Skip to main content
AI model responses are the raw material powering all your metrics. Understanding how a scan is structured, how AI evaluates your brand against competitors, and where it gathers its information is essential for optimizing your presence in generative search (GEO - Generative Engine Optimization).

​
Anatomy of a Scan

In the main AI Scans table and the interactive side panel (Response Drawer), each entry breaks down an individual query sent to a generative engine.

Overview of AI scans in Wudlet

Each row and expanded view contains the following core elements:
  • Scan Date: The local date and time when the prompt was executed and the model’s response was captured.
  • Model: The generative engine that processed the query (for example: ChatGPT, Claude, Perplexity, Gemini, Grok, DeepSeek).
  • Prompt Text: The contextual query or question submitted to the AI model, designed to mirror real user search intent in your market.
  • Response Text: The textual content generated by the model. Includes an arrow (→) shortcut to open the full interactive viewer.
  • Brand Mentioned: A boolean indicator showing whether your brand or domain was identified in the response.
  • Mention Count: The total number of times your brand appears in the response body.
  • Sentiment: The semantic tone with which the model refers to your brand:
    • Positive: Direct recommendations, praise for features, added value, or industry leadership.
    • Neutral: Purely factual mentions or appearances in lists without explicit bias.
    • Negative: Criticism of limitations, pricing warnings, support complaints, or unfavorable remarks.
  • Competitors: Avatars and badges of competing brands detected in the same response, allowing you to assess direct competition for that query.
  • Created At: Timestamp when the record was saved in your project database.

​
Citations vs Sources

One of the most critical distinctions when analyzing web-enabled generative AI engines is the difference between Explicit Citations and Browsing Sources.
In the interactive side panel (Response Drawer), Wudlet automatically categorizes both types of links so you know which pages the AI reads and which ones it chooses to explicitly cite and recommend.

​
1. Citations

These are the links and bibliographic references embedded directly within the text of the model’s response.
  • They indicate that the model used specific data from that URL to formulate that sentence, recommendation, or comparison table.
  • They have the highest potential click-through rate when human users interact with the AI.

​
2. Sources

These are all the URLs that the model accessed and indexed in real time during its preliminary browsing phase before synthesizing the answer.
  • They include contextual sources consulted by web search tools (web_search_tool_result), even if the model chose not to cite them directly in the final output text.
  • Analyzing your sources helps you understand which domains are educating the model on your industry, even if they don’t visibly link to you.

​
Platform-Specific AI Behavior

Every artificial intelligence engine features distinct architectures, search policies, and citation heuristics:
  • Dynamic web browsing: Depending on prompt intent, ChatGPT may answer directly from its internal training weights or trigger a real-time web search (live web search).
  • Source behavior: If the search tool is not triggered, the response will not display web citations or sources. This is normal behavior and does not indicate a system error.
  • Advertising and sponsored content: In ChatGPT web environments, Wudlet automatically detects whether the query triggered commercial mentions or sponsored cards (Sponsored).
  • Search-first architecture: Virtually 100% of Perplexity responses perform an immediate web search.
  • High volume of sources: Typically consults a large number of background URLs, but displays a more selective number of explicit in-text citations.
  • High volatility: Because it indexes the web in real time, sources and rankings on Perplexity can fluctuate more frequently than in closed-parameter models.
  • Structured search tool calling: Claude utilizes native tool-use blocks (web_search).
  • Strict separation: Cleanly differentiates between pages it inspected (sources) and pages it explicitly linked using citation tags (citations) in the prose.
  • Analytical tone: Tends to produce longer, highly detailed explanatory responses and in-depth comparisons.
  • Gemini: Taps directly into Google’s search index to retrieve fresh, locally contextualized information.
  • Grok: Accesses real-time social context and breaking trends via X / the live web.
  • DeepSeek: Emphasizes deductive logical reasoning, making it ideal for comparing technical specs across software or services.

​
Position Rankings & Advanced Metrics

When an AI recommends products, tools, or solutions to an open question (for example: “What are the best marketing platforms in 2026?”), the order in which it mentions you is decisive.

​
Brand Rank

Indicates where your brand appears in the recommendation ranking:
  • Position #1: Peak authority. The AI considers your brand the benchmark and preferred option for that use case.
  • Position #2 to #3: Strong contender. Generally considered part of the leading group.
  • Position #4+ or absent: Secondary mention or lagging behind. Indicates that competitors hold greater authority in the documents indexed by the model.

​
Share of Voice (SoV)

Measures your brand’s visibility percentage relative to the total mentions detected in the response: Share of Voice (%)=(Brand MentionsBrand Mentions+∑Competitor Mentions)×100\text{Share of Voice (\%)} = \left( \frac{\text{Brand Mentions}}{\text{Brand Mentions} + \sum \text{Competitor Mentions}} \right) \times 100

​
Opportunity Gaps

Wudlet flags a scan as an Opportunity Gap (is_gap = true) when:
  1. The query caused the model to recommend one or more competitors from your tracking list.
  2. Your brand was not mentioned at all.
Filter or search by opportunity gaps to identify queries where competitors are winning visibility and create targeted content to close that gap.

​
Chat & Interactive Drawer Features

Clicking the arrow button (→) on any row in the scans table opens the Response Drawer:

Exact Chat Replay

Inspect both the exact prompt sent and the complete response returned by the model, preserving original formatting and line breaks.

Visual Brand Highlighting

The viewer automatically highlights your brand mentions in a distinctive color alongside competitor mentions so you can assess context instantly.

Direct Citations Panel

A dedicated list with outbound links to all URLs cited directly within the model’s text.

Browsing Sources Panel

A list of additional background web sources consulted by the model before formulating the response.

​
Tips for Filtering and Navigating Results

  1. Keyword Search: Use the top search bar to find mentions of specific product features, competitor names, or prompt keywords.
  2. Time Range & Model Filtering: Use the global filters at the top to compare how your visibility has evolved across ChatGPT versus Perplexity or Claude over time.
  3. Dynamic Pagination: Adjust the table view to 25, 50, or 100 rows per page to audit large batches of scans efficiently.