into a hierarchical thematic structure
STEP 2
Intelligently structure information
WHAT
GAP
Current technologies crumble. The volume of content relevant to disinformation is enormous — tens of millions of traditional news articles per year and hundreds of millions of social media posts per day.
Other search engines fail in terms of speed, precision, or language coverage. LLMs are too slow, costly, and energy-intensive to tackle this at scale.
HOW
A new hyper-fast, hyper-efficient AI. We created deepAI, a new proprietary NLP technology made in Germany. It generates a precise characteristic fingerprint for each document, that is used to precisely index, search, and cluster at scale.
Numbers matter. deepAI processes ~5 million characters or ~1,000 documents per second per CPU core. It forms thematic clusters in real-time (over 800 documents/second/CPU-core). It can manage over 250,000,000 documents with a single one of Apple’s new Mac Studios with 512GB of RAM.
IMPACT
deeptrust handles even the largest information tsunamis. Today, deeptrust analyses all available news in real-time. In the coming months, we will expand coverage to our first social network: Telegram.
Precisely fingerprints and indexes
~1,000 docs/CPU-core/s
thanks to throughput of ~5,000,000 characters per second
Intelligently clusters over
800 docs/CPU-core/s
into a hierarchical thematic structure
STEP 2
Intelligently structure information
WHAT
GAP
Current technologies crumble. The volume of content relevant to disinformation is enormous — tens of millions of traditional news articles per year and hundreds of millions of social media posts per day.
Other search engines fail in terms of speed, precision, or language coverage. LLMs are too slow, costly, and energy-intensive to tackle this at scale.
HOW
A new hyper-fast, hyper-efficient AI. We created deepAI, a new proprietary NLP technology made in Germany. It generates a precise characteristic fingerprint for each document, that is used to precisely index, search, and cluster at scale.
Numbers matter. deepAI processes ~5 million characters or ~1,000 documents per second per CPU core. It forms thematic clusters in real-time (over 800 documents/second/CPU-core). It can manage over 250,000,000 documents with a single one of Apple’s new Mac Studios with 512GB of RAM.
IMPACT
deeptrust handles even the largest information tsunamis. Today, deeptrust analyses all available news in real-time. In the coming months, we will expand coverage to our first social network: Telegram.
Precisely fingerprints and indexes
~1,000 docs/CPU-core/s
thanks to throughput of ~5,000,000 characters per second
Intelligently clusters over
800 docs/CPU-core/s
into a hierarchical thematic structure
STEP 2
Intelligently structure information
WHAT
GAP
Current technologies crumble. The volume of content relevant to disinformation is enormous — tens of millions of traditional news articles per year and hundreds of millions of social media posts per day.
Other search engines fail in terms of speed, precision, or language coverage. LLMs are too slow, costly, and energy-intensive to tackle this at scale.
HOW
A new hyper-fast, hyper-efficient AI. We created deepAI, a new proprietary NLP technology made in Germany. It generates a precise characteristic fingerprint for each document, that is used to precisely index, search, and cluster at scale.
Numbers matter. deepAI processes ~5 million characters or ~1,000 documents per second per CPU core. It forms thematic clusters in real-time (over 800 documents/second/CPU-core). It can manage over 250,000,000 documents with a single one of Apple’s new Mac Studios with 512GB of RAM.
IMPACT
deeptrust handles even the largest information tsunamis. Today, deeptrust analyses all available news in real-time. In the coming months, we will expand coverage to our first social network: Telegram.
Precisely fingerprints and indexes
~1,000 docs/CPU-core/s
thanks to throughput of ~5,000,000 characters per second
Intelligently clusters over
800 docs/CPU-core/s
into a hierarchical thematic structure
STEP 3
Identify disinformation candidates
WHAT
GAP
Instant, precise full-text search is hard. To identify fake news at scale, a system needs full-text search. Yet, while traditional search systems perform well with queries of fewer than 5 words, their quality plummets with longer queries, to say nothing of entire paragraphs or whole articles.
LLM-based comparisons already break down. Even at modest volumes: If LLMs had speeds of 1,000 comparisons/sec, checking 5,000 verified Fake News against 2 million articles would take ~3.8 months.
HOW
deepAI to the rescue. Instead of losing quality with longer queries and full texts, deepAI’s precision increases.
Without compromising on performance. Our system and infrastructure support around 1,000 queries per second, per CPU-core, on up to hundreds of millions of documents.
Conquering complexity. With the precise and fast full-text search, the amount of articles that need to be verified is reduced from many millions to just a few hundred.
IMPACT
Identifying Fake News. Finally possible at scale. deepAI’s breakthrough is what makes combatting disinformation possible in the first place.
Where LLMs would take months, deeptrust needs hours. And it doesn’t stop: After the initial set-up it accelerates to a fully continuous, real-time process.
STEP 3
Identify disinformation candidates
WHAT
GAP
Instant, precise full-text search is hard. To identify fake news at scale, a system needs full-text search. Yet, while traditional search systems perform well with queries of fewer than 5 words, their quality plummets with longer queries, to say nothing of entire paragraphs or whole articles.
LLM-based comparisons already break down. Even at modest volumes: If LLMs had speeds of 1,000 comparisons/sec, checking 5,000 verified Fake News against 2 million articles would take ~3.8 months.
HOW
deepAI to the rescue. Instead of losing quality with longer queries and full texts, deepAI’s precision increases.
Without compromising on performance. Our system and infrastructure support around 1,000 queries per second, per CPU-core, on up to hundreds of millions of documents.
Conquering complexity. With the precise and fast full-text search, the amount of articles that need to be verified is reduced from many millions to just a few hundred.
IMPACT
Identifying Fake News. Finally possible at scale. deepAI’s breakthrough is what makes combatting disinformation possible in the first place.
Where LLMs would take months, deeptrust needs hours. And it doesn’t stop: After the initial set-up it accelerates to a fully continuous, real-time process.
STEP 3
Identify disinformation candidates
WHAT
GAP
Instant, precise full-text search is hard. To identify fake news at scale, a system needs full-text search. Yet, while traditional search systems perform well with queries of fewer than 5 words, their quality plummets with longer queries, to say nothing of entire paragraphs or whole articles.
LLM-based comparisons already break down. Even at modest volumes: If LLMs had speeds of 1,000 comparisons/sec, checking 5,000 verified Fake News against 2 million articles would take ~3.8 months.
HOW
deepAI to the rescue. Instead of losing quality with longer queries and full texts, deepAI’s precision increases.
Without compromising on performance. Our system and infrastructure support around 1,000 queries per second, per CPU-core, on up to hundreds of millions of documents.
Conquering complexity. With the precise and fast full-text search, the amount of articles that need to be verified is reduced from many millions to just a few hundred.
IMPACT
Identifying Fake News. Finally possible at scale. deepAI’s breakthrough is what makes combatting disinformation possible in the first place.
Where LLMs would take months, deeptrust needs hours. And it doesn’t stop: After the initial set-up it accelerates to a fully continuous, real-time process.
STEP 3
Identify disinformation candidates
WHAT
GAP
Instant, precise full-text search is hard. To identify fake news at scale, a system needs full-text search. Yet, while traditional search systems perform well with queries of fewer than 5 words, their quality plummets with longer queries, to say nothing of entire paragraphs or whole articles.
LLM-based comparisons already break down. Even at modest volumes: If LLMs had speeds of 1,000 comparisons/sec, checking 5,000 verified Fake News against 2 million articles would take ~3.8 months.
HOW
deepAI to the rescue. Instead of losing quality with longer queries and full texts, deepAI’s precision increases.
Without compromising on performance. Our system and infrastructure support around 1,000 queries per second, per CPU-core, on up to hundreds of millions of documents.
Conquering complexity. With the precise and fast full-text search, the amount of articles that need to be verified is reduced from many millions to just a few hundred.
IMPACT
Identifying Fake News. Finally possible at scale. deepAI’s breakthrough is what makes combatting disinformation possible in the first place.
Where LLMs would take months, deeptrust needs hours. And it doesn’t stop: After the initial set-up it accelerates to a fully continuous, real-time process.
STEP 4
Validate disinformation
WHAT
HOW
Fine-tuned LLM. We use a specially fine-tuned LLM that distinguishes between reporting on disinformation and its deliberate reproduction and propagation.
IMPACT
Reliability. Only articles that actively spread disinformation are flagged — increasing accuracy, minimizing false positives, and building trust.
STEP 4
Validate disinformation
WHAT
HOW
Fine-tuned LLM. We use a specially fine-tuned LLM that distinguishes between reporting on disinformation and its deliberate reproduction and propagation.
IMPACT
Reliability. Only articles that actively spread disinformation are flagged — increasing accuracy, minimizing false positives, and building trust.
STEP 4
Validate disinformation
WHAT
HOW
Fine-tuned LLM. We use a specially fine-tuned LLM that distinguishes between reporting on disinformation and its deliberate reproduction and propagation.
IMPACT
Reliability. Only articles that actively spread disinformation are flagged — increasing accuracy, minimizing false positives, and building trust.
STEP 4
Validate disinformation
WHAT
HOW
Fine-tuned LLM. We use a specially fine-tuned LLM that distinguishes between reporting on disinformation and its deliberate reproduction and propagation.
IMPACT
Reliability. Only articles that actively spread disinformation are flagged — increasing accuracy, minimizing false positives, and building trust.
STEP 5
Visualize disinformation networks
WHAT
GAP
No statistical basis. To deduce these insights it isn’t enough to have isolated examples. You need a holistic view of most instances, reproductions and how these relate to one another. This wasn’t possible up until now.
HOW
Holistic sampling. deeptrust comprehensively scans the world of news and soon the world of social media for disinformation.
By aggregating all instances of disinformation identified in the previous steps, deeptrust is able to identify and then visualize the hidden network of disinformation.
IMPACT
Actionable transparency. For the first time, the disinformation landscape becomes visible — structured, evidence-based, and actionable for democratic institutions.
STEP 5
Visualize disinformation networks
WHAT
GAP
No statistical basis. To deduce these insights it isn’t enough to have isolated examples. You need a holistic view of most instances, reproductions and how these relate to one another. This wasn’t possible up until now.
HOW
Holistic sampling. deeptrust comprehensively scans the world of news and soon the world of social media for disinformation.
By aggregating all instances of disinformation identified in the previous steps, deeptrust is able to identify and then visualize the hidden network of disinformation.
IMPACT
Actionable transparency. For the first time, the disinformation landscape becomes visible — structured, evidence-based, and actionable for democratic institutions.
STEP 5
Visualize disinformation networks
WHAT
GAP
No statistical basis. To deduce these insights it isn’t enough to have isolated examples. You need a holistic view of most instances, reproductions and how these relate to one another. This wasn’t possible up until now.
HOW
Holistic sampling. deeptrust comprehensively scans the world of news and soon the world of social media for disinformation.
By aggregating all instances of disinformation identified in the previous steps, deeptrust is able to identify and then visualize the hidden network of disinformation.
IMPACT
Actionable transparency. For the first time, the disinformation landscape becomes visible — structured, evidence-based, and actionable for democratic institutions.
STEP 5
Visualize disinformation networks
WHAT
GAP
No statistical basis. To deduce these insights it isn’t enough to have isolated examples. You need a holistic view of most instances, reproductions and how these relate to one another. This wasn’t possible up until now.
HOW
Holistic sampling. deeptrust comprehensively scans the world of news and soon the world of social media for disinformation.
By aggregating all instances of disinformation identified in the previous steps, deeptrust is able to identify and then visualize the hidden network of disinformation.
IMPACT
Actionable transparency. For the first time, the disinformation landscape becomes visible — structured, evidence-based, and actionable for democratic institutions.
STEP 6
Strengthen the early warning system
WHAT
GAP
No scalable pattern recognition. For an early warning system to work, it needs to have all relevant data instantly available and be able to precisely assess the risk of any new given piece of information.
Up until now neither the scalabilty and speed, nor the precise pattern recognition have been brought together at the same time to combat disinformation.
HOW
A comprehensive, intelligent system. deeptrust continuously expands its pattern recognition to assess whether new pieces of information have known markers of disinformation. It factors in topical matches, source history as well as how quickly the given topic is growing.
IMPACT
A partner in truth. deeptrust shares these insights with fact-checkers and subsequently with authorities and journalists to enable them to strengthen and secure our democracy.
STEP 6
Strengthen the early warning system
WHAT
GAP
No scalable pattern recognition. For an early warning system to work, it needs to have all relevant data instantly available and be able to precisely assess the risk of any new given piece of information.
Up until now neither the scalability and speed, nor the precise pattern recognition have been brought together at the same time to combat disinformation.
HOW
A comprehensive, intelligent system. deeptrust continuously expands its pattern recognition to assess whether new pieces of information have known markers of disinformation. It factors in topical matches, source history as well as how quickly the given topic is growing.
IMPACT
A partner in truth. deeptrust shares these insights with fact-checkers and subsequently with authorities and journalists to enable them to strengthen and secure our democracy.
STEP 6
Strengthen the early warning system
WHAT
GAP
No scalable pattern recognition. For an early warning system to work, it needs to have all relevant data instantly available and be able to precisely assess the risk of any new given piece of information.
Up until now neither the scalabilty and speed, nor the precise pattern recognition have been brought together at the same time to combat disinformation.
HOW
A comprehensive, intelligent system. deeptrust continuously expands its pattern recognition to assess whether new pieces of information have known markers of disinformation. It factors in topical matches, source history as well as how quickly the given topic is growing.
IMPACT
A partner in truth. deeptrust shares these insights with fact-checkers and subsequently with authorities and journalists to enable them to strengthen and secure our democracy.
STEP 6
Strengthen the early warning system
WHAT
GAP
No scalable pattern recognition. For an early warning system to work, it needs to have all relevant data instantly available and be able to precisely assess the risk of any new given piece of information.
Up until now neither the scalability and speed, nor the precise pattern recognition have been brought together at the same time to combat disinformation.
HOW
A comprehensive, intelligent system. deeptrust continuously expands its pattern recognition to assess whether new pieces of information have known markers of disinformation. It factors in topical matches, source history as well as how quickly the given topic is growing.
IMPACT
A partner in truth. deeptrust shares these insights with fact-checkers and subsequently with authorities and journalists to enable them to strengthen and secure our democracy.