ClinLit AIBlogAboutTry ClinLit AI

What ClinLit AI is

ClinLit AI is an AI agent with access to curated clinical science data. Today that means the ClinicalTrials.gov registry, the PubMed literature and FDA’s drug data. It can also read the protocols, statistical analysis plans (SAPs) and consent forms that trials have posted, so it can go deep with you on the details of clinical science.

It can follow a compound through the published literature, from preclinical work to the trials testing it, and on to its FDA approval, its label and the adverse events reported after it reached the market. It can chart and map what it finds, tell you which sites and investigators run the trials in your therapeutic area, and show you the trends. And it can work through hard questions with you, about protocol design, endpoints or statistical analysis, citing the trial records, articles and guideline sections it relies on.

Who it’s for

You know clinical research inside out but not data engineering. You know your data but not AI. Or you’re working on a tight budget. If any of these is you, you’re in the right place. Think of ClinLit AI as a co-thinker and co-investigator with access to a vast amount of public data, a good memory, and a knack for synthesis. Give it a try: I’m excited to see what it can do, and what you create together.

Who it’s not for

If you have deep pockets and already use enterprise data products such as Citeline, ClinLit AI may not be what you need. It’s built for people who don’t have that option. However, I do believe my AI agent is smarter, so please still give it a try and let me know what you think. Thanks!

What’s in it

ClinLit AI is in a pilot phase, running on modest computing power and memory, so some questions may take a while.

  • Clinical trials: about 431,000 interventional and expanded-access studies from ClinicalTrials.gov, through AACT (the CTTI’s copy of the registry). It’s updated every day from AACT’s daily copy, so a trial’s changes show up within a day or two of ClinicalTrials.gov’s. To keep the data and the searches manageable, I left out observational studies and Phase 4 trials.
  • How trials changed: every earlier version of each trial’s record, monthly since July 2024 and daily since October 2026, so it can tell you when a trial’s status, enrollment, dates, endpoints or eligibility changed (for example, when it added or removed a site). Earlier snapshots are in a format that doesn’t compare with today’s.
  • Trial documents: about 73,000 protocols, SAPs and consent forms that studies posted to ClinicalTrials.gov, with their full text searchable for most of them (scanned documents without a text layer can’t be searched), and the earlier versions of the ones that were revised. Make the most of these: they hold the details the registry doesn’t.
  • Published literature: about 17.6 million PubMed articles published since 2010 that have an abstract, with their titles, abstracts, MeSH topics, chemicals and publication types. It starts from the 2026 PubMed baseline and takes in PubMed’s daily updates every day, so new articles, corrections and retractions arrive within a day of PubMed’s.
  • FDA drug data: from openFDA, refreshed every week. Approvals and supplements for about 29,000 applications (Drugs@FDA), Orange Book patents and exclusivities, the NDC directory, drug recalls and shortages, and about 91,000 current prescription and biologic labels, searchable section by section, by meaning or by keyword. Every drug is linked by FDA’s ingredient identifier to the trials that test it and the articles about it, under any of its names: generic, brand or development code.
  • Adverse event reports: about 7.8 million serious reports to FDA’s Adverse Event Reporting System (FAERS) from the last ten years, updated each quarter, with the statistics that show which reactions are reported more often than expected for a drug. These are voluntary, unchecked reports, shaped by publicity and by who reports, and not scientific proof that a drug caused anything.
  • Good Clinical Practice: the ICH E6(R3) guideline, which it can quote and cite by section.
  • The web: when a question needs something the data doesn’t hold, such as the latest news or another regulator’s decision, it can search the web and read the pages it finds, including the FDA approval letters and reviews the data links to, and links each point it takes from there to the page it came from.

Coming next: patents beyond the Orange Book, to complete a drug’s path from the lab to the market.

How you can help

I’d love to hear from you about:

  • questions ClinLit AI didn’t answer well;
  • data sources you’d like it to have;
  • features you’d like it to have;
  • your experience using it, such as how long answers take;
  • any other ideas worth a try;
  • how much you’d pay for a subscription to a product like this.

Use the Feedback button at the bottom left of the page, or email [email protected].

A few things to know

  • It sticks to its subject. A filter checks each question before the AI sees it, and politely declines anything outside clinical research, drugs, the biomedical literature and regulation (the weather, say), so the AI’s budget goes on the questions that matter. Greetings are fine. If it turns down a question it should have answered, please tell me.
  • The data is updated regularly. PubMed articles arrive within a day of PubMed’s, new papers, corrections and retractions included, and trial records within a day or two of ClinicalTrials.gov’s. FDA’s drug data is refreshed every week, and adverse event reports each quarter, when FDA publishes them. It isn’t live, though: for something that happened recently, check the source.
  • It can make mistakes. Like any AI, it can get things wrong. Check its answers, especially before relying on them; it cites its sources so you can. Its answers are not medical, investment or regulatory advice.

Your data

  • The AI model: ClinLit AI currently uses DeepSeek’s model to reason and write answers, because it’s affordable and capable. That means your questions, and the conversation so far, are sent to DeepSeek through its API. If that’s a concern for you, please tell me: if enough people want another model, I’ll switch.
  • Web search: when it searches the web, the search terms it writes from your question and the addresses of the pages it reads, including any link you give it (not your conversation), are sent to Tavily, a web search service.
  • What we keep: your questions and ClinLit AI’s answers, with your IP address, are recorded in our technical logs so I can fix problems; these logs are deleted after 30 days. Our web server’s access logs, which include IP addresses, are deleted after 14 days. Cloudflare, which delivers and protects the site, also sees your IP address. Your conversation history itself stays in your browser, not on our servers.
  • Please don’t enter personal health information, other personal details, proprietary data, business secrets or anything else sensitive.

The full details are in our privacy policy.

Who I am

I’m a solo developer with hands-on experience in data science, AI and software engineering. I also work in consulting and business strategy, with domain expertise in clinical operations and pharma R&D. I built ClinLit AI to see what that combination of skills could make, and to offer a useful tool to people who, like me, are interested in AI and healthcare.

Send feedback