Overview
Deep Research is a powerful data enrichment feature that allows you to find information about leads and companies by searching the web. When enabled for an AI Field, our system actively searches the internet, visits relevant web pages, and extracts the information you need.
This document explains what Deep Search can and cannot do, so you can set the right expectations with customers and help them get the most out of this feature.
How Deep Search Works
When you enable Deep Research for an AI Field, here's what happens behind the scenes:
The AI reads your prompt and understands what information you're looking for
It performs web searches using relevant keywords based on the lead/company data
It visits web pages to extract detailed content (like company websites, news, articles, etc.)
It analyzes the content and provides an answer based on what it found
It can iterate up to 5 times, refining searches and exploring different sources
The technology behind Deep Search uses Firecrawl, a specialized web scraping tool designed to extract content from websites and convert it into a format that AI can understand.
What Deep Search CAN Do ✅
Capability | Description |
Search the web | Find information across public websites, news articles, and online directories |
Read website content | Extract text, data, and information from company websites and other public pages |
Follow links | Navigate from one page to another to find relevant information |
Extract structured data | Pull out specific information like company size, industry, technology stack, etc. |
Filter by recency | Focus on recent news or content from specific time periods |
Handle dynamic content | Access pages that require JavaScript to render |
Limitations Compared to ChatGPT Web Search ⚠️
If customers have used ChatGPT's web search feature, they might expect similar behavior. Here are the key differences:
1. Search Scope
Our Deep Search | ChatGPT Web Search |
Searches based on specific lead/company context | Searches the entire open web freely |
Limited to 5 search iterations per field | Can search extensively until it finds the answer |
Focuses on extracting data for a specific question | Can have conversational back-and-forth exploration |
What this means: Deep Search works best when you have specific, well-defined questions. It may not find information if the question is too broad or if the answer requires extensive exploration across many sources.
2. Real-Time Information
Our Deep Search | ChatGPT Web Search |
Fetches data at the time of enrichment | Has real-time access to breaking news, stock prices, live scores |
Cached results may be up to 2 days old | Always fetches the latest information |
Best for stable company information | Better for time-sensitive data |
What this means: Deep Search is excellent for finding company information, job history, technology stacks, and other relatively stable data. For breaking news or rapidly changing information, there may be a slight delay.
3. Content Access
Our Deep Search | ChatGPT Web Search |
Can only access public, unrestricted websites | Has partnerships for premium content access |
May be blocked by anti-bot measures on some sites | Better access to protected content |
Limited to 100 seconds per page timeout | More flexible timeouts |
What this means: Some websites actively block automated access. If a piece of information is only available on a site that blocks scrapers, Deep Search may not be able to retrieve it.
4. Understanding & Reasoning
Our Deep Search | ChatGPT Web Search |
Focused on answering the specific prompt | Can reason across multiple topics conversationally |
Returns structured, consistent output | Can provide nuanced, contextual explanations |
Limited to 5 iterations of search + analysis | Can continue reasoning indefinitely |
What this means: Deep Search is optimized for batch data enrichment, answering the same question across many leads/companies consistently. ChatGPT is better for ad-hoc research where you might need to refine your question based on initial findings.
When to Use Deep Search (Best Use Cases) 🎯
Deep Search works best for:
Finding company information not available in your existing data (website, size, industry, technologies used)
Researching specific facts about a lead's company (recent news, funding rounds, product launches)
Extracting data from known websites (company website, Wikipedia, Google)
Enriching with publicly available information that requires visiting multiple pages
When Deep Search May Struggle ⚡
Consider alternative approaches when:
Information is behind login walls (private LinkedIn data, gated content)
Data changes by the minute (stock prices, live scores, breaking news)
Questions are very open-ended ("Tell me everything about this company")
Sources actively block web scrapers (some financial or news sites)
You need to compare hundreds of sources for a single answer
Setting Customer Expectations 💬
When discussing Deep Search with customers, use these talking points:
✅ DO say:
"Deep Search can find publicly available information about companies and leads"
"It works best with specific questions that have clear, factual answers"
"The AI will search the web and visit relevant pages to find your answer"
"It's optimized for consistent data enrichment across many records"
❌ DON'T promise:
"It can find anything that's on the internet"
"It works exactly like ChatGPT"
"It can access information behind login pages"
"It will always find an answer"
Cost Considerations 💰
Deep Search uses more credits than regular AI Fields because:
Web searches incur additional costs per search query
Page scraping uses credits to extract content from websites
Multiple iterations may be needed to find the right answer
If a customer has many leads/companies to enrich, the costs can add up. Consider:
Using Deep Search only for fields where the information isn't available elsewhere
Testing with a small batch first to estimate total costs
Using more specific prompts to reduce the number of search iterations needed
Troubleshooting Common Issues
Issue | Likely Cause | Solution |
Returns "UNKNOWN" frequently | Information may not be publicly available | Verify the information exists online manually; refine the prompt |
Slow enrichment times | Multiple search iterations + page loading | Expected behavior; Deep Search prioritizes accuracy over speed |
Inconsistent results | Sources may have conflicting information | Add more context to the prompt to narrow down sources |
Blocked on certain sites | Anti-bot measures | Some sites cannot be accessed; try alternative sources |
Summary
Deep Search is a powerful tool for enriching lead and company data with information from the web. It's particularly effective for finding specific, publicly available facts and works best when you provide clear, focused prompts.
While it may not match every capability of consumer-facing AI assistants like ChatGPT, it's optimized for what matters most in sales workflows: consistent, structured data enrichment at scale.