The Search pipeline
You enable the Search pill and send your message
When you toggle the Search pill and send, ZeroTwo receives your message along with a signal to fetch live web data. The active model, conversation history, selected focus mode, and any other active tools are all factored into the request.
ZeroTwo formulates search queries
ZeroTwo analyzes your message and generates one or more targeted search queries — often phrased differently from your original prompt to maximize result quality. For a prompt like “What happened at the Fed meeting last week?”, ZeroTwo might search for
Federal Reserve FOMC meeting [current month and year] decision rate.Top results are fetched and read
ZeroTwo retrieves the top results for each query. For each result, it reads the full page content — not just the meta description or snippet shown in a traditional search engine. The number of pages read depends on query complexity and your plan.
ZeroTwo synthesizes the results
Retrieved content is combined with your original question and passed through the model. ZeroTwo synthesizes a response that weaves together findings from multiple sources into a coherent, readable answer.
Sources panel is populated
Every web page that contributed to the response is listed in the Sources panel on the right side of the chat. Each entry shows the source number, page title, domain, and full URL.
Auto-triggered Search
ZeroTwo monitors every message for signals that real-time data is needed — even when you have not manually enabled the Search pill. It may auto-trigger Search for:- Questions about today’s date, current time, or “latest” anything
- Queries about ongoing events, live scores, or breaking news
- Requests involving recent product releases, software versions, or prices
- Anything with explicitly time-sensitive language (“this week,” “right now,” “as of today”)
Web Search vs. Deep Research
Both tools retrieve web content, but they are optimized for very different use cases.| Feature | Web Search | Deep Research |
|---|---|---|
| Number of queries | 1–5 targeted queries | Dozens of iterative queries |
| Time to complete | Seconds | 2–10 minutes |
| Output format | Inline response with citations | Structured Canvas document (report) |
| Source depth | Top results per query, full page content read | Comprehensive crawl via Firecrawl |
| User interaction | Single message → response | Review and approve a research plan first |
| Best use case | Quick lookups, current events, fact-checking | Reports, analysis, literature reviews, due diligence |
| Free plan | 1/day | 1/day |
What ZeroTwo reads from each page
When ZeroTwo fetches a search result, it reads the full text content of the page — not just the meta description. This means:- Long articles and reports are fully processed
- Tables and structured data within the page are parsed
- Paywalled or JavaScript-rendered content may not be accessible
- PDFs linked from search results are readable when using PDF focus mode
If you specifically need PDF content, switch your focus mode to PDF before sending. See Focus Modes for details.
Query reformulation
ZeroTwo does not submit your exact message as a search query. It reformulates your question into effective search queries designed to retrieve the most relevant results. You can sometimes see the queries ZeroTwo used in the Sources panel header. To influence query formulation, be explicit in your prompt:- “Search for the latest research on X published in 2024 or 2025”
- “Find the official documentation for Y from the vendor’s website”
- “Look up recent news about Z from reputable news outlets only”
- “Search specifically for
[exact phrase]”
Limitations
- ZeroTwo cannot access pages that require a login or are behind a paywall
- Very recently published pages (within the last few hours) may not yet be indexed
- Search accuracy depends on the quality of available web content for the topic
- ZeroTwo may occasionally misrepresent a source — always verify critical information by clicking through to the original URL