AI is moving fast, but there were only a handful of genuinely new,
broadly relevant items reported in the last 24 hours, followed by a
steady drumbeat of platform refinements.
The big pattern is long‑context reasoning becoming normal instead of exotic. Models from OpenAI, Anthropic, Google, DeepSeek, and SubQ are all pushing context windows into the hundreds of thousands of tokens and beyond, which means you can realistically load entire research folders, multi‑generation research logs, or book‑length county histories into one sustained conversation instead of chopping them into pieces.[mindstudio]
At the same time, grounded, cited answers are improving, especially in Google Gemini and in tools that layer models over live web search (like Perplexity). For genealogy, that translates to more reliable “where are the records?” help: instead of hallucinated collections, you increasingly get links to real archive catalogs, digital collections, and institution homepages, plus a clearer distinction between primary sources and derivative summaries.[docs.perplexity]
Finally, we’re seeing more agent‑style features like Perplexity’s Personal Computer and Claude/MAI/GPT agents that can orchestrate across local files, cloud drives, and the web. For working genealogists, this means you can begin to treat AI as an assistant that not only explains a deed abstract but also pulls the next five deeds in the series, renames files according to your schema, and drafts a research plan in your preferred template.[youtube][docs.perplexity]
Plug‑and‑play AI micro‑workflows you can try today
Below are twenty very concrete workflows tied directly to the releases above; they assume you’re comfortable with tools like Zotero, RootsMagic, and AI chat interfaces.[familytreewebinars][youtube][grip.ngsgenealogy]
One‑conversation “whole project” review (SubQ or DeepSeek V4)
Use an ultra‑long‑context open‑weight model (e.g., SubQ or DeepSeek V4‑Flash) via a notebook or local UI to ingest your entire research log, a cluster of county histories, and all transcribed records for one problem ancestor.[felloai]
Prompt: “Given everything loaded in this session, produce: a timeline, a list of conflicts, and a prioritized list of record types and jurisdictions to search next for [ancestor].”[denyseallen.substack]
Multigenerational FAN‑club scan (Claude Opus 4.8 / Sonnet 4.7)
Paste or upload a long bundle of deeds, tax rolls, and probate abstracts into Claude Sonnet 4.7 or Opus 4.8 and ask it to extract every recurring neighbor, witness, bondsman, and administrator.[youtube][felloai]
Follow‑up: “Cluster these by surname and geography; propose likely kinship or social network patterns to investigate.”[denyseallen.substack]
Repository‑finding via grounded Gemini 3.1 / AI Overviews
Use Gemini with AI Overviews enabled to search for specific record sets: “Original probate packet series for Muskogee County, Oklahoma, 1907–1930, and how to access them.”[mindstudio]
Check the “Highly Cited” and “Preferred Sources” indicators to focus on archive catalogs, state archives, and major genealogy sites.[evertune]
“What records exist for this place/time?” (Perplexity + Gemini)
Ask Perplexity or Gemini: “List likely record types and repositories for African American families in [county], [state], 1870–1910,” then confirm the output by following cited links to archive guides.[youtube][docs.perplexity]
Use the resulting list as a checklist in your research log.
Automated research log diagnosis (GPT‑5.5 / Claude)
Paste a long research log into GPT‑5.5 Thinking or Claude Sonnet 4.7.[felloai]
Prompt: “Summarize research completed, identify record types not yet searched, highlight conflicting evidence, and recommend next steps, formatted as a research plan for [surname] in [county, state].”[grip.ngsgenealogy]
Cross‑model sanity check for a tough brick wall (Perplexity Model Comparison)
Use Perplexity’s multi‑model comparison tooling to send the same brick‑wall summary to GPT‑5.5, Claude Opus, Gemini 3.5 Pro, and an open‑weight model.[docs.perplexity]
Compare: where do they agree on next steps? Where do they hallucinate repositories that don’t exist? Use consensus only when confirmed with real catalogs.[mindstudio]
Probate packet narrative builder (GPT‑5.5 / Claude)
After abstracting a probate file, feed the abstracts to GPT‑5.5 Pro and ask: “Write a concise research narrative that explains the probate sequence, all heirs and their relationships, and any inferred migrations, with a separate section listing unresolved questions.”[youtube][familytreewebinars]
Use this as a draft for your research report, then manually verify every assertion.
Quick handwriting triage with Gemini + images
Capture images of difficult 19th‑century probate or deed pages and feed them to Gemini 3.x multimodal.[youtube][mindstudio]
Ask for a rough reading and a list of key names, dates, and places; treat it as a helper, then refine transcription manually.[familytreewebinars][youtube]
Grounded “where is this church register?” hunt (Gemini + AI Overviews)
Search: “Physical location and access instructions for [denomination] church registers in [county/state] 1880–1910.”[mindstudio]
Use the AI Overview citations to navigate to archive catalogs, diocesan archives, or microfilm notes rather than relying on the summary itself.[evertune]
Local‑file orchestrated cleanup (Perplexity Personal Computer)
On a machine with Perplexity Personal Computer enabled, point it at a folder of downloaded PDFs and images for one project.[docs.perplexity][youtube]
Task it to: rename files according to your naming convention, move them into case‑specific subfolders, and produce an inventory list (filename, type, jurisdiction, date) in CSV format.[grip.ngsgenealogy]
Timeline + gap analysis over 1M tokens (DeepSeek V4 / SubQ)
Load a full lifetime of records (vital, census, land, court, city directories) for a single person into a 1M‑context model.[evertune]
Prompt: “Build a detailed timeline, then list chronological gaps of five years or more with suggested record types to fill them, focusing on [jurisdiction].”[denyseallen.substack]
Side‑by‑side narrative drafts (GPT‑5.5 vs Claude vs Gemini)
Give the same structured facts (names, dates, places, occupations) to GPT‑5.5, Claude Sonnet 4.7, and Gemini 3.5 Pro, asking each to draft a 500‑word life sketch.[youtube][mindstudio]
Compare tone and factual discipline; pick the best as a starting draft and then annotate it with citations from your RootsMagic database.[youtube][grip.ngsgenealogy]
DNA correspondence template generator (GPT‑5.5 / MAI)
Ask GPT‑5.5 Instant or a Microsoft MAI model to draft outreach messages to DNA matches: “Given this match’s predicted relationship and shared segment data, draft a clear, non‑intrusive message that asks about ancestors from [place].”[felloai]
Save templates in your correspondence log and adapt per match.
Local history chapter digests (open‑weight long‑context)
Load an entire county history or tribal history volume into a long‑context open‑weight model (DeepSeek V4‑Pro or SubQ).[felloai]
Ask for chapter‑by‑chapter digests focusing only on your target surname, township, or tribal group, then export summaries into Zotero notes.[familytreewebinars]
Record‑type teaching aids for classes (GPT‑5.5 / Gemini)
For a workshop, feed a short description of a record set (e.g., Dawes packets, Oklahoma probate dockets) into GPT‑5.5 and ask it to generate a one‑page handout explaining what the records are, what they typically contain, and how to cite them.[thefhguide]
Optionally, use Gemini 3.1 Flash TTS to generate an audio version for students who prefer listening.[youtube]
Micro‑lesson: AI record critique (Claude / Gemini)
Take one AI‑generated “summary of a record” (from Claude or GPT) and, in class, have students compare it to the original image.[youtube][familytreewebinars]
Use Claude or Gemini to highlight omissions, misinterpretations, and unproven inferences, reinforcing the difference between AI assistance and sound proof arguments.[grip.ngsgenealogy]
Automated catalog search scripts (DeepSeek / MiniMax / MAI)
Using an open‑weight model that handles code and long context (DeepSeek V4, MiniMax M3, or a MAI model), have it generate Python scripts that query online catalog APIs (where allowed) for recurring surname/place combinations.[youtube][felloai]
Run scripts locally to build a running list of call numbers and online collections for a surname across multiple repositories.[grip.ngsgenealogy]
Multi‑modal case file “explainer” (Gemini 3.5 Pro)
Combine text notes, images of records, and maps in one Gemini session.[mindstudio]
Prompt: “Explain this case to a fellow genealogist: summarize the key evidence, identify conflicts, and show how the geography changes over time.”[youtube][familytreewebinars]
Triaged “next best action” list (Grok R1 / GPT‑5.4 Thinking)
Feed a narrative brick‑wall summary into Grok R1 and GPT‑5.4 Thinking.[evertune]
Ask both: “List the next five concrete actions, each with a specific repository, record type, time frame, and rationale; avoid speculative leaps.” Compare results and adopt only those you can verify in catalog systems.[mindstudio]
Cross‑tool “AI for genealogy” syllabus refresh (Perplexity + Model Tracker)
Use Perplexity plus an AI model tracker (like Evertune’s) to list current top models and their strengths (long context, multimodal, cost).[docs.perplexity]
Map these to your course topics—for example, “Claude Sonnet for transcriptions, GPT‑5.5 for writing, Gemini for locating repositories, open‑weight SubQ for archival‑scale projects”—and update your workshop syllabus accordingly.[youtube][grip.ngsgenealogy]
20+ concrete AI use cases for genealogists
Each of these is phrased so you could drop it straight into your own workflow or blog as an illustrated example. They are grounded in practices already being demonstrated by working genealogists and AI‑for‑genealogy educators.
1. Summarize a long probate or pension file
Upload or paste a full probate packet or pension application and ask the model to produce a structured summary: parties, relationships, dates, places, property, and key inferred relationships.
Then request a second output: a plain‑language paragraph you can paste into a research log or report, making sure you verify each point against the original images.
2. Extract names, dates, and relationships into tables
Give AI a transcription or good OCR of a will, deed series, or church‑record page and ask it to list all persons, roles, dates, and locations in a table you can paste into Excel or Google Sheets.
This is particularly helpful with compiled genealogies or county histories where people and relationships are buried in dense prose.
3. Clean up OCR and messy text
Paste messy OCR from digitized books, newspapers, or city directories and ask AI to correct spelling, line breaks, and obvious formatting issues while preserving all names and numbers exactly.
You can then run a second pass asking it to highlight candidate personal names, organizations, and place names for verification.
4. Transcribe historical handwriting
Use a dedicated handwriting tool (e.g., Transkribus) or a vision‑enabled general LLM to produce a draft transcription of a difficult will, deed, or church register entry.
Follow by asking the model to mark uncertain words, suggest alternatives, and explain letter‑forms by comparison to other words on the page.
5. Translate foreign‑language records
Paste a parish record, civil registration entry, or notarial act in German, Latin, French, Spanish, or another language and ask for both a literal translation and a genealogist‑friendly paraphrase.
Prompt it to keep all names and numbers unchanged and to flag abbreviations and formulaic phrases typical of that record type.
6. Build structured research timelines
Export a person’s facts from your online tree (Ancestry, FamilySearch, MyHeritage) or from RootsMagic, then paste the timeline into AI and ask it to normalize dates, sort events, and add columns for “evidence source” and “research gaps.”
You can instruct the model to highlight chronological conflicts and possible identity problems in red‑flag notes.
7. Ask for next‑step research suggestions
Give AI your current objective, the jurisdiction, a short timeline, and a list of records you have already searched; then ask for additional record types, repositories, or strategies you may have missed.
This “thinking partner” role is now widely recommended by genealogists experimenting with AI: you still decide what is plausible, but it often surfaces fresh angles.
8. Draft research plans and work logs
Ask AI to convert a narrative brick‑wall description into a structured research plan with clear tasks, prioritized record types, and specific jurisdictions.
You can also have it convert a messy research journal into a dated log with columns for task, repository, result, and follow‑up actions.
9. Turn raw findings into narrative reports
Paste your research notes and citations and have AI produce a first‑draft narrative report or proof argument in your preferred voice, with explicit sections for background, evidence, analysis, and conclusion.
Many genealogists are using AI to handle structure and sentence‑level polish, then revising to ensure every assertion is source‑backed and accurately interpreted.
10. Create ancestor biographies and blog posts
Provide a timeline of events with sources, plus a short paragraph about local context, and ask AI to draft a readable ancestor biography suitable for a blog or society newsletter.
You can instruct it to keep speculative language clearly marked and to include a “Research Notes” section that distinguishes proven from hypothesized relationships.
11. Generate locality and historical context briefs
Ask AI for a concise overview of a town, county, or tribal jurisdiction during a specific time frame, including boundary changes, major migrations, and predominant record types.
Then have it list context questions tailored to your research objective, such as “What land policies applied in this county in the 1880s?” as prompts for further archival digging.
12. Analyze clusters and FAN clubs
Paste a list of neighbors, witnesses, bondsmen, and godparents across multiple records and ask AI to group them by surname, location, and recurring associations.
You can then request a narrative interpretation of possible kinship or community patterns, which you can test against maps and additional record searches.
13. Land description parsing and plat hints
Provide metes‑and‑bounds land descriptions or series of deeds and ask AI to normalize the text, identify repeated landmarks, and list adjoining landowners over time.
Some genealogists are also prompting AI to outline step‑by‑step instructions for manually platting the land in a spreadsheet or GIS tool, even if the AI cannot draw the plat itself.
14. Relationship and pedigree consistency checks
Feed AI a family group sheet or short register‑style outline and ask it to highlight improbable ages, chronological conflicts, and relationship contradictions.
You can have it produce a list of “consistency checks” you should perform in your genealogy software (e.g., parents too young/too old, children born after parent’s death).
15. Teaching aids and class handouts
For a Sunday afternoon genealogy class or society presentation, ask AI to generate simple definitions, analogies, and one‑page handouts explaining concepts like “reasonably exhaustive research” or “FAN club,” which you then edit for accuracy.
Instructors are also using AI to turn syllabi or slide decks into plain‑language summaries and assignment sheets for adult‑education learners.
16. Slide decks and lesson outlines
Paste your outline for a workshop (e.g., “Using AI in Probate Research”) and ask AI to propose slide titles, key bullet points, and timing estimates for a 50‑minute lecture.
You can then request example prompts and before/after text snippets to demonstrate in class, making it easier to build live demos.
17. Source citation scaffolds (with manual review)
Provide AI with a full description of a record (collection title, archive, film, image number, page, entry) and ask it to generate a draft citation in Evidence Explained–style prose, which you then correct.
This can be handy when teaching beginners: show them AI’s draft, then walk through what needs to be fixed and why.
18. Research log normalization and tagging
Paste free‑form research notes and ask AI to break them into discrete log entries with date, repository or website, search terms, and whether the search was positive or negative.
You can then request keyword tags for each entry (surname, locality, record type) to help you migrate the log into Zotero, Notion, or your preferred system.
19. Training data for students: practice problems
Ask AI to invent realistic but clearly labeled “fictional” record sets (census entries, deeds, probate abstracts) that you can use as exercises in a class or blog post.
Have it include built‑in conflicts or ambiguous clues so students must practice correlation and conflict resolution.
20. Image‑based headstone analysis
Use an AI‑powered image tool to read inscriptions from gravestone photos, enhance faint lettering, and propose transcriptions, especially when erosion or lichen makes text difficult.
Some experimental tools also attempt to interpret common symbols or iconography, which you can then verify against epigraphic reference works.
21. AI‑indexed collections as discovery aids
On FamilySearch, test AI‑indexed collections by searching for variant spellings and then comparing the machine‑read index entry with the original image.
Use the AI index as a pointer only, then construct your own citations and extractions from the digitized record itself.
22. Long‑document “single‑pass” review
With newer long‑context models (like GLM‑5.2 and others), you can load an entire 200‑page county history or compiled genealogy in one session and ask for every mention of a specific surname and locality.
Follow up by having AI list all candidate passages you must read closely, with page references, instead of trying to skim the whole volume visually.
No comments:
Post a Comment