Difference between revisions of "Genealogy"

From
Jump to: navigation, search
(Created page with "{{#seo: |title=PRIMO.ai |titlemode=append |keywords=artificial, intelligence, machine, learning, models, algorithms, data, singularity, moonshot, Tensorflow, Google, Nvidia, M...")
 
(→‎Records and Search)
(12 intermediate revisions by the same user not shown)
Line 1: Line 1:
 +
__NOTOC__
 
{{#seo:
 
{{#seo:
−
|title=PRIMO.ai
+
|title=Genealogy
 
|titlemode=append
 
|titlemode=append
−
|keywords=artificial, intelligence, machine, learning, models, algorithms, data, singularity, moonshot, Tensorflow, Google, Nvidia, Microsoft, Azure, Amazon, AWS
+
|keywords=genealogy, AI genealogy, family history, DNA, artificial intelligence, handwriting recognition, full-text search, FamilySearch, MyHeritage, Ancestry, Transkribus, family trees, historical records, home genealogy
−
|description=Helpful resources for your journey with artificial intelligence; videos, articles, techniques, courses, profiles, and tools
+
|description=A practical guide to using AI tools for genealogy at home: free-first workflows, handwriting recognition, DNA match analysis, photo restoration, verification discipline, privacy, and cost.
 
}}
 
}}
−
[https://www.youtube.com/results?search_query=Bioinformatics+Bioinformatics+genetics+CRISPR+gene+dna+artificial+intelligence+deep+machine+learning Youtube search...]
+
[https://www.youtube.com/results?search_query=AI+genealogy+family+history+tools YouTube]
−
[https://www.google.com/search?q=Bioinformatics+Bioinformatics+genetics+CRISPR+gene+dna+artificial+intelligence+deep+machine+learning ...Google search]
+
[https://www.quora.com/search?q=AI%20genealogy%20family%20history ... Quora]
 +
[https://www.google.com/search?q=AI+genealogy+tools+family+history ...Google search]
 +
[https://news.google.com/search?q=AI+genealogy+family+history ...Google News]
 +
[https://www.bing.com/news/search?q=AI+genealogy+family+history&qft=interval%3d%228%22 ...Bing News]
  
−
* [[Case Studies]]
+
* [[Bioinformatics]]
−
** [[Paleontology]]
+
* [[Paleontology]]
−
** [[Healthcare]]
+
* [[Healthcare]]
−
*** [[Pharmaceuticals]]
+
* [[Life Sciences]]
−
*** [[Drug Discovery]]
+
* [https://familytreemagazine.com/websites/ai-tools-roundup-big-websites/ Roundup: AI Tools at Ancestry.com, FamilySearch and MyHeritage — Family Tree Magazine] ... Comparison of the competing research chatbots announced at RootsTech 2026, including MyHeritage's GAIA and Ancestry's forthcoming assistant.
−
**** [[Protein Folding & Discovery]] ...[[Protein Folding & Discovery#Google DeepMind AlphaFold|Google DeepMind AlphaFold]]
+
* [https://www.familysearch.org/en/blog/what-to-expect-from-familysearch-in-2026 What to Expect from FamilySearch in 2026 — FamilySearch] ... Full-Text Search expanding to additional languages, a help chatbot, a discovery assistant for filtering hints, and data-quality alerts on the shared tree.
−
*** [[Genealogy]]
+
* [https://www.familysearch.org/en/blog/ai-research-assistant-home-page-hints Introducing Tree-Extending Hints from the AI Research Assistant — FamilySearch] ... Home-page hints that flag likely tree extensions for review.
−
** [[Life Sciences]]
+
* [https://denyseallen.substack.com/p/ai-safety-family-historians-genealogy AI Safety for Family Historians — Denyse Allen] ... On fabricated sources, synthetic media, and pressure on the volunteer infrastructure that free genealogy depends on.
 +
 
 +
 
 +
== Overview ==
 +
 
 +
Between 2024 and 2026, AI moved from an experimental add-on to standard equipment in consumer genealogy. The major platforms now run AI record matching by default, handwriting recognition has made billions of previously browse-only document images searchable, and general-purpose chatbots have become common research companions.
 +
 
 +
For the home researcher, the useful question is no longer ''whether'' to use AI, but '''which tool for which job''' — and how to keep an eager, confident machine from quietly corrupting a family tree. No single tool does everything well. The gains come from matching a specific tool to a specific stage of research, and from verifying everything it produces.
 +
 
 +
This page is organized around that practical framing. The technical and professional material has been kept but moved toward the end.
 +
 
 +
== Quick Start: A Free-First Workflow ==
 +
 
 +
A home researcher can get a long way without paying anything. A sensible order of operations:
 +
 
 +
# '''Records first.''' Build the skeleton of the tree on [https://www.familysearch.org/ FamilySearch] (free, nonprofit, billions of records and one shared collaborative world tree). Do this before spending money anywhere.
 +
# '''Then unindexed images.''' Run names through [https://www.familysearch.org/en/search/full-text FamilySearch Full-Text Search], which uses handwriting recognition to surface records that traditional name indexes never captured — deeds, probate files, land records, and court papers where ancestors are mentioned but not indexed.
 +
# '''Then the hard handwriting.''' For documents the platforms can't read — German Kurrent, old Italian scripts, eighteenth-century legal hands — use a dedicated HTR tool such as [https://www.transkribus.org/ Transkribus]. For a single census line on your phone, Google Lens or Apple Live Text is often enough.
 +
# '''Then DNA.''' Test, then upload to free comparison sites and use match-analysis tools to organize clusters before drawing conclusions.
 +
# '''Then photos.''' Restoration, colorization, dating, and animation are the most emotionally rewarding stage and the least evidentiary. Do them last.
 +
# '''Then write it up.''' Use a general assistant to turn verified facts into readable narrative — never to supply the facts themselves.
 +
 
 +
Budget guidance: most active home researchers can work effectively at $0–25 per month. A paid subscription starts paying for itself only once you are researching most weeks.
 +
 
 +
== AI Tools by Task ==
 +
 
 +
=== Records and Search ===
 +
 
 +
{| class="wikitable"
 +
! Tool !! What the AI does !! Cost !! Notes for home users
 +
|-
 +
| [https://www.familysearch.org/ FamilySearch] || AI-indexed records, Full-Text Search over handwriting, help chatbot, and an AI Research Assistant that surfaces "tree-extending hints" on the home page || Free || The strongest free offering. AI does a first pass on records; volunteers check its work through Get Involved
 +
|-
 +
| [https://www.ancestry.com/ Ancestry] || Hints and ThruLines surface likely records and connect your tree to other users' trees; automatic transcription and tagging make handwriting searchable || Subscription (roughly $25–50/month by plan) || Hints become genuinely useful once a tree passes about 50 people. An "Ask AncestryAI" feature appears on some accounts from the LifeStory tab, currently limited to preset questions
 +
|-
 +
| [https://www.myheritage.com/ MyHeritage] || Automatic Record Matches, AI Record Finder™ (chat-based record search), AI Biographer™, Scribe AI document analysis, Cousin Finder, Theory of Family Relativity || Free tier covers trees up to 250 people; full access via subscription (all-inclusive plans around $299/year, first-year discounts common) || Strongest international and European record coverage of the big three. A broader research assistant, '''GAIA''', was demonstrated at RootsTech 2026 and beta-tested that spring; check current availability
 +
|-
 +
| [https://www.findmypast.com/ Findmypast] || Tree-linked record hints, plus searchable text recognition across the [https://www.britishnewspaperarchive.co.uk/ British Newspaper Archive], which it runs in partnership with the British Library || Family Tree tier £9.99/month, Everything tier £24.99/month (prices in GBP) || The pick for British, Irish and Commonwealth lines. Important caveat: the cheaper Family Tree tier surfaces records '''only''' through hints — you cannot search the collections independently. Over a million newspaper pages are free to view with a free account via the "Free To View" access filter
 +
|-
 +
| [https://www.wikitree.com/ WikiTree] || Almost nothing, deliberately. No AI hints and no automated tree-matching; GEDCompare flags likely duplicates on upload, but conclusions get argued out by people on the profile page || Free, always || Useful as a check on the other three. Roughly 1.2 million members maintaining one shared tree of 40+ million profiles, with members signing an Honor Code that requires citing sources. GEDCOM uploads are capped at 5,000 people. FamilyTreeDNA added direct tree linking in March 2026; about 15 million profiles carry DNA connections
 +
|-
 +
| [https://www.goldiemay.com/ Goldie May] || Third-party research assistant that runs searches across FamilySearch, Ancestry, MyHeritage, Newspapers.com, GenealogyBank and Storied, then evaluates which results are most promising; also drafts citations, transcribes handwriting, and lets you choose among several underlying models || Subscription — verify current tiers || Sits on top of subscriptions you already hold rather than replacing them. States that user data is not used for model training
 +
|}
 +
 
 +
=== Handwriting and Transcription ===
 +
 
 +
{| class="wikitable"
 +
! Tool !! Best for !! Cost !! Caveat
 +
|-
 +
| [https://www.transkribus.org/ Transkribus] || Difficult, old, or uncommon scripts; can be trained on a specific clerk's hand or a particular record set || Credit-based; limited free tier || Makes ordinary recognition errors but does not hallucinate — it will not invent text that isn't there. Accuracy varies sharply by document type
 +
|-
 +
| Google Lens / Apple Live Text || A single record photographed on a phone || Free || Surprisingly capable for casual use; no batch processing
 +
|-
 +
| Platform-native HTR (FamilySearch, MyHeritage Scribe AI, Kindex, Goldie May) || Documents already inside a platform workflow || Varies || Convenient, but generalized models tuned for volume rather than for unusual scripts
 +
|-
 +
| General chatbots (ChatGPT, Claude, Gemini) || Reading a page and explaining archaic terms, abbreviations, and legal formulae || Free tiers available || '''Will confidently misread and then smooth over the gap.''' Useful for interpretation, risky for raw transcription
 +
|}
 +
 
 +
=== DNA Match Analysis ===
 +
 
 +
Consumer DNA testing itself is largely statistical rather than generative AI, but the analysis layer has become increasingly automated.
 +
 
 +
* '''Testing companies:''' AncestryDNA (largest database), MyHeritage DNA (strong European matching), 23andMe.
 +
* '''Match organization:''' [https://dnapainter.com/ DNA Painter] offers chromosome mapping plus tools for testing hypothetical relationships, which is particularly useful for unknown-parentage cases. Free tier available; a paid tier (around $65/year) adds clustering.
 +
* '''Relationship prediction:''' MyHeritage's cM Explainer and similar tools estimate likely relationships from shared centimorgans.
 +
* '''A caution:''' shared DNA constrains the ''range'' of possible relationships; it rarely proves one. AI tools that name a specific relationship are presenting a probability as a conclusion.
 +
 
 +
=== Photographs ===
 +
 
 +
{| class="wikitable"
 +
! Feature !! Platform !! What it does
 +
|-
 +
| Photo Enhancer, Photo Repair, MyHeritage In Color™ || MyHeritage / Reimagine app || Sharpens, repairs damage, and colorizes. Free uses are limited, then subscription
 +
|-
 +
| PhotoDater™ || MyHeritage || Estimates when a photo was taken from clothing, hairstyles, photographic style, and paper stock. '''Free.''' Can miss by a decade or more on unusual photos, but narrows an otherwise open search window
 +
|-
 +
| Deep Nostalgia™, LiveMemory™ || MyHeritage || Animates faces or whole scenes into short video clips. Premium after a small number of free uses
 +
|-
 +
| Photo Tagger || MyHeritage || Face clustering across a photo collection, with human confirmation required
 +
|-
 +
| Remini, VanceAI, Topaz || Third-party || Restoration and upscaling; often stronger on badly damaged originals
 +
|}
 +
 
 +
'''Important distinction.''' Enhancement, colorization, and animation are ''interpretations'', not evidence. Colorized clothing is a guess. An animated ancestor is a generated fiction of a real person. Keep the original scan, label derivatives clearly, and never let an enhanced image become the archival copy.
 +
 
 +
=== General-Purpose Assistants ===
 +
 
 +
ChatGPT, Claude, Gemini, and Perplexity are all used widely by genealogists. The practical differences between them are real but narrowing, and the skills transfer. Experienced practitioners generally advise picking one, using it for a month, and getting fluent rather than tool-hopping.
 +
 
 +
Good uses:
 +
* Building a research plan or locality guide
 +
* Summarizing a long document you have already obtained
 +
* Explaining archaic legal, occupational, or medical terminology
 +
* Translating older foreign-language records
 +
* Drafting an ancestor biography from facts ''you'' supply
 +
* Improving your own prompts — ask the model how to rephrase a request that isn't working
 +
 
 +
Poor uses:
 +
* Asking for records, citations, or sources directly
 +
* Asking who someone's parents were
 +
* Anything where you cannot check the output against an original
 +
 
 +
== What AI Does Well, and Where It Fails ==
 +
 
 +
'''Reliable strengths:''' searching enormous corpora quickly, reading handwriting at scale, translating, clustering and organizing, summarizing documents you already hold, and generating readable prose from verified facts.
 +
 
 +
'''Persistent weaknesses:'''
 +
* '''Hallucination.''' Models invent plausible-sounding names, dates, and entire citations. A model may correctly identify ''where'' a record ought to be and then fabricate the record itself. This is the single largest risk to a home researcher's tree.
 +
* '''No concept of truth.''' AI cannot distinguish a correct inference from a confident guess, and will fill a data gap rather than report it.
 +
* '''Missing cultural and historical context.''' Naming conventions, migration patterns, religious practice, and jurisdictional boundary changes are routinely mishandled.
 +
* '''Silent propagation.''' An unverified AI suggestion merged into a shared collaborative tree spreads to every researcher downstream and is difficult to unwind.
 +
 
 +
== Verification Discipline ==
 +
 
 +
The consensus rule across FamilySearch, professional genealogists, and experienced AI-genealogy educators is the same: '''AI assists, you verify.'''
 +
 
 +
# Treat every AI output as a '''hint''', never as a source.
 +
# Follow the hint to the original record image and read it yourself.
 +
# Cite the original record, not the AI.
 +
# If a model gives you a citation, assume it may not exist until you locate it.
 +
# Note '''"AI-assisted"''' in your research log for any step where a tool contributed. This preserves transparency for anyone reviewing your work later, including future you.
 +
# Do not merge AI-suggested relationships into a shared public tree until they are independently supported.
 +
 
 +
One widely taught framework is '''AIM''' — ''Assess'' what the tool gives you, ''Interact'' to refine it, ''Measure'' it against your sources.
 +
 
 +
== Privacy, Cost, and Data Ownership ==
 +
 
 +
* '''Uploads are disclosures.''' Putting photographs, records, or DNA data into a commercial AI system means entrusting a company with sensitive material about living relatives who did not consent.
 +
* '''Living people.''' Avoid pasting details about living individuals into general-purpose chatbots. Redact before uploading.
 +
* '''DNA data is different.''' It implicates biological relatives who never tested. Review privacy and match-visibility settings on every service, and revisit them after policy changes.
 +
* '''Free before paid.''' FamilySearch is free and permanent. Exhaust it before subscribing.
 +
* '''Subscription drift.''' Photo animation and enhancement credits are the most common upsell. They are also the least research-critical feature.
 +
* '''Export regularly.''' Keep a GEDCOM backup and local copies of original scans. Do not let a subscription service be your only archive.
 +
 
 +
== Emerging Risks ==
 +
 
 +
* '''Synthetic ancestors.''' Generated portraits and animations of real deceased people circulate quickly through families and social media, and are rapidly mistaken for authentic photographs. Label them.
 +
* '''Fabricated sources at scale.''' AI-written genealogy content with invented citations is now common enough that secondary sources online require more scrutiny than they used to.
 +
* '''Tree pollution.''' Collaborative trees are vulnerable to confident, unverified AI-generated additions.
 +
* '''Pressure on volunteer infrastructure.''' Much of free genealogy depends on volunteer indexers, moderators, and maintainers. Automated submissions and AI-generated content increase their review burden — a structural concern raised within the community during 2026.
 +
 
 +
== Under the Hood ==
 +
 
 +
=== Handwritten Text Recognition ===
 +
Modern HTR moved beyond character-level OCR to sequence models that read whole lines in context. Domain adaptation is the key advantage for genealogy: a model fine-tuned on one parish's registers, or one clerk's hand, dramatically outperforms a general model on that specific collection. This is why a trainable tool can beat a larger commercial platform on unusual material.
 +
 
 +
=== Record Linkage ===
 +
Natural language processing parses census returns, vital registrations, and parish registers, then links individuals across disparate datasets using name variants, ages, locations, and household composition. The output is a probabilistic candidate set, not an identification — which is precisely the step where home users most often over-trust the machine.
 +
 
 +
=== Genetic Analysis ===
 +
Machine learning imputes missing genetic markers from incomplete samples and assigns haplogroups by pattern-matching against reference panels. Relationship estimation from shared segments is statistical inference over recombination, producing likelihood distributions across possible relationships rather than single answers.
 +
 
 +
== Professional and Research Tier ==
 +
 
 +
{| class="wikitable"
 +
! Category !! Focus !! Typical tools
 +
|-
 +
| Consumer-facing || Ease of use, tree building, record matching, DNA matching, photo tools || Ancestry, FamilySearch, MyHeritage, 23andMe, DNA Painter, Transkribus
 +
|-
 +
| Institutional / archival || Bulk digitization, HTR at archive scale, structured extraction || Transkribus collections, custom HTR pipelines, NARA-style AI indexing (used for the 1950 US Census)
 +
|-
 +
| Bioinformatics || High-throughput sequencing, variant calling, ancestral inference, ancient DNA || GATK (Genome Analysis Toolkit), PLINK, ADMIXTURE, population-genetics pipelines
 +
|}
 +
 
 +
= Learning Resources =
 +
 
 +
* [https://www.familysearch.org/en/blog/ai-developments-genealogy FamilySearch: AI and Genealogy — Advancements You Can Use] — the platform's own guidance, including limitations and responsible use
 +
* [https://www.familysearch.org/en/rootstech/ RootsTech] — the largest family history conference; virtual attendance is free and the session archive includes extensive AI instruction
 +
* [https://aigenealogyinsights.com/ AI Genealogy Insights] (Steve Little) — ongoing practical coverage, prompts, and a plain-language glossary of AI terminology for genealogists
 +
* [https://familylocket.com/ Family Locket] — annual roundups of handwriting-recognition tools and the ''Research Like a Pro with AI'' methodology
 +
* [https://vitabrevis.americanancestors.org/-ai-in-genealogical-research American Ancestors: Limitations of AI in Genealogical Research] — a sober account of hallucination in practice
 +
 
 +
== Video Tutorials: Low-Cost and Free AI Tools ==
 +
 
 +
''Dates reflect publication as reported at the time of writing. Tutorials tied to a specific model version go stale quickly — verify before relying on version-specific steps. Note also that many genealogy channels carry affiliate links to the subscription services they review.''
 +
 
 +
=== FamilySearch Full-Text Search (free) ===
 +
 
 +
The highest-payoff free AI tool for the home researcher, and the best-documented.
 +
 
 +
'''How To Find Your Ancestors with Full-Text Search''' — FamilySearch official, May 2026. The place to start.<br>
 +
<youtube>8KGBbBvrqAs</youtube>
 +
 
 +
'''Step-by-Step GUIDE to FamilySearch Full-Text Searching''' — April 2026. Finding ancestors in handwritten records without waiting for an index.<br>
 +
<youtube>mZc2-Z-TROw</youtube>
 +
 
 +
'''DON'T MISS this IMPORTANT Full Text Search Update 2026''' — July 2026. Covers a change to how the tool handles searches; the most recent tutorial listed here.<br>
 +
<youtube>gX-d5Szg9XY</youtube>
 +
 
 +
'''How to Use FamilySearch's Simple Search for Easier Full-Text Searching''' — Kathryn Grant webinar, April 2026. Longer and more methodical than the short-form tutorials.<br>
 +
<youtube>_szy7VbYPjE</youtube>
 +
 
 +
'''Most Family Historians are Missing this Tool… Are YOU?''' — April 2026 interview with FamilySearch on how the tool was built and where it is going. Background rather than how-to.<br>
 +
<youtube>D3cFvqNZUnY</youtube>
 +
 
 +
'''What is FamilySearch's Full-Text Search?''' — August 2025. Concise overview.<br>
 +
<youtube>hpWf-VxjtyE</youtube>
 +
 
 +
* [https://www.youtube.com/playlist?list=PLC6xLbn7JeDmqpgT9mW6SBrjdCuutlsRc FamilySearch Full-Text Search playlist] — ongoing collection
 +
 
 +
=== General-Purpose Assistants (free tiers) ===
 +
 
 +
'''The 4 Most Useful 2026 AI Tools for Genealogy''' — December 2025. Organized around which tool suits which task rather than a feature tour.<br>
 +
<youtube>hyD3Q0sz5Kg</youtube>
 +
 
 +
'''How to Use ChatGPT for Genealogy Research — Easy Guide 2026''' — April 2026.<br>
 +
<youtube>a6saa-yYlCo</youtube>
 +
 
 +
'''Must Use ChatGPT 5 Features for Family History''' — August 2025. Useful method, dated on version specifics.<br>
 +
<youtube>EBNHm6ecSyw</youtube>
 +
 
 +
'''Getting Started: The 3 Free AI Tools I Use for Genealogy Research''' — September 2024. Older, but a clean beginner's introduction to Gemini, Perplexity, and ChatGPT.<br>
 +
<youtube>Q8oJTK3sDB0</youtube>
 +
 
 +
* [https://denyseallen.substack.com/p/practical-ai-toolkit-genealogy-2026-chatgpt-claude-gemini-perplexity Practical AI Toolkit for Genealogy (2026)] — May 2026, video embedded in the post. Divides the work four ways: planning, transcription, writing, and record location.
 +
 
 +
=== Handwriting Transcription (free route) ===
 +
 
 +
* [https://ancestorsandai.buzzsprout.com/2525649/episodes/18422484-ai-for-genealogy-how-to-transcribe-handwritten-genealogy-documents-with-free-ai-gemini-3-tutorial Ep. 20: Transcribe Handwritten Genealogy Documents with Free AI — Gemini 3 Tutorial] — 35 minutes, December 2025. Covers free access via Google AI Studio, copy-ready prompts for pension records, wills, deeds, letters, and Bible records, and when verification matters for proof standards. '''Caveat:''' its headline error-rate figures come from the creator's own testing, not independent benchmarking, and other 2026 comparisons report noticeably different numbers. The method is sound; treat the accuracy claims as unsettled.
 +
* [https://aigenealogyinsights.com/prompts/ AI Genealogy Insights — Prompt Library] — free, Creative Commons licensed, tested prompts emphasizing diplomatic transcription that preserves the page exactly and marks uncertain readings. A good companion to any of the above.
 +
 
 +
=== Paid, but Worth Knowing ===
 +
 
 +
* [https://familytreewebinars.com/webinar/top-5-ai-tools-for-transcribing-handwritten-historical-documents Top 5 AI Tools for Transcribing Handwritten Historical Documents] — Nicole Dyer, Legacy Family Tree Webinars, September 2026. Membership required. Compares ChatGPT, Claude, Gemini, Archive Studio, and Handwriting OCR, and teaches prompting strategies that get a model to flag its own uncertainties for human review — the most valuable technique on this list and poorly covered elsewhere.
 +
<br>
 +
<youtube>-OlZYLs410E</youtube>

Revision as of 14:40, 11 September 2026

YouTube ... Quora ...Google search ...Google News ...Bing News


Overview

Between 2024 and 2026, AI moved from an experimental add-on to standard equipment in consumer genealogy. The major platforms now run AI record matching by default, handwriting recognition has made billions of previously browse-only document images searchable, and general-purpose chatbots have become common research companions.

For the home researcher, the useful question is no longer whether to use AI, but which tool for which job — and how to keep an eager, confident machine from quietly corrupting a family tree. No single tool does everything well. The gains come from matching a specific tool to a specific stage of research, and from verifying everything it produces.

This page is organized around that practical framing. The technical and professional material has been kept but moved toward the end.

Quick Start: A Free-First Workflow

A home researcher can get a long way without paying anything. A sensible order of operations:

  1. Records first. Build the skeleton of the tree on FamilySearch (free, nonprofit, billions of records and one shared collaborative world tree). Do this before spending money anywhere.
  2. Then unindexed images. Run names through FamilySearch Full-Text Search, which uses handwriting recognition to surface records that traditional name indexes never captured — deeds, probate files, land records, and court papers where ancestors are mentioned but not indexed.
  3. Then the hard handwriting. For documents the platforms can't read — German Kurrent, old Italian scripts, eighteenth-century legal hands — use a dedicated HTR tool such as Transkribus. For a single census line on your phone, Google Lens or Apple Live Text is often enough.
  4. Then DNA. Test, then upload to free comparison sites and use match-analysis tools to organize clusters before drawing conclusions.
  5. Then photos. Restoration, colorization, dating, and animation are the most emotionally rewarding stage and the least evidentiary. Do them last.
  6. Then write it up. Use a general assistant to turn verified facts into readable narrative — never to supply the facts themselves.

Budget guidance: most active home researchers can work effectively at $0–25 per month. A paid subscription starts paying for itself only once you are researching most weeks.

AI Tools by Task

Records and Search

Tool What the AI does Cost Notes for home users
FamilySearch AI-indexed records, Full-Text Search over handwriting, help chatbot, and an AI Research Assistant that surfaces "tree-extending hints" on the home page Free The strongest free offering. AI does a first pass on records; volunteers check its work through Get Involved
Ancestry Hints and ThruLines surface likely records and connect your tree to other users' trees; automatic transcription and tagging make handwriting searchable Subscription (roughly $25–50/month by plan) Hints become genuinely useful once a tree passes about 50 people. An "Ask AncestryAI" feature appears on some accounts from the LifeStory tab, currently limited to preset questions
MyHeritage Automatic Record Matches, AI Record Finder™ (chat-based record search), AI Biographer™, Scribe AI document analysis, Cousin Finder, Theory of Family Relativity Free tier covers trees up to 250 people; full access via subscription (all-inclusive plans around $299/year, first-year discounts common) Strongest international and European record coverage of the big three. A broader research assistant, GAIA, was demonstrated at RootsTech 2026 and beta-tested that spring; check current availability
Findmypast Tree-linked record hints, plus searchable text recognition across the British Newspaper Archive, which it runs in partnership with the British Library Family Tree tier £9.99/month, Everything tier £24.99/month (prices in GBP) The pick for British, Irish and Commonwealth lines. Important caveat: the cheaper Family Tree tier surfaces records only through hints — you cannot search the collections independently. Over a million newspaper pages are free to view with a free account via the "Free To View" access filter
WikiTree Almost nothing, deliberately. No AI hints and no automated tree-matching; GEDCompare flags likely duplicates on upload, but conclusions get argued out by people on the profile page Free, always Useful as a check on the other three. Roughly 1.2 million members maintaining one shared tree of 40+ million profiles, with members signing an Honor Code that requires citing sources. GEDCOM uploads are capped at 5,000 people. FamilyTreeDNA added direct tree linking in March 2026; about 15 million profiles carry DNA connections
Goldie May Third-party research assistant that runs searches across FamilySearch, Ancestry, MyHeritage, Newspapers.com, GenealogyBank and Storied, then evaluates which results are most promising; also drafts citations, transcribes handwriting, and lets you choose among several underlying models Subscription — verify current tiers Sits on top of subscriptions you already hold rather than replacing them. States that user data is not used for model training

Handwriting and Transcription

Tool Best for Cost Caveat
Transkribus Difficult, old, or uncommon scripts; can be trained on a specific clerk's hand or a particular record set Credit-based; limited free tier Makes ordinary recognition errors but does not hallucinate — it will not invent text that isn't there. Accuracy varies sharply by document type
Google Lens / Apple Live Text A single record photographed on a phone Free Surprisingly capable for casual use; no batch processing
Platform-native HTR (FamilySearch, MyHeritage Scribe AI, Kindex, Goldie May) Documents already inside a platform workflow Varies Convenient, but generalized models tuned for volume rather than for unusual scripts
General chatbots (ChatGPT, Claude, Gemini) Reading a page and explaining archaic terms, abbreviations, and legal formulae Free tiers available Will confidently misread and then smooth over the gap. Useful for interpretation, risky for raw transcription

DNA Match Analysis

Consumer DNA testing itself is largely statistical rather than generative AI, but the analysis layer has become increasingly automated.

  • Testing companies: AncestryDNA (largest database), MyHeritage DNA (strong European matching), 23andMe.
  • Match organization: DNA Painter offers chromosome mapping plus tools for testing hypothetical relationships, which is particularly useful for unknown-parentage cases. Free tier available; a paid tier (around $65/year) adds clustering.
  • Relationship prediction: MyHeritage's cM Explainer and similar tools estimate likely relationships from shared centimorgans.
  • A caution: shared DNA constrains the range of possible relationships; it rarely proves one. AI tools that name a specific relationship are presenting a probability as a conclusion.

Photographs

Feature Platform What it does
Photo Enhancer, Photo Repair, MyHeritage In Color™ MyHeritage / Reimagine app Sharpens, repairs damage, and colorizes. Free uses are limited, then subscription
PhotoDater™ MyHeritage Estimates when a photo was taken from clothing, hairstyles, photographic style, and paper stock. Free. Can miss by a decade or more on unusual photos, but narrows an otherwise open search window
Deep Nostalgia™, LiveMemory™ MyHeritage Animates faces or whole scenes into short video clips. Premium after a small number of free uses
Photo Tagger MyHeritage Face clustering across a photo collection, with human confirmation required
Remini, VanceAI, Topaz Third-party Restoration and upscaling; often stronger on badly damaged originals

Important distinction. Enhancement, colorization, and animation are interpretations, not evidence. Colorized clothing is a guess. An animated ancestor is a generated fiction of a real person. Keep the original scan, label derivatives clearly, and never let an enhanced image become the archival copy.

General-Purpose Assistants

ChatGPT, Claude, Gemini, and Perplexity are all used widely by genealogists. The practical differences between them are real but narrowing, and the skills transfer. Experienced practitioners generally advise picking one, using it for a month, and getting fluent rather than tool-hopping.

Good uses:

  • Building a research plan or locality guide
  • Summarizing a long document you have already obtained
  • Explaining archaic legal, occupational, or medical terminology
  • Translating older foreign-language records
  • Drafting an ancestor biography from facts you supply
  • Improving your own prompts — ask the model how to rephrase a request that isn't working

Poor uses:

  • Asking for records, citations, or sources directly
  • Asking who someone's parents were
  • Anything where you cannot check the output against an original

What AI Does Well, and Where It Fails

Reliable strengths: searching enormous corpora quickly, reading handwriting at scale, translating, clustering and organizing, summarizing documents you already hold, and generating readable prose from verified facts.

Persistent weaknesses:

  • Hallucination. Models invent plausible-sounding names, dates, and entire citations. A model may correctly identify where a record ought to be and then fabricate the record itself. This is the single largest risk to a home researcher's tree.
  • No concept of truth. AI cannot distinguish a correct inference from a confident guess, and will fill a data gap rather than report it.
  • Missing cultural and historical context. Naming conventions, migration patterns, religious practice, and jurisdictional boundary changes are routinely mishandled.
  • Silent propagation. An unverified AI suggestion merged into a shared collaborative tree spreads to every researcher downstream and is difficult to unwind.

Verification Discipline

The consensus rule across FamilySearch, professional genealogists, and experienced AI-genealogy educators is the same: AI assists, you verify.

  1. Treat every AI output as a hint, never as a source.
  2. Follow the hint to the original record image and read it yourself.
  3. Cite the original record, not the AI.
  4. If a model gives you a citation, assume it may not exist until you locate it.
  5. Note "AI-assisted" in your research log for any step where a tool contributed. This preserves transparency for anyone reviewing your work later, including future you.
  6. Do not merge AI-suggested relationships into a shared public tree until they are independently supported.

One widely taught framework is AIM — Assess what the tool gives you, Interact to refine it, Measure it against your sources.

Privacy, Cost, and Data Ownership

  • Uploads are disclosures. Putting photographs, records, or DNA data into a commercial AI system means entrusting a company with sensitive material about living relatives who did not consent.
  • Living people. Avoid pasting details about living individuals into general-purpose chatbots. Redact before uploading.
  • DNA data is different. It implicates biological relatives who never tested. Review privacy and match-visibility settings on every service, and revisit them after policy changes.
  • Free before paid. FamilySearch is free and permanent. Exhaust it before subscribing.
  • Subscription drift. Photo animation and enhancement credits are the most common upsell. They are also the least research-critical feature.
  • Export regularly. Keep a GEDCOM backup and local copies of original scans. Do not let a subscription service be your only archive.

Emerging Risks

  • Synthetic ancestors. Generated portraits and animations of real deceased people circulate quickly through families and social media, and are rapidly mistaken for authentic photographs. Label them.
  • Fabricated sources at scale. AI-written genealogy content with invented citations is now common enough that secondary sources online require more scrutiny than they used to.
  • Tree pollution. Collaborative trees are vulnerable to confident, unverified AI-generated additions.
  • Pressure on volunteer infrastructure. Much of free genealogy depends on volunteer indexers, moderators, and maintainers. Automated submissions and AI-generated content increase their review burden — a structural concern raised within the community during 2026.

Under the Hood

Handwritten Text Recognition

Modern HTR moved beyond character-level OCR to sequence models that read whole lines in context. Domain adaptation is the key advantage for genealogy: a model fine-tuned on one parish's registers, or one clerk's hand, dramatically outperforms a general model on that specific collection. This is why a trainable tool can beat a larger commercial platform on unusual material.

Record Linkage

Natural language processing parses census returns, vital registrations, and parish registers, then links individuals across disparate datasets using name variants, ages, locations, and household composition. The output is a probabilistic candidate set, not an identification — which is precisely the step where home users most often over-trust the machine.

Genetic Analysis

Machine learning imputes missing genetic markers from incomplete samples and assigns haplogroups by pattern-matching against reference panels. Relationship estimation from shared segments is statistical inference over recombination, producing likelihood distributions across possible relationships rather than single answers.

Professional and Research Tier

Category Focus Typical tools
Consumer-facing Ease of use, tree building, record matching, DNA matching, photo tools Ancestry, FamilySearch, MyHeritage, 23andMe, DNA Painter, Transkribus
Institutional / archival Bulk digitization, HTR at archive scale, structured extraction Transkribus collections, custom HTR pipelines, NARA-style AI indexing (used for the 1950 US Census)
Bioinformatics High-throughput sequencing, variant calling, ancestral inference, ancient DNA GATK (Genome Analysis Toolkit), PLINK, ADMIXTURE, population-genetics pipelines

Learning Resources

Video Tutorials: Low-Cost and Free AI Tools

Dates reflect publication as reported at the time of writing. Tutorials tied to a specific model version go stale quickly — verify before relying on version-specific steps. Note also that many genealogy channels carry affiliate links to the subscription services they review.

FamilySearch Full-Text Search (free)

The highest-payoff free AI tool for the home researcher, and the best-documented.

How To Find Your Ancestors with Full-Text Search — FamilySearch official, May 2026. The place to start.

Step-by-Step GUIDE to FamilySearch Full-Text Searching — April 2026. Finding ancestors in handwritten records without waiting for an index.

DON'T MISS this IMPORTANT Full Text Search Update 2026 — July 2026. Covers a change to how the tool handles searches; the most recent tutorial listed here.

How to Use FamilySearch's Simple Search for Easier Full-Text Searching — Kathryn Grant webinar, April 2026. Longer and more methodical than the short-form tutorials.

Most Family Historians are Missing this Tool… Are YOU? — April 2026 interview with FamilySearch on how the tool was built and where it is going. Background rather than how-to.

What is FamilySearch's Full-Text Search? — August 2025. Concise overview.

General-Purpose Assistants (free tiers)

The 4 Most Useful 2026 AI Tools for Genealogy — December 2025. Organized around which tool suits which task rather than a feature tour.

How to Use ChatGPT for Genealogy Research — Easy Guide 2026 — April 2026.

Must Use ChatGPT 5 Features for Family History — August 2025. Useful method, dated on version specifics.

Getting Started: The 3 Free AI Tools I Use for Genealogy Research — September 2024. Older, but a clean beginner's introduction to Gemini, Perplexity, and ChatGPT.

Handwriting Transcription (free route)

  • Ep. 20: Transcribe Handwritten Genealogy Documents with Free AI — Gemini 3 Tutorial — 35 minutes, December 2025. Covers free access via Google AI Studio, copy-ready prompts for pension records, wills, deeds, letters, and Bible records, and when verification matters for proof standards. Caveat: its headline error-rate figures come from the creator's own testing, not independent benchmarking, and other 2026 comparisons report noticeably different numbers. The method is sound; treat the accuracy claims as unsettled.
  • AI Genealogy Insights — Prompt Library — free, Creative Commons licensed, tested prompts emphasizing diplomatic transcription that preserves the page exactly and marks uncertain readings. A good companion to any of the above.

Paid, but Worth Knowing

  • Top 5 AI Tools for Transcribing Handwritten Historical Documents — Nicole Dyer, Legacy Family Tree Webinars, September 2026. Membership required. Compares ChatGPT, Claude, Gemini, Archive Studio, and Handwriting OCR, and teaches prompting strategies that get a model to flag its own uncertainties for human review — the most valuable technique on this list and poorly covered elsewhere.


Retrieved from "https://primo.ai/index.php?title=Genealogy&oldid=43715"