If you've read our About page, the short version is "four researchers, one judge." This page is the longer version, written for transparency and disclosure.

Genealogy Genie AI uses AI to analyze information surfaced through web searches, FamilySearch integration, and open archives, highlighting new avenues of research for family historians.

The AI does not modify or quote source material. Two reasons:

Copyright integrity. The records, indexes, and articles we point you to belong to their publishers, archives, and rights holders. We don't reproduce their content; we send you to it.
Research integrity. If the AI quoted, paraphrased, or summarized a source inline, the source could drift — a date misread here, a name flattened there. Suggestive becomes definitive without anyone noticing. We don't allow it.

Instead, the AI generates a short analysis of what the source appears to contain and why it may be relevant to your research. The original source itself is presented to you by non-AI code, which renders a direct link to the rights holder's site so you can read the unaltered original in its original context. The AI cannot alter that link, the source, or the surrounding context.

AI participates in four distinct, bounded steps. None of them writes to your tree.

1. Search term derivation. The AI reads what is known about an individual or research question and proposes search terms likely to surface new information. The functional (non-AI) code runs those searches against web sources, FamilySearch, and open archives. The AI proposes that deterministic code executes.

2. Result analysis. Once results come back, the AI produces a short written analysis addressing the original research question — what looks relevant, what looks tangential, where the apparent gaps are. This analysis is the AI's own work product; it is not a quotation of any source.

3. Independent review by a second AI. A separate model, operating under a different persona and prompt, reviews the first AI's analysis against the actual retrieved data before any of it reaches you. The reviewing AI's only job is to catch claims that the underlying results don't support. As with step 2, the second AI cannot alter source URLs or source content — it can only approve, flag, or send the analysis back for correction.

4. Periodic tree review. AI periodically reviews events already in your tree to identify factual inconsistencies (children born before parents, implausible ages, conflicting locations), cultural or historical implausibilities, and openings for further research. Findings are returned with a priority grade. Nothing is written to your tree. You decide what, if anything, to act on.

The models we depend on shift as the field evolves. What's the best fit for family history research today may not be the best fit in six months. We continually re-evaluate and rotate models in and out based on their performance against our internal evaluations for source-grounded analysis.

If a model can't meet our standards for grounded, verifiable analysis, it doesn't go into rotation regardless of how impressive its marketing is.

As of today, the lineup is:

You may have read about AI systems that aggressively crawl the web, scraping entire sites, repeatedly hitting the same servers, and occasionally taking smaller operations offline. That behavior is incompatible with what Genealogy Genie AI is built to do, and our architecture is designed to avoid it.

We don't operate our own web crawler. We start with web searches through a commercial search provider, which means we draw from an index that has already been built in accordance with each site's robots.txt settings. If a publisher, archive, or genealogical society has told automated systems to stay out, those preferences are honored at the search-provider layer before a result ever reaches us.

For FamilySearch, we use the platform's own search. We don't crawl FamilySearch records; we use the search features FamilySearch provides for that purpose.

We don't scrape entire sites, and we don't pound a server looking for a single nugget. Each research question generates a small, targeted set of searches based on what we already know about the individual or event. We're looking for a few promising leads to pass to you, not building a private mirror of someone else's collection.

The whole point of the product is to send you to the rights holder's site to read the original in its full context. Hammering that site to extract fragments would defeat the purpose.

Genealogy Genie AI uses only commercial API subscriptions that have either opted out of training data collection or do not collect customer data for training in the first place. Your family information is not used to train any model — not ours, not the providers', not anyone's.

When you reject a finding inside the application, that signal stays within our system. We use it to refine future research for you — the AI learns what you've already evaluated and looks elsewhere. That rejection signal is not sent back to the model or API providers as training data.

This sits alongside the broader data protections described in our FAQ and Privacy Policy: names are transformed into system-generated identifiers before processing, events are recorded using years rather than full dates, and your data is encrypted, stored securely, and never sold.

Every AI output in Genealogy Genie AI is a suggestion attached to a real link to a real source. Confirming, rejecting, and recording evidence is your work, and it stays that way. We built the product to make it faster and more thorough, not to do it for you, and not to pretend it has.

If you have questions about how the AI handles a particular type of record or research scenario, send us a note. The email goes to someone who reads it.