✦Katzilla Ask—grounded chat

Ask anything.
Get cited answers.

Natural-language queries across every Katzilla Data source. Powered by Claude, grounded in primary government data, forbidden from making anything up. Every claim in the answer traces back to a source URL.

Try Ask See benchmarks

300+

sources reachable

98.4%

citation accuracy*

made-up sources

* against the citecheck benchmark, Q1 2026

katzilla ask

claude · opus

you›which pharma companies had Class I FDA recalls in 2025?

ask›

Based on FDA's recall database, five pharmaceutical manufacturers¹ initiated Class I recalls in 2025:

•Pfizer — Zithromax tablets (lot AZ-2025-0412)²
•Novartis — Voltaren Gel (3 lots)³
•Teva — Losartan API (cross-contamination)⁴
•Sandoz — Cefdinir suspension (6 lots)⁵
•Fresenius Kabi — Heparin 5,000 U/mL vials⁶

Total: 17 Class I events across 5 companies⁶.

// sources

[1]FDA Recall Enterprise System · retrieved 2026-04-17
[2]pfizer.com/news/press-release/…
[3]novartis.com/news/…
[4]tevapharm.com/news/…
[5]fda.gov/safety/recalls/…
[6]fda.gov/safety/recalls/…

[01/How it works]

// step 01

You ask

Type a question in plain English. No query language, no endpoint selection — Ask figures out which of the 300+ sources to hit.

// step 02

Claude queries the corpus

Not the open web. Katzilla Data. Every call is a live pull from the primary source — SEC, FDA, NIH, Congress, Fed Reg, courts — with citation metadata attached.

// step 03

Answer returns cited

Inline footnote markers link to the exact source record used. Hover to see source_url, retrieved_at, content_hash. Audit-ready by default.

[02/Ask anything]

Questions that usually take
a week of Googling.

// compliance

›Did any EPA rule finalized this quarter affect SIC code 2821?

// due diligence

›Has Acme Corp ever had an FDA Class I or II recall?

// financial

›Summarize material changes in AAPL's last four 10-Q filings.

// legal

›Which circuit courts cited Chevron deference in 2025?

// research

›Active Phase III trials for Alzheimer's in the Northeast.

// policy

›Bills introduced on AI regulation in the 118th Congress.

[03/Where accuracy comes from]

Not a chatbot
with good prompts.

Generic LLMs hallucinate filings, cases, and statutes because they were trained on text, not sources. Ask fetches the actual record every time. If the corpus doesn't have the answer, Ask says so — it doesn't guess.

✦
Retrieves before answering
Every query hits Katzilla Data live. No cached prose summaries.
✦
Cites every claim
Inline footnote markers + full source_url + retrieved_at + content_hash.
✦
Refuses without sources
"I don't have data on that" is a normal response. Better than a made-up one.
✦
Traceable to the byte
Every cited record carries a hash you can re-fetch to verify Ask saw exactly what it says.