Skip to main content

​
Overview

In Orvi AI, sources are the websites and URLs that AI models rely on when generating answers to your prompts. You can’t directly control an AI model’s internal reasoning — but you can influence what the model reads and trusts by improving your presence on the sources that shape its responses. Orvi AI stores this source-level data in the source_events table, at a per-chat level:
  • domain and url
  • domain_type and url_type
  • whether the URL was an explicit citation is_citation)

​
Sources vs citations (what’s the difference?)

Not all sources are citations — but every citation is a source.
  • Source: a URL the model used or considered during response generation (stored as a row in source_events).
  • Citation: a source that was explicitly referenced in the response text source_events.is_citation = TRUE).
Why this matters:
  • Citations tend to drive traffic (they’re explicitly referenced).
  • Non-cited sources still influence answers (they shape the content even without an explicit reference).

​
Where to find sources in Orvi AI

Use the Sources area in the sidebar to explore source performance and opportunities:
  • Sources (Brand): sources across your brand prompts (supports Brand Topics filtering and Gap Analysis).
  • Products Sources: sources across your product prompts (supports Product + Prompt filtering; Gap Analysis may be limited).

​
How to read the Sources page

Both Sources pages have two views:
  • Domains: aggregate performance for a whole website (e.g. nytimes.com)
  • URLs: performance for specific pages (e.g. a single article or product page)
You’ll also see two charts:
  • Source usage over time (top 5 domains/URLs)
  • Source type chart: a breakdown of citations by domain_type (Domains view) or url_type (URLs view)

​
Filters (focus your analysis)

Depending on which Sources page you’re on, you can filter by:

​
Brand sources filters

  • Brand: your brand or a competitor
  • Date range: preset or custom
  • Models: e.g. ChatGPT vs Perplexity
  • Topics: Brand Topics

​
Product sources filters

  • Products: one or many product URLs
  • Date range: preset or custom
  • Models: AI platforms included in analysis
  • Prompts: product prompt IDs

​
Domain view (Sources → Domains)

Domains help you answer:
  • Which websites influence answers in my category?
  • Are competitors appearing in sources where I’m missing?
  • Which domain categories dominate my space (Editorial vs Corporate vs UGC)?
In the domains table, you’ll typically see:
  • Source: the domain name (e.g. g2.com)
  • Domain Type: Orvi’s classification of the domain (see below)
  • Used %: how often the domain appeared as a source
  • Avg Citations: how often the domain is explicitly cited when it’s used

​
URL view (Sources → URLs)

URLs help you answer:
  • Which specific pages are shaping AI answers?
  • What content formats get used and cited most (articles vs listicles vs product pages)?
  • Which URLs should I target for outreach, updates, partnerships, or content creation?
In the URL table, you’ll typically see:
  • URL: the exact page used as a source
  • URL Type: Orvi’s classification of the page type
  • Used total: how many chats used this URL as a source
  • Avg Citations: how often the URL is explicitly cited when it’s used
  • Updated: the last time Orvi AI fetched/processed the page content
  • Details: what Orvi AI extracted from that page (what the model can “read”)

​
Source Domain Types (domain classification)

Orvi AI assigns a domain_type to help you quickly understand what kind of site you’re looking at: | Domain type | What it usually means | | --- | --- | | CORPORATE | Official company websites and corporate pages | | EDITORIAL | News sites, blogs, online magazines, and publications | | INSTITUTIONAL | Government, educational, and non-profit organizations | | UGC | User-generated content (forums, communities, social platforms) | | REFERENCE | Encyclopedias, documentation, and reference material | | OTHER | Uncategorized or mixed sources |

​
Source URL Types (page-level classification)

Orvi AI assigns a url_type to identify the content format of a specific page: | URL type | What it usually means | | --- | --- | | HOMEPAGE | A site’s main entry page | | CATEGORY PAGE | A page that lists products, articles, or subcategories | | PRODUCT PAGE | A page describing a specific product or SKU | | LISTICLE | “Top 10…” style list content | | HOW-TO GUIDE | Instructional content with steps | | ARTICLE | General editorial content | | OTHER | Anything that doesn’t fit the above |

​
Source metrics (how Orvi AI measures influence)

​
Used %

Used % measures the percentage of chats where a domain (or URL) contributed to the response as a source — even without an explicit citation. 
Used % = (Number of chats where the source appears / Total chats) × 100
This is the key “influence” signal: it tells you which sources AI models rely on behind the scenes.

​
Avg Citations

Avg Citations measures how often a source is explicitly cited when it’s used. At a high level: 
Avg Citations = Average # of citations per chat, among chats where the source was used
Multiple citations in a single response typically indicate the model found that source highly relevant and trustworthy for the question being answered.

​
Total Citations

Total Citations is the total count of citation events in the selected period: 
Total Citations = COUNT(source_events) WHERE is_citation = TRUE

​
Gap Analysis (find high-leverage opportunities)

When available, Gap Analysis helps you prioritize sources where:
  • the source is frequently used (**high Used %**), and
  • other brands are mentioned, but
  • your brand is not mentioned.
In Orvi AI, Gap Score is designed as a simple opportunity score (0 → 1): 
Gap Score = usage_weight × competitor_presence × (1 - your_presence)
Practical interpretation:
  • High Gap Score: the source influences answers and competitors (or other brands) are present — a strong target for outreach or content updates.
  • Gap Score = 0: your brand is already present (or the source is not a meaningful opportunity).

​
Important limitation: AI readability

Orvi AI can only analyze what’s accessible as page content. In practice:
  • AI models (and automated scrapers) rely on HTML-rendered content
  • content hidden behind paywalls or not available in the loaded page may not be usable
  • pages that depend heavily on client-side JavaScript can be harder to interpret consistently

​
How to use source insights (quick playbook)

  • Editorial EDITORIAL): prioritize PR, journalist outreach, and being included in “best of” articles.
  • Corporate CORPORATE): pursue partnerships, integrations, listings, and co-marketing pages.
  • UGC UGC): participate in communities where your category is discussed (authentically and consistently).
  • Reference REFERENCE): improve public documentation and listings; fix outdated or incomplete entries.
  • Your own site: structure pages clearly (headings, summaries, comparisons, unique data) so it’s easy for models to extract and cite.
Last modified on January 25, 2026