What happens to what you type
Last updated 14 July 2026.
Most privacy policies are written to protect the company. This one is written to tell you what actually happens, in the order it happens. The short version: TrialThread stores nothing about you — no account, no profile, no database, no record of your search. An AI service reads your description in order to do the search, and does not keep it. The detail is below, including the two places where that sentence needs a footnote.
Do not type anything that identifies a person
No names. No date of birth. No medical record number, insurance number, or Social Security number. No address, phone number, or email. TrialThread does not need any of it and cannot use any of it. Describe the medical situation, not the person: “My mother is 58, HER2-positive metastatic breast cancer, spread to the liver, two prior treatments, we live near Chicago” is everything the search requires. TrialThread is not a covered entity under HIPAA and does not operate under a Business Associate Agreement — which is a formal way of saying it is not the right place to put identifiable medical records, and it never needs to be.
The path your words take
1. Your browser. You type a description into the box. It is sent to TrialThread over an encrypted connection. It is never put in the web address, so it never lands in your browser history, in a bookmark, or in a link you might share by accident. TrialThread sets no cookies and stores nothing in your browser.
2. TrialThread’s server. The description is held in memory for the length of your search and then it is gone. There is no database — not an empty one, none at all. It is never written to a log file: when a search fails, the error report records the stage that failed and a scrubbed error message, never your text. Nobody, including the person who built this, can look up what you searched for, because it was never written down.
3. The AI provider. To read your description and read the trial criteria, TrialThread sends the text to Anthropic’s Claude API. This is the one place your words leave TrialThread, so here is exactly what happens to them there. Under Anthropic’s API terms, conversation content — your prompts and the model’s outputs — is not retained by default, and it is never used to train their models without express permission. The one exception: Anthropic’s automated safety systems screen content, and if a session is flagged as violating their usage policy, they may retain it for up to two years. Describing a cancer diagnosis is not the sort of thing those systems flag — but you are entitled to know the mechanism exists, which is why it is written here instead of left out.
4. ClinicalTrials.gov. TrialThread queries the public U.S. registry for trials. Those queries carry medical search terms and a city-level location. They never carry your free text.
What the analytics record — and what they cannot
TrialThread counts how the product is used, because a free tool nobody can measure is a tool that quietly breaks. The counting is deliberately blunt. It records events like “a search started,” “a search returned nine results,” and “someone clicked the third one.” It records no free text, no condition names, no biomarkers, no locations, no trial identifiers, and no ages. Medical interest is medical data, and none of it is collected. The analytics are cookieless, set no advertising or cross-site trackers, and cannot follow you anywhere. That rule is enforced in the code, in a file you can read: lib/analytics.ts. Your IP address reaches the hosting provider, as it does with every website on the internet, and is used to rate-limit abuse.
Said precisely, one claim at a time
These sound like one sentence and they are not, so here they are separately. No account: true — there is no way to create one. No database: true — the application has no data store of any kind. Not logged by TrialThread: true — patient text is scrubbed from error reporting by design. Not retained by the AI provider: true by default, with the trust-and-safety exception named above. Not used to train AI models: true. Never seen by anyone but you: not a claim we can make — the text is processed by a third-party AI service in order to work at all, and that is the honest boundary of the promise.
What we would have to tell you
If TrialThread ever adds saved searches or accounts, it will be opt-in, explained in plain language before you choose, and deletable — and this page will change first. If the AI provider changes, or its retention terms change, or TrialThread switches to a model with different retention rules, this page changes first. If any of the above turns out to be wrong, it gets corrected here and the correction is noted, rather than quietly edited. That has already happened once: an earlier draft of this page overstated how long the AI provider keeps your text. It was corrected against the provider’s own documentation before this page ever went live.
Security
If you find a security problem, please report it privately — the how to report a security problem page explains the channels. Never include real patient information in any report.
Questions this page does not answer belong on the questions page, or in an issue.