How It Works
The Seekr Score
The Seekr Score helps you find reliable news articles by rating the content in your search results. We call it a Seekr Score which is an information reliability rating. Read on to learn more...
Seekr holds seven patents in natural language processing (#humblebrag), meaning not only that our search technology is impressive, but also that our information evaluation capabilities are a revolutionary step for online content as well. Seekr uses core semantic analysis and natural language processing technology to find and score content. Say what?
Think of it like this: Semantic analysis is how computers understand language—the sentiment, emotions, and context. This means it can do things like identify the difference between a fact statement and a judgment. For example, comparative words like “smarter” indicate a judgment, which is not a fact. This is how computers can derive meaning from the data.
We use data scientists and a diverse range of expert journalists who train models to recognize specific Score Factors: Subjectivity, Personal Attack, Clickbait, Title Exaggeration and the Presence/Absence of a Byline. (So, we find the right people for the job—top notch qualifications.)
Additionally, our models are rigorously tested, and human reviewed before going live using blind studies and statistical matching. This systemic approach ensures that content is evaluated equally, and scoring is consistent.
Once generated, The Seekr Score is sorted into five levels of reliability, from very low to very high. For transparency, an explanation for each level is provided below. The Seekr Score is a number between 1 and 100 to showcase the granularity of the evaluation.
Article uses straight reporting, meaning it tells us what happened with very little influential, predictive, or opinionated terminology associated with the Seekr Score Factors. It may even be a primary source for other publications, which may be evident in the byline. When articles don’t stray from the most essential information (who, what, when), it’s less likely that our AI will detect negative score factors. This is what an A+ looks like! (Note: Opinion pieces are expected to have some subjectivity. That does impact the score; however, less severely than other article types such as beat reporting.)
Article uses straight reporting, but our AI may be detecting some analysis terminology via the Score Factors . It is common in reporting to use facts to support some assumptions, opinions, predictions to understand the why. Our AI picks up on that but has not detected levels of analysis that indicate the author is straying far from the topic. It’s also possible the problem is simply that there is no author! A missing byline can impact an otherwise high-quality article. Not an A+ but predicted as still quite informative and reliable.
This article is sprinkled with generally informative useful statements; however, it’s not considered an authoritative source. Here’s why: it could lack a byline and/or our AI is detecting levels of analysis terminology that could indicate attempts to persuade, a bias, or deviation from the main topic as shown by our Score Factors. It’s not likely a terrible read, but you may want to consider cross-referencing with another article to ensure you’re getting the bigger picture. Not too shabby—but keep your wits about you.
Rather than to inform, this article’s intent may be to misrepresent, draw clicks, or provoke a response. Reader may still find useful information in the article; however, the way the statements are supported is considered sloppy. This article has high amounts of terminology flagged by Seekr Score Factors that demonstrate low-level analysis which are often boldly represented in a headline and throughout the article. Reader be warned.
Rather than to inform, the intent of this article is rather extreme in its attempt to support an agenda. The tactics are deceptive with little concern of presenting valid, straight reporting. In short, you deserve much better.
* Seekr AI can detect direct statements but cannot determine the veracity (level of truth) behind those statements. For example, “The sky turned green on Wednesday, April 5th, 2023” is a statement that does not contain any opinionated, predictive, or influential language, however it is not true. Seekr does not tell readers what is true or not true.
Reliable information means that facts are presented (The who, what, when) and if analysis is offered (The why), it does not stray from the facts. Reliable analysis can be presented as opinions and predictions reasonably supported by facts; that’s fair. But when the analysis strays from fact statements (Subjectivity), or uses exaggerated, sensational, or insulting language (Title Exaggeration, Clickbait, Personal Attack), the analysis becomes less reliable.
Information is also less reliable if the source does not identify an author; it lacks accountability. (No Byline)
Strict journalistic principles are the foundation for the Seekr Score Factors. An absence of these factors means the information is of higher quality and likely more reliable.
The higher presence of a Score Factor that the AI detects in an article, the lower the Seekr Score will be. Byline is an exception; if there is no Byline (named author), the overall score will be lower.
When clicking on the Seekr Score icon next to an article, a window reveals The Score, which is a number between 1 – 100. You can also learn how the Score Factors impacted the overall score.
Here’s what the AI is looking for and why it’s important:
| Score Factor | How the AI works | Why it matters for reliability | Special circumstances |
|---|---|---|---|
| Title Exaggeration | Seekr AI looks for headlines that portray something as more important than it is or hyperbolizes a claim. | Credible headlines use superlatives only when they can be proven or reference a study or explicit opinion. | |
| Subjectivity | The AI looks for thoughts, beliefs, or attitudes that go beyond just reporting facts. Words that contain sentiment (“outstanding”), that cannot be measured (“quickly”) or comparative words (“smarter”) are a few examples. | News tells us what happened. Anything beyond facts is analysis or input. Personal assessments, speculation, and predictions are subjective and should be assessed with critical thought rather than taken at face value. | Article types, like opinion pieces, by nature, contain opinions making them subjective. The penalty for subjectivity is less impactful when detected in Op-Eds vs Beat Reporting. |
| Clickbait | The AI looks for sensational promises. “You won’t believe...” is typically followed by content that under-delivers: The ol’ bait and switch. Clickbait appeals to our curiosity and teases the reader with inaccurate and sometimes inappropriate articles. | Credible headlines should prioritize reporting the truth behind the real topic as accurately as possible rather than prioritizing clicks. | |
| Personal Attack | The AI looks for Ad Hominem arguments, which show language that attacks the character, motive, or other characteristics of a person rather than their actual argument. | A credible argument focuses on evidence without having to resort to name-calling or other hateful language. | Interview transcripts or articles describing an interview may contain personal attacks. The Seekr Score does not penalize the article in this scenario as it would in typical news articles because it is quoting an individual. |
| Byline | A byline names the author(s) of an article. The AI looks for the absence of a byline. | When there is no byline it means the author is not named and the article lacks accountability. | Some news organizations removed bylines, like for Russia-Ukraine coverage, to protect journalists. The Seekr Score does not penalize these instances for not having a byline as the intent is not to deceive but to protect. |
| SCORE FACTOR | Tittle Exaggeration |
|---|---|
| HOW THE AI WORKS | Seekr AI looks for headlines that portray something as more important than it is or hyperbolizes a claim. |
| WHY IT MATTERS FOR RELIABILITY | Credible headlines use superlatives only when they can be proven or reference a study or explicit opinion. |
| SCORE FACTOR | Subjectivity |
| HOW THE AI WORKS | The AI looks for thoughts, beliefs, or attitudes that go beyond just reporting facts. Words that contain sentiment (“outstanding”), that cannot be measured (“quickly”) or comparative words (“smarter”) are a few examples. |
| WHY IT MATTERS FOR RELIABILITY | News tells us what happened. Anything beyond facts is analysis or input. Personal assessments, speculation, and predictions are subjective and should be assessed with critical thought rather than taken at face value. |
| SPECIAL CIRCUMSTANCES | Article types, like opinion pieces, by nature, contain opinions making them subjective. The penalty for subjectivity is less impactful when detected in Op-Eds vs Beat Reporting. |
| SCORE FACTOR | Clickbait |
| HOW THE AI WORKS | The AI looks for sensational promises. “You won’t believe...” is typically followed by content that under-delivers: The ol’ bait and switch. Clickbait appeals to our curiosity and teases the reader with inaccurate and sometimes inappropriate articles. |
| WHY IT MATTERS FOR RELIABILITY | Credible headlines should prioritize reporting the truth behind the real topic as accurately as possible rather than prioritizing clicks. |
| SCORE FACTOR | Personal Attack |
| HOW THE AI WORKS | The AI looks for Ad Hominem arguments, which show language that attacks the character, motive, or other characteristics of a person rather than their actual argument. |
| WHY IT MATTERS FOR RELIABILITY | A credible argument focuses on evidence without having to resort to name-calling or other hateful language. |
| SPECIAL CIRCUMSTANCES | Interview transcripts or articles describing an interview may contain personal attacks. The Seekr Score does not penalize the article in this scenario as it would in typical news articles because it is quoting an individual. |
| SCORE FACTOR | Byline |
| HOW THE AI WORKS | A byline names the author(s) of an article. The AI looks for the absence of a byline. |
| WHY IT MATTERS FOR RELIABILITY | When there is no byline it means the author is not named and the article lacks accountability. |
| SPECIAL CIRCUMSTANCES | Some news organizations removed bylines, like for Russia-Ukraine coverage, to protect journalists. The Seekr Score does not penalize these instances for not having a byline as the intent is not to deceive but to protect. |
- The Seekr Score does not represent likes and popularity. For example, a high score does not mean it is news everyone is reading
- The Seekr Score does not mean the information is censored or informed based on user profiles.
- The Seekr Score is not a rating for an entire website or source. A score is assigned for an individual article.
- The Seekr Score does not use a “pay-to-play” model. This means that news outlets and advertisers cannot pay to receive a high score for their content to be prioritized. Seekr does not score content for financial gain but to keep people better informed.
- The Seekr Score does not indicate if information is true or false.
Political Lean
Seekr uses advanced machine learning to analyze articles for their political lean.
When Seekr AI detects a particular political stance in an article, a Political Lean icon is shown next to the Seekr Score.
The Political Lean labels are based on the standard U.S. classification system for political ideology: Left, Left Center, Center, Right Center, or Right.
Choose articles across the political spectrum to get the full picture with the Political Lean Filter. You can select more than one Political Lean at a time. By choosing to include Non-Political results, it delivers the broadest view.
As with the Seekr Score, Seekr AI also extracts and deeply analyzes a text for Political Lean. Using advanced machine learning and natural language processing, it detects the stance of an article around political topics. These may cover familiar hot-button topics like taxation, reproductive rights, climate change, immigration, social injustice, etc.
It is easy to state that a particular website always leans left or right; however, to reduce bias and overgeneralization, Seekr AI analyzes the exact content of an article, not the website itself. Sometimes, opinion pieces can depart from the usual bias of a publication.
Seekr AI also detects expressions, keywords, and semantics strongly associated with a particular political lean. A current example of this in action is with the word ‘Woke,’ which appears in right-leaning content used to describe left-leaning agendas. As expressions come and go in our culture, their associated perceptions change as well; therefore, Seekr continually tests and improves its model.
The Seekr model is based on machine learning, which analyzes a variety of high-quality, broad, and diverse datasets. The datasets originate from the public domain—such as congressional documents — through to our vast multiple-annotated datasets. These use a precise, unbiased, and guided data creation process performed by a diverse group of annotators (they’re real people) from across the political spectrum.
Machine-learning systems must be explainable and transparent to be trusted. Checks and balances are in place; The Political Lean Indicator must prove why an article exhibits a certain lean. An example of this evidence could be that the article has taken a “pro-life” or a “pro-choice” position. Seekr employs ongoing automated and manual testing to ensure accurate lean detection and prevent bias or inaccurate drifts in the model. We accomplish this through techniques such as:
- Comprehensive testing and blind studies: we ask a wide range of people to evaluate the predictions of our systems by “hiding” information that may lead to bias or opinion (such as the originating domain or author); that way, we only consider and analyze the content of the article.
- Specific machine-learning architectures that will not result in “black-box” results but instead outputs that are easily explainable.
- Meticulous testing models for biases against the entire library of articles in our index.
Non-Political Articles: Seekr analyzes every news article; those where the AI detects political content will be scored through the Political Lean Indicator. Seekr does not attempt to detect political lean within non-political articles. For example, an NFL-related article will not be analyzed because it’s a sports story and is unlikely to contain any political topics. The same is true for most entertainment, technology, business, and health articles.
Short Text: If the text extracted from an article is too short, it will not be analyzed for a political lean. To do so would lead to predicting a political lean without enough confidence to be correct or trusted. Reasons for short text can include:
- Short breaking stories with no details (i.e., Breaking: Gunshots heard on Capitol Hill, awaiting details)
- Paywall-protected websites that require a subscription
- Pages that are too complex to extract text automatically
- Primarily a video or image slideshow
- Non-English sites
- Accompanying tweets, videos, and photos