The “AI Researcher” LeadTables Data Module

The “AI Researcher” Data Module lets you ask an AI to research something on the web for each row in your LeadTable and save the answers back as real columns.

This is especially useful for ICP filtering and lead qualification because it turns “web research” into structured data you can sort, filter, segment, and use in downstream workflows.

How It Works:

• Write a prompt describing what you want researched for each lead (using your existing row data as context).
• Optionally restrict which websites the research can pull from (allowlist/denylist).
• Define the specific fields you want back (your “Output Structure”), then run a small test.
• Deploy it to all rows matching your current filters and review results in your new columns.

---

Why It’s Awesome:

• Automates web research: Get “research-grade” answers for every lead without manual Googling.
• Turns research into filterable columns: Save answers as structured fields so you can slice your list instantly (instead of reading notes).
• Great for ICP filtering: Score fit, detect red flags, classify industries, validate claims, and prioritize the best leads.
• More auditable than typical AI outputs: Optionally store explanations (and the source list) so you can QA and build trust.
• Scales from quick tests to full-table rollouts: Run it on 1–10 rows first, iterate, then deploy to everyone when it looks good.
• Downstream unlock: Use the saved outputs for routing, personalization, segmentation, and follow-on enrichments.

---

Output Fields:

• Research Sources (json) — A list of the sources the AI used (URLs/titles/snippets) so you can audit where answers came from.
• your_key (text | number | boolean) — One answer per output you define (saved as a real column you can filter/sort).
• your_key_explanation (text) — Optional. A short explanation of why the AI gave that answer (useful for QA and trust).

---

Configuration:

• AI Prompt: What you want researched and how you want the AI to think about each lead (use your row’s existing data as context).
• Response Output Structure: The list of fields you want back (label, key, data type, and per-field guidance).
• Intelligence Level: Higher intelligence usually means better reasoning, but higher cost.
• Search Result Context Size: How much source material the AI considers before answering (larger tends to improve coverage, but costs more).
• Search Domain Whitelist/Blacklist: Restrict or exclude specific domains/URLs for higher trust, less noise, or tighter focus.

---

Data Quality Considerations:

• Web content changes over time, so reruns can produce different answers.
• Some sites may be blocked, require logins, or limit automated access; that can reduce coverage or quality.
• Sources can conflict; the AI may summarize imperfectly or miss nuance, especially on ambiguous topics.
• Domain rules are best-effort and in the past have been unreliable (e.g. I’d set a domain rule and still get results that ignore the rule)
• I personally find AI research to be not as ideal as scraping a homepage first and then analyzing it with a regular AI prompt Data Module for anything that requires “definitely true data markers about the lead.” Often I’ve noticed that Perplexity might pull from search results about different companies with similar names, etc., and produce incorrect output.

---

Current Data Provider:

This module is powered by Perplexity.

---

Misc Tips:

• Start with “Test on 1” or “Test on 10,” then refine your prompt and output fields before deploying broadly.
• Keep outputs focused. Too many fields (or overly long guidance) can reduce quality.
• Turn on explanations while you’re QA’ing a new setup, then disable them later if you don’t need the extra detail. (Though I anecdotally believe keeping them enabled improves the model’s output quality)
• If you care about trust/accuracy, restrict sources to a small set of high-quality domains and review Research Sources on a sample — make sure you test that Perplexity’s heeding your domain rules though!