How MatterRank Works
MatterRank gives you control over how search results are evaluated and ranked based on criteria that matter to you. This guide explains the key concepts and how to use the platform effectively.
Building Your Search Engine
Creating a custom search engine is a step-by-step process where each choice affects how your results will be scored and ranked. Here's how to build an engine that matches your specific needs:
Name Your Engine
Choose a name that reflects what you're looking for. The name helps you identify the engine's purpose, like "Academic Research" or "Product Reviews".
Define Your Ranking Criteria
This is the most important step. You'll define what matters to you in search results by creating two types of criteria:
Positive Criteria
Add criteria to increase a result's score:
- "are written by experts or academics"
- "include detailed analysis"
- "cite reliable sources"
Negative Criteria
Add criteria to decrease a result's score:
- "contain excessive ads"
- "have clickbait titles"
- "require paid subscriptions"
Setting Importance
For each criterion, you can set its importance from 25% (Low) to 100% (Critical). This determines how much the criterion affects the final score.
The more specific your criteria, the better MatterRank can understand your preferences and deliver personalized results.
Set Relevance Sensitivity
Adjust the balance between your criteria and traditional search relevance:
Criteria First
Balanced
Query First
Choose where to place emphasis: on your custom criteria (left) or on traditional search relevance (right). The balanced middle setting gives equal weight to both.
Choose Output Format
Select how you want search results displayed:
Standard View
A traditional list of search results with snippets, but each result includes a score and matching criteria tags.
Summary + Results
An AI-generated summary of the top search results with citations, followed by the full list of ranked results.
You can always switch between formats later if you change your mind.
Add Custom Instructions (Optional)
For Summary + Results format, you can provide custom instructions to guide the AI:
Example Instructions
- "Present the information as a step-by-step guide"
- "Compare different viewpoints and highlight areas of consensus"
- "Organize the summary as pros and cons with a recommendation"
Custom instructions only affect how the summary is formatted and presented, not which sources are used.
Configure Advanced Options
AI Query Enhancement
Enabled by default
When enabled, MatterRank will generate additional search queries related to your original query. This helps find relevant results that might not contain your exact search terms.
Impact: Adds 3-5 seconds to search time but improves result quality.
Full Webpage Evaluation
Enabled by default
When enabled, MatterRank will render and analyze the entire webpage for each search result, rather than just evaluating the search snippet.
Impact: Adds 15-30 seconds to search time but significantly improves scoring accuracy. Recommended for most users.
Set Date Range Filters
Limit search results to specific time periods to find the most relevant content based on publication date:
No Filter
Search all content regardless of when it was published. Best for finding evergreen information or when recency doesn't matter.
Recent Content
Focus on content from the past days, weeks, or months. Perfect for news, current events, or rapidly evolving topics.
Custom Date Range
Specify exact start and end dates to search within. Ideal for historical research or finding content from a specific time period.
Date Range Options
Recent Content: Choose from preset timeframes (24 hours, 3 days, 1 week, 2 weeks, 1 month, 3 months, 6 months, or 1 year)
Custom Range: Set optional start and end dates to create a specific time window for your search
Date range filters help you focus on content that's most relevant to your search timeframe, whether you need the latest information or historical context.
How Scoring Works
MatterRank uses LLMs to analyze and score each search result based on your criteria. Here's a detailed look at how the scoring process works:
The Evaluation Process
Content Analysis
For each search result, an LLM analyzes either the full webpage (if Full Webpage Evaluation is enabled) or just the search snippet (if disabled).
Criteria Matching
Each of your criteria is evaluated against the content, and the LLM determines how well the content matches each criterion on a scale from 0 to 100.
Weighted Calculation
Each criterion's score is weighted according to the importance you assigned (25% to 100%), then combined into positive and negative subtotals.
Criteria Score Calculation
The initial criteria score is calculated by taking the weighted positive scores and subtracting the weighted negative scores.
Relevance Balancing
The criteria score is combined with a traditional search relevance score based on your Relevance Sensitivity setting. This produces the final score that determines ranking.
The Scoring Formula
Criteria Score = (Weighted Positive Scores) - (Weighted Negative Scores)
Final Score = (Criteria Score × Sensitivity Factor) + (Relevance Score × (1 - Sensitivity Factor))
Each criterion score is normalized to a 0-100 scale, then multiplied by its importance weight. The combined criteria score is then balanced with traditional search relevance according to your sensitivity setting.
Example Calculation
For a search result with the following scores:
Positive criteria:
"Contains expert analysis" (Score: 85, Weight: 100%) = 85
"Includes citations" (Score: 70, Weight: 75%) = 52.5
Total weighted positive: 137.5
Negative criteria:
"Contains excessive ads" (Score: 60, Weight: 50%) = 30
Total weighted negative: 30
Criteria Score = 137.5 - 30 = 107.5
Relevance Score = 75 (traditional search relevance)
Sensitivity Factor = 0.6 (slightly favors criteria over relevance)
Final Score = (107.5 × 0.6) + (75 × 0.4) = 64.5 + 30 = 94.5
Normalized Score (0-100 scale) = 95
The Ranking Process
After all search results are scored, MatterRank sorts them from highest to lowest score, displaying the results that best match your criteria at the top.
Important Distinction
Unlike traditional filters that completely remove results, MatterRank shows all results but prioritizes the ones that match your criteria. This means you'll never miss potentially useful content that might not perfectly match your preferences.
Relevance Sensitivity
Relevance Sensitivity is a powerful feature that lets you control the balance between your custom criteria and traditional search relevance. Here's how it works:
Understanding Relevance Sensitivity
The Relevance Sensitivity slider determines how much weight is given to your custom criteria versus traditional search relevance (keyword matching, page authority, etc.).
Criteria First
Query First
0
25
50
75
100
Sensitivity Settings Explained
Strongly Criteria First (0)
Results are ranked almost entirely based on how well they match your criteria, with minimal consideration for traditional search relevance. Best when you're more concerned about quality than keyword matching.
Criteria First (25)
Your criteria are prioritized, but query relevance still matters. Good for finding high-quality content related to your topic.
Balanced (50)
Equal weight is given to both your search query and your criteria. A good all-purpose setting for most searches.
Query First (75)
Results closely matching your search query are prioritized, with criteria as secondary factors. Good for finding specific information.
Strongly Query First (100)
Exact matches to your search query are strongly prioritized. Your criteria have minimal impact. Best for precise, targeted searches.
When to Adjust Sensitivity
Tips for Adjusting Sensitivity
Move toward "Criteria First" when quality matters more than exact keyword matching, or when searching for conceptual topics.
Move toward "Query First" when searching for specific facts, exact terms, or when you need precise information.
Start with "Balanced" for most searches, then adjust based on your results.
Custom Instructions
When using the Summary + Results format, you can provide custom instructions to guide how the AI generates your summary. Here's how to make the most of this feature:
What Custom Instructions Do
Custom instructions affect how your summary is structured, organized, and presented. They guide the AI's writing style and format, but don't change which sources are used or how results are scored and ranked.
Important Note
Custom instructions only affect the summary's presentation—the same top-ranked results are always used as sources for the summary, regardless of your instructions.
Effective Instruction Types
Structure Instructions
Guide the format and organization of the summary.
Examples:
"Organize as a step-by-step guide"
"Present as bullet points of key facts"
"Structure as problem/solution pairs"
Content Focus Instructions
Emphasize certain types of information in the summary.
Examples:
"Focus on actionable recommendations"
"Emphasize scientific evidence"
"Highlight different perspectives"
Style Instructions
Guide the tone and language style of the summary.
Examples:
"Write in a beginner-friendly style"
"Use technical language appropriate for experts"
"Keep explanations concise and straightforward"
Comparison Instructions
Request analysis of different positions or options.
Examples:
"Compare different approaches and their tradeoffs"
"Present pros and cons with a recommendation"
"Analyze areas of consensus and disagreement"
Tips for Writing Effective Instructions
Be clear and specific about what you want. "Present as a step-by-step guide" works better than "Make it good."
Combine instruction types for more nuanced summaries: "Present a comparison of approaches in beginner-friendly language."
Keep instructions concise—a few clear sentences work better than lengthy paragraphs.
Match instructions to your query—for a "how to" query, step-by-step instructions work well; for a complex topic, comparison instructions may be better.
Understanding Your Results
MatterRank's search results are designed to give you clear insights into why content was ranked the way it was. Here's how to interpret what you see:
Result Components
85
Score
Each result shows a score from 0-100 in the left margin. Higher scores indicate better matches with your criteria and search query, balanced according to your Relevance Sensitivity setting.
Expert
Cited
Tags
Green tags show which positive criteria the result matches best. These give you a quick overview of the result's strengths.
Details
Details Button
Click the "Details" button to see a breakdown of how the result scored on each of your criteria, with explanations for why it received those scores.
The Details View
Detailed Scoring Breakdown
Positive Criteria
"Contains expert analysis"
85/100
This article was written by a professor with 15+ years of experience in the field and includes detailed technical explanations.
"Includes citations"
70/100
The article includes 8 citations to academic papers and reliable sources, though some claims could use additional support.
Negative Criteria
"Contains excessive ads"
60/100
The page has moderate ad presence with 4 display ads, including one popup, which slightly disrupts the reading experience.
This detailed view helps you understand exactly why a result received its score, with specific explanations for each criterion evaluation.
Summary View
If you selected the "Summary + Results" output format, you'll see an AI-generated summary at the top of your results:
AI Summary Example
Recent studies show that regular exercise significantly improves cognitive function[2]while also reducing stress levels and enhancing sleep quality[5].A minimum of 150 minutes of moderate activity per week appears to be the threshold for measurable benefits[3],though even light activity has been shown to have positive effects compared to sedentary behavior.
About Citations
The numbers in blue boxes are citations that link to specific search results. Click on a citation to jump directly to the referenced result.
Example Engines
Here are some example search engines you can create, along with their criteria and sample searches:
Both Sides
Example Searches:
"affirmative action", "universal basic income", "vaccine mandates"
Finds content that presents multiple perspectives on complex or controversial topics.
Higher Scores For:
- present multiple perspectives
- cite diverse sources
- use neutral language
- are relevant to the search query
Lower Scores For:
- use biased language
- present only one viewpoint
- misrepresent opposing views
- use emotional manipulation
Multiple Topics
Example Searches:
"ai agents, the industrial revolution, the economy", "space exploration, mars, government funding"
Finds content that specifically discusses certain topics or concepts together.
Higher Scores For:
- mention all specified topics
- include in-depth discussion
- provide concrete examples
- analyze relationships between topics
Lower Scores For:
- only mention one topic
- contain superficial mentions
- require payment to access
- lack connections between topics
Well Researched Stories
Example Searches:
"the future of remote work", "climate change solutions", "space tourism progress"
Finds well-written, researched content that isn't social media posts or clickbait.
Higher Scores For:
- contain original reporting
- include expert interviews
- demonstrate quality writing
- cite credible sources
Lower Scores For:
- are social media posts
- are clickbait or listicles
- lack proper sources
- are auto-generated content
Coming Soon
We're continuously improving MatterRank with new features. Here's what's on the horizon:
Video Search
Find videos based on content within their transcripts, not just titles and descriptions. This will allow you to discover videos that discuss specific topics in depth, even if those topics aren't explicitly mentioned in the video metadata.
Example: Search for "quantum computing explanation" and find videos where experts actually explain quantum principles in detail, not just videos with those keywords in the title.
Image Search
Find images based on visual content, style, and quality, not just keywords. A multimodal LLM will analyze the actual content of images to help you discover exactly what you're looking for.
Example: Create an engine that scores higher for "professional photography with natural lighting" and lower for "stock photos with artificial poses," then search for "business meeting" to find authentic-looking images.
Ready to Build Your Custom Search Engine?
Start creating search engines that understand what matters to you and deliver personalized results.