Looking at Record Hints between the Big Three Genealogy Family Tree Websites

 

FamilySearch.org, Ancestry.com, and MyHeritage.com all provide record hints. Over the past few years, these matching systems have increasingly incorporated artificial intelligence. I use all three websites daily for genealogical research, and I have definite opinions regarding the accuracy, availability, and overall usefulness of their hints. These observations are not based on a formal study; they reflect my personal experience and methodology.

Each of these massive websites hosts its own unique collections, and hint generation naturally depends on the records each platform holds. These collection differences are among the most influential factors when judging utility. Because I am familiar with all three, I rely on each for its specific strengths. This diversity makes direct, 1-to-1 comparisons tricky—each platform is uniquely valuable in its own right.

Both Ancestry and MyHeritage offer an abundance of immediately actionable record hints (Record Matches on MyHeritage). When working on either platform, one could spend nearly full-time simply evaluating and attaching quality matches. FamilySearch.org remains the outlier. When adding a new person to the FamilySearch Family Tree, you might receive a trickle of hints, but more often the system seems to wait for you to build out the profile before suggesting relevant records. Because of this lag, I typically do the heavy lifting on Ancestry or MyHeritage first. Once I transfer that verified data over to FamilySearch, hints finally begin to populate—though they often simply mirror the records I just added from the other sites.

Ancestry and MyHeritage deliver hints seamlessly. On Ancestry, AI increasingly surfaces potential tree extensions and leaf hints that are generally well-supported and helpful, though complex or conflicting lines can still trip them up. MyHeritage’s Record Matches have consistently been among the most dependable, often capable of driving an entire line's research on their own.

FamilySearch has placed its AI-assisted research tools front and center. However, these AI-generated hints and suggestions can be deceptively tricky, frequently leading researchers into messy genealogical quagmires. Unwary users risk accepting these suggestions without proper source verification, quickly injecting errors into a shared tree.

I encountered a clear example of this recently with a FamilySearch AI-assisted suggestion. The tool suggested a spouse for an individual in England whose existing profile showed siblings born across five widely scattered locations—a classic red flag that the family grouping was already suspect. Accepting the AI recommendation at face value would have compiled bad data on top of bad data.

AI can and does make mistakes. Evaluating historical context, geographic plausibility, and family structure should never be left entirely to an algorithm. Future iterations of these tools will undoubtedly improve, but for now, everyday researchers need to approach AI-generated suggestions with healthy skepticism and rigorous verification.

Can AI Fix the FamilySearch Family Tree -- Part Two Concerns and Challenges

 

The first big question is whether AI can actually improve the quality of the massive FamilySearch.org Family Tree, and whether it can consolidate and search the scattered record collections across the website. FamilySearch currently maintains eight distinct record collections that are either searched individually or, at best, appear together in only one or two combined queries.

Here are the eight primary search areas under the "Search" tab:

• Records -- The core portal for searching traditional indexed records, currently supplemented by computer-indexed records that are increasingly AI-assisted.

• Full Text -- An AI-powered search (currently in FamilySearch Labs) that indexes every word in processed documents and generates transcriptions alongside the digital images.

• Images -- A vast archive of browse-only digital images, largely unindexed and gathered by FamilySearch camera crews around the world.

• Genealogies -- A diverse aggregate of unrelated databases, lineage collections, oral genealogies, and contributed family trees.

• Catalog -- The comprehensive bibliographic guide to the Family History Library's collection, indexed using traditional library methods at the collection or roll level rather than the individual record level.

• Books -- A collection of digitized family histories, local histories, and periodicals that are searchable full-text, word-for-word.

• Cemeteries -- A dedicated database of burial grounds and individual headstones searchable via the Cemeteries interface.

• Newspapers -- A historical newspaper database indexed for full-text search, provided through a partnership with NewspaperArchive.com.

It is obvious that AI can perform remarkable feats—particularly in automated transcription, paleographic deciphering, translation, and surrogate search. A key question, however, is whether AI is reliable enough to replace FamilySearch's traditional catalog and query systems, tools built on decades of painstaking volunteer indexing.

Full-text search can surface ancestral records that never made it into traditional field-indexed collections. However, it currently covers only a small percentage of total holdings, and there is no simple way to track its overall impact across the other major collections. While icon badges in the Catalog indicate certain microfilms or DGS folders that have been processed by Full Text, the remaining collections offer few clear indicators of inclusion.

If FamilySearch eventually applies full-text AI search across the vast majority of its holdings, will traditional searches—such as the Historical Records Collection and the Catalog—become obsolete? Furthermore, could full-text retrieval be paired with specialized AI agents capable of actively evaluating and contextualizing search results against existing Family Tree profiles? Ultimately, will AI agents assist in researching all the records on the FamilySearch website?

The AI technology answer all three questions with a "yes" is available today. The biggest challenge is processing the billions of records on the FamilySearch website and then redesigning the website to take advantage of the complete Full-text search capabilities. 

The results would then be similar to what is currently done on the Ancestry.com and MyHeritage.com websites. For example, MyHeritage.com has for years provided highly accurate Record Matches that search virtually all the pertinent records on the website in 42 languages. Also, Ancestry.com has AI assisted searches that suggest the next individuals on many of the end of lines. 

I guess the last question is how fast can FamilySearch start implementing the programming to solve the questions?

Can AI fix the FamilySearch.org Family Tree?

 

Many years ago, I had the opportunity to travel to Salt Lake City and to attend a very limited conference with FamilySearch about the future of the FamilySearch Family Tree. This was at the junction of the failure of the new.FamilySearch.org Family Tree and the desire to replace it with something else. The results of the day-long meeting was that the FamilySearch Family Tree should be an extensive wiki. However, at that date, relatively far in the past, it was not possible to imagine the size and complexity of the current FamilySearch Family Tree.

The FamilySearch Family Tree has grown in its size to an immense repository for a vast amount of information about the family of families of people throughout the world. The original concepts of allowing unlimited individual changes and unlimited individual contributions has become its greatest obstacle for continued use. Slowly over the years since it was introduced, the FamilySearch Family Tree has been modified from its original completely open wiki structure to be more restricted. 

One of the initial problems with the creation of the Family Tree was the addition of seed data that was incomplete and in many cases entirely inaccurate. This data had been created by years of individual genealogists acting separately with various levels of competency. Initial basic challenges of the Family Tree included, a massive amount of duplication, a considerable amount of inaccuracy and a substantial amount of imaginary data.

Efforts at regularizing and correcting the data entered into the FamilySearch Family Tree includes the creation of user trees that are independent of the FamilySearch Family Tree and maintained by individuals or organizations. Although a very valuable step in providing a venue for additional information that is separate from the FamilySearch Family Tree, the addition of user trees is only a small bandage covering the issues of the main tree itself.

FamilySearch additionally, resisted the suggestions that the wiki family tree be moderated. The amount of effort and the number of people to actually moderate the FamilySearch Family Tree seemed overwhelming. However, today, there is the possibility of using AI agents and AI oversight of the FamilySearch Family Tree is certainly a way to overcome the current difficulties.

Major issues facing the FamilySearch.org Family Tree include a substantial amount of duplication, changes made to well-documented and accurate information that is already present in the family tree, the addition of unsubstantiated and unsourced information, much of which is entirely inaccurate, other minor issues with the consistency of the data. All these issues fall into categories of maintenance that could be supplied by AI agents.

There is a major concern on the part of FamilySearch that requiring more accuracy and eliminating duplicates will discourage people from using the Family Tree. This is particularly true for first-time users entering their own information from personal records or memory. AI agents could easily be developed to assist with this initial entry, ensuring contributions are supported by an identified source—even if that source is simply personal knowledge. Beyond those initial generations, however, there is little justification for allowing constant, unchecked changes to records that are already thoroughly documented and substantiated.

Substantial parts of the family tree where the data has been substantiated and where all the ordinance work has been done by those doing the work in the temples of The Church of Jesus Christ of Latter-day Saints should be partially read-only. The basic information available which is well substantiated should not be subject to change, although the addition of additional sources and memories should be possible. My constant example of this is the descendants of the Mayflower.

Information about the steps being taken by FamilySearch to begin this process of adding AI agents is not readily available to the public. But it would help those who are now abandoning the family tree for other venues due to the constant changes in their own genealogical data. It would also help if FamilySearch used some of the most talented genealogists to provide the structure for the AI agents. 

Presently, AI is changing dramatically, almost weekly. And it may be an obstacle to implementing any particular type of AI agency until the AI begins to settle down. However, it would also be a good idea to use AI programming that would be at least in part independent of the major AI websites and providers.

The original obstacle in creating a moderated wiki was the overwhelming need for people to monitor the quickly growing family tree. Today, we have the technology to add these monitors and improve the operation of the family tree so that it may become useful even to those who are now abandoning it.

Can you compare AI websites?

 

AI hasn't reached the level of variety presently available to someone trying to buy a car, but there are still some of the same factors involved in deciding which AI website to use. Each type of car available is scrupulously reviewed in online YouTube channel videos. Likewise, the different AI websites come under tremendously detailed examination. Like cars, sometimes the differences in AI websites turn out to be preferences. The usual car video compares two, three, four, or five different models of cars that are essentially the same. There are also major online reporting agencies, such as Consumer Reports, that rank vehicles on their own set of criteria. Likewise, AI websites are compared based on differences that are often totally unapparent to the general AI users. 

When I was very young, I lived in a very small town, and there was a major division between those in the town who drove Chevrolets and those who drove Fords. Members of each of the two groups were enthusiastically defensive of their choices and viewed the opposite car/truck owners as ill-informed and sometimes slightly un-American. Each year, of course, there was a scramble to compare the newest versions of the offerings from both car companies. 

We are in about the same condition today with the AI websites as happened during the past few weeks, new models or new versions of nearly all the websites have been introduced, and the online critics immediately begin the comparison. Unfortunately, the comparisons usually fall into the area of preferences rather than differences discernible by the ordinary user. For example, once again comparing the AI websites to vehicle purchases: the difference is that if you are a building contractor, you probably are more interested in a truck. If you are an individual whose use of the vehicle is limited, you probably drive a smaller, more economical vehicle. 

This is the case with AI users around me. Most of them are very happy using the free versions of any of the AI websites. These users don't need the enterprise versions of the AI websites any more than the average driver needs a  Ford F-450 Super Duty, Ram 3500 or a Chevrolet Silverado 3500HD, 

It is also apparent that, just as there is constant attention paid to the features of the new automobiles, the AI websites are continually compared by the new features that are implemented. Trying to keep up with AI is complicated by the fact that the features that are being reviewed are often so esoteric that only trained computer professionals can understand the nuances of the differences. 

What I think the individual AI user is really looking for is the AI equivalent of a Toyota Corolla.

If you are interested in exploring the differences between the various AI websites, the best way to do so is to begin asking your preferred website about how all the different offerings from Google Gemini to Claude are ranked under different criteria. By the way, the gas mileage or electrical mileage of the AI industry is measured in tokens. In one category, the websites with the highest number of usable tokens win the game. However, just like focusing on vehicle mileage ignores all of the other considerations involved in the purchase of a vehicle, focusing just on tokens produces the same results. The real difficulty is that the current state of the AI industry is that there is not a discernible difference in features that allows users to make educated choices. I find that most users make the choice of which AI website to use based mostly on recommendations from those that they view as more educated.

I spend a considerable amount of time comparing and contrasting various AI offerings and make my choice primarily on the breadth of utility, i.e., the features available. For some of the same reasons that I drive a Subaru, at this point, my ownership of TrailSeeker is based on economics and the fact that the car has high clearance and that I live in a town where there are terrible roads, steep hills, ice and snow, and other obstacles.

I suggest, just as I do with cars, that you do your homework, look at a lot of options, and make your choices based on what fits your needs the most.

Recently seen on Facebook: "I don't touch AI, and certainly don't want it anywhere near my genealogy !"

I am puzzled how this person was able to avoid using AI, unless they are conducting genealogical research completely on paper. If they have used any of the large family history platforms—from FamilySearch to Ancestry or MyHeritage—they have already been using AI. The record matching algorithms, automated indexing, and handwritten text recognition (HTR) behind these sites all rely on machine learning. In fact, anyone using Google Chrome or Microsoft Edge to search the web, compose emails with predictive text, or browse social media is regularly interacting with AI. There is simply no longer a practical way to participate online without it.

I maintain trees on FamilySearch, Ancestry, MyHeritage, and other platforms. AI is a remarkable research assistant if you understand how to harness it.

Even if we assume someone has somehow avoided direct contact with AI tools, their online "genealogy" is almost certainly built upon data, indexes, and digital collections surfaced by artificial intelligence. The massive shared trees online demonstrate just how intertwined modern research is with automated technology.

Years ago, I saw the exact same resistance from some of the attorneys in my law office. They were bound and determined to avoid computers. That seemed viable until they realized the rest of us were completing research in a fraction of the time using legal databases like Westlaw. Complete avoidance became impossible once the Arizona court system mandated electronic filing. Very shortly, even attorneys across the United States will find it impossible to practice law without utilizing some form of AI.

It will only be a matter of time before AI is entirely ubiquitous.

Introducing GAIA -- A Real AI Research Agent from MyHeritage

I was one of the very early testers of the new GAIA AI Research Agent from MyHeritage.com. I had time to test GAIA over a period of weeks with some extensive research projects. Here is what I said at the end of the test run. 

The addition of GAIA as a research assistant on MyHeritage is a milestone in the development of the website. I have been testing new genealogy programs for many years, and I always begin with information I already know thoroughly. This lets me determine whether a program provides new clues, generates inaccurate data, or merely copies what already exists in my tree.

Working with GAIA was a new experience. As a test, I chose one of my direct-line sixth-great-grandmothers, Lydia Stewart. Generally, maternal lines are far less researched and documented than paternal lines, but Lydia Stewart happens to be well documented. Even so, many questions about families living in the late 1700s and early 1800s remain difficult to resolve.

As I began using GAIA, I was impressed by the questions it asked to advance the research. Drawing on existing genealogical sources and data from other family trees, its conclusions were consistent and well-grounded. Lydia Stewart died in Bolton, Warren County, New York, and GAIA located a link and brought up documents pointing to a possible probate search. Although I already had substantial sources for Lydia, any reference to a probate record was missing, making this a valuable suggestion. The Warren County online records look to be a promising resource. GAIA also provided details on Lydia’s children that were absent from my own files. While I might eventually have discovered these sources myself, GAIA’s ability to find them demonstrates its real value as a research assistant.

You find GAIA on an individual's profile page. It is opened by clicking on the "Research this person" link. This opens a page with a list of possible sources and a further link to "Research Agent Suggestions."

Here is an example of the beginning research for my direct line ancestor, Garrard Morgan. 

Here is a summary of which MyHeritage subscription levels provide access to GAIA. 

• Complete Plan: Grants access to the full historical data collections (billions of records) and advanced research tools required for the AI agent to search, extract, and match historical records against your tree.

• Omni Plan: MyHeritage’s all-inclusive subscription. It includes everything in the Complete plan plus higher usage allowances/quotas for advanced AI tools, full access to OldNews.com, Geni Pro, and photo tools. Notice the orange "Go Omni" banner in your header—users on lower tiers or with standard plans are prompted to upgrade to Omni for expanded usage limits.

• Data Plan (or standalone Data / Tree subscription): A basic Data subscription allows record matching, but specialized AI features typically operate on a quota system with limited searches or trial prompts before requiring an upgrade to Complete or Omni.

• Free / Basic Accounts: Free accounts are generally limited to a small initial preview or trial run of AI features before being prompted to purchase a subscription to view full source matches, conduct additional sessions, or extract records into their tree.

You can check your exact remaining monthly allowance or session quota at any time by clicking the "GAIA usage" link located at the bottom left of your chat panel. I currently have the Complete plan. AI usage allowances operate on a rolling monthly cycle that refreshes on the same day each month corresponding to your subscription renewal or initial allocation date (not necessarily the 1st of the calendar month).

For a lot more information go to the MyHeritage Blog:  MyHeritage Introduces GAIA, a Groundbreaking AI Agent for Genealogical Research

What is a Genealogical Source?

The simple definition of a genealogically valuable historical record or source is a record created at or near the time of an event by someone who participated in it, witnessed it, or had an official duty to record it. Genealogically valuable records are not limited to a particular era or format; an oral history or personal recollection from a relative fits this description just as well as an archival document.

Unfortunately, the word "source" often carries multiple meanings. In genealogical practice, what we often need most is a clear record of where that source resides, that is a citation telling exactly where and how the information was obtained. Over a lifetime of reading footnotes, I have found that few explain this access route in detail. Most merely cite a title or publication without indicating where a researcher might actually locate it. Citations vary widely in style and depth, but even a straightforward bibliographic citation can establish this connection:

Here is a perfectly acceptable citation to a book. 

Bennett, Archibald F. A Guide for Genealogical Research. Genealogical Society of the Church of Jesus Christ of Latter-day Saints, 1956.

From a genealogical perspective, the citation is missing the key element of where the book was found. In this case, the book is still subject to copyright protection and is only available in libraries and for purchase online. If the citation included a reference to the location of the book, subsequent researchers would not have to go through and try to find the book themselves. Another example of the need for specificity is a citation such as this one for this record:

“Santa Ana. Confirmation Records 1730–1736.” Accessed October 24, 2026. 

This citation tells us where the record was found. However, that particular set of records has 835 images, and lacking a more exact citation, I would possibly have to go through all 835 records to find this particular image. Here is a more complete citation:

“Santa Ana. Confirmation Records 1730–1736.” Accessed October 24, 2026. https://www.familysearch.org/ark:/61903/3:1:33SQ-GGZK-8NV?cat=koha%3A686299&view=explore&lang=en&groupId=M986-8MC

Fortunately, many of the online genealogy programs have a way to save a complete citation to a specific record. However, this type of feature is not universal, and occasionally you will get a citation to the entire set of records. Here is the citation to the record from FamilySearch.org.

"Santa Ana, Oaxaca, Mexico records," images, FamilySearch (https://www.familysearch.org/ark:/61903/3:1:33SQ-GGZK-8NV?view=explore : Oct 24, 2026), image 14 of 103; Archivo Diocesano de Oaxaca. Image Group Number: 004782273

Both these citations take you to the original document. Another important reason for having more specific citations is that you, in doing your own research, may need to go back and look at the original record. Without a specific citation, you may end up doing the research over again.

There are a number of programs which will automate the process of creating citations. A free app is Zotero.org from the Massachusetts Institute of Technology. 

Be sure and remember that the source is the description of the origin of the information. The citation is the format in which it is presented. 

There are possibly hundreds of different citation formats. However, genealogists primarily use the Chicago Manual of Style. If you use an app such as Zotero.org, the app itself will format your citations in the style you select.