Evaluating the FamilySearch AI Research Assistant

 

FamilySearch.org

The notices of the FamilySearch.org AI Research Assistant have become ubiquitous and I have been receiving constant feedback about their use and effectiveness. I have been evaluating the usefulness of several AI Chatbots over the past two years in cluding Google Gemini, OpenAI ChatGPT, Microsoft Copilot, Perplexity, Claude, and others. Initially, my conclusions about their ability to either do genealogical research or even assist with doing genealogical research were negative, to say the least. However, because my initial reaction to the FamilySearch.org AI Research Assistant (AIRA) was unfruitful and negative, I haven't really looked at it for a while. 

The questions being asked by the genealogists I usually work with are often also not very positive and also question the AIRA's usefulness. In addition, over time, I have found the latest versions of some of the Chatbots to be very helpful. I have recently written about my remarkable experience with Google Gemini 3. See Can Gemini 3 do valid genealogical research?

The AIRA from FamilySearch is not a chatbot. It is merely a specifically trained finding aid. Here is a recent entry from the website showing the results of clicking on the first suggested item. 

The two choices are to View Record or Attach. Basically, there is no way to know anything about this entry other than an entry for a "Frank Hamilton" in the Illinois, Births and Christenings, 1824-1940 record collection. The AIRA entry suggests that this record for this "Hamilton" who "is listed as a child of Frank James Hamilton in the record, but is not found in your tree." The record for Frank James Hamilton has three named children listed. 

There are also two record hints listed for the same record suggested by the AIRA. 

Neither of these two Record Hints had a direct reference to anyone named Hamilton. The second Record Hint (aka Research Help) suggested a Full-text search of the Births: Hancock, Illinois. Birth Certificates 1891 collection. I did the Full-text search and examined all 43 results. I did find an interesting entry for this family but not the one linked to the AIRA suggestion. 

I have no question about the value of full-text searches. Here is the other record suggested by the AIRA. It does not appear to have an entry for anyone named Frank or Hamilton. 

I find the record hints to be useful when verified by either existing information or additional research. I do not think that giving people the chance to attach a record without verification or additional research is a good idea. 

By the way, the records I found by full text search mostly have the name Hamilton because of a town name not a person's name. In addition, as noted above, the Frank Hamilton family in the FamilySearch Family Tree has a child born in October of 1892, Clara Marguerite Hamilton L5NK-1LB, with death records with a death date of October 26, 1892, the same month and date as the full-text search document but a different year. So this unnamed document is probably their first child with the correct birth date. 

What does all this indicate to me? I see no difference between a record hint and the so-called AIRA. Both can be inaccurate or helpful but both usually require a significant amount of research to validate the suggested information. It does not help to label the suggested research as being AI-generated.