Here’s today's AI briefing tailored for working genealogists and family historians, focused on what actually changed and how you can use it this week.
A. Named releases & features (last 48–72 hours)
OpenAI · ChatGPT model lineup change (GPT‑5.1 retirement) – As of early June 2026, GPT‑5.1 Instant, Thinking, and Pro are no longer available in ChatGPT, with newer 5‑series and o‑series reasoning models now carrying the load.[help.openai]
Anthropic · Claude Opus 4.8 launch – New top‑end Claude model with better judgment, stronger long‑running “agentic” workflows, and cheaper, faster “fast mode” for heavy tasks, plus new effort‑control and dynamic workflows features.[anthropic]
Anthropic · Opus 4 / 4.1 retirement in Claude UI – Older Opus 4 and 4.1 models have been removed from the Claude selector and Claude Code, consolidating users onto Opus 4.7+ and 4.8.[support.claude]
Anthropic · Effort control slider in Claude.ai & Cowork – You can now explicitly tell Claude how hard to “think,” trading speed vs depth per task (all plans).[anthropic]
Anthropic · Dynamic Workflows in Claude Code (research preview) – Claude can now spin up hundreds of parallel sub‑agents in one run, verify results, and return synthesized outputs for large, complex jobs.[anthropic]
Google · Gemini Apps – June shutdown of older 2.0 models – Gemini 2.0 Flash and other early 2.0 models are being shut down in favor of the newer Gemini 3.x models; API release notes now point users to Gemini 3.1/3.2 Flash and Pro.[ai.google][youtube][gemini]
Google · Gemini 3.2 Flash emerging in apps – A faster, cheaper Flash‑tier model is quietly appearing in the Gemini model picker, approaching 3.1 Pro quality at a fraction of the price for routine tasks.[youtube][ai.google]
xAI · Grok 4.20 Beta 2 as current flagship – Grok 4.20 Beta 2 remains the current public model with better instruction‑following, fewer hallucinations, and improved LaTeX and image handling while Grok 5 is still in Q2 testing.[nxcode]
xAI · Grok model retirement notice (API) – Several older Grok 4‑series “fast” reasoning endpoints were officially retired from the xAI API in mid‑May, routing traffic to newer Grok variants.[docs.x]
Perplexity · Computer feature widely available to Pro users – Perplexity Computer (always‑on agent running on a dedicated Mac mini or cloud) is now generally available to Pro subscribers and has just been upgraded with faster models and better enterprise connectivity.[perplexity]
Perplexity · Independent document‑audit pass – Computer can now perform an “audit” pass on documents to check for logical consistency, structure, and factual issues without rewriting the content.[perplexity]
Open‑weight · Mistral Large 2 in Snowflake Cortex – Mistral Large 2, with a larger 128K context window and improved reasoning, is now available for serverless inference in Snowflake’s Cortex AI environment (relevant if your tree/research data is sitting in Snowflake).[docs.snowflake]
Open‑weight · Mistral‑Small 3.2 (local‑friendly) – A fresh Mistral‑Small 3.2 release just hit Hugging Face and Ollama, offering a stronger, still‑lightweight model genealogists can run locally for private, offline tasks.[simonwillison]
B. Implications for genealogists this week
Claude Opus 4.8 and the new effort control slider make Claude a stronger “deep analysis” partner for long probate files, case studies, and knotty conflicting‑evidence problems. You can now reserve high‑effort mode for tough analysis and run quick, fast‑mode passes for simpler chores like drafting research logs or email summaries without burning through limits.[anthropic]
Gemini’s quiet shift away from 2.0 models and toward 3.x Flash/Pro means any saved workflows built on “Gemini 2.0 Flash” should be reviewed and updated to the 3.x equivalents, especially if you rely on scripts, notebooks, or API‑based tools. The emerging 3.2 Flash tier looks like the new “workhorse” option for image‑heavy tasks—like cemetery photos, map snippets, and form‑style records—where lower cost and speed matter.[ai.google][youtube]
On the tooling side, Perplexity Computer and Anthropic’s dynamic workflows are both early signals of more “agentic” research helpers that can operate across multiple files and sessions. For genealogists, that’s less about flashy demos and more about: letting an AI crawl your research corpus to build timelines, cross‑check sources, and flag logical contradictions before you publish. Meanwhile, Mistral‑Small 3.2 and Mistral Large 2 expand options for privacy‑sensitive or institutional projects where you want local or controlled‑environment AI over cloud consumer tools.[perplexity]
C. Plug‑and‑play AI micro‑workflows you can try today
Below are 20+ concrete, genealogy‑specific workflows, each tied to one of the releases or changes above. You can mix and match based on which tools you use.
1–4: Deep document analysis with Claude Opus 4.8
High‑effort probate file dissection (Claude Opus 4.8 + effort control)
Upload a long probate packet or guardianship file to Claude.
Set effort to “high” and ask: “Create a detailed chronological timeline of events, identify all named individuals with inferred relationships, and list unresolved identity conflicts.”[anthropic]
Use the output to update your research log and highlight areas needing original jurisdiction checks.
Fast‑mode indexing of multi‑page deeds (Claude Opus 4.8 fast mode)
Batch upload several deed images or text transcriptions.
Use low effort/fast mode to produce a compact table: parties, date, land description, witnesses, and immediate family clues.[anthropic]
Paste that table into your locality study or Excel log.
Conflicting‑evidence brief (Claude Opus 4.8)
Provide a narrative summary plus citations, then ask Claude in high‑effort mode: “Evaluate each conclusion against the evidence and point out weak inferences or missing negative evidence.”[anthropic]
Treat this as an AI “peer review” step before submitting an article or client report.
Multi‑language church‑book sweep (Claude Opus 4.8)
Upload several pages of German or Dutch parish registers transcribed as text.
Ask Claude to scan the whole batch for your target surnames and produce a per‑family entry list, without translating everything.[anthropic]
5–8: Large‑scale, agentic tasks with Claude dynamic workflows
Automated locality‑study builder (Claude dynamic workflows)
Give Claude a list of URLs (county histories, digital collections, catalog entries) and ask it—via dynamic workflows—to:
visit each link, 2) extract record‑set descriptions, and 3) output a structured locality guide (collection name, years, access, coverage).[anthropic]
Research‑plan aggregator from scattered notes (Claude dynamic workflows)
Drop in multiple prior research reports, to‑do lists, and log snippets.
Ask Claude to run a dynamic workflow that: parses each document, merges all open research questions, removes duplicates, and outputs a ranked research plan with repositories and record types.[anthropic]
Surname cluster analysis across multiple counties (Claude dynamic workflows)
Provide CSV exports from several databases (e.g., tax lists, land grants).
Have Claude run a workflow to find clusters by surname, time, and jurisdiction, then output candidate FAN club groupings for a brick‑wall ancestor.[anthropic]
Bulk citation sanity check (Claude dynamic workflows)
Paste a long list of source citations (from a manuscript or RM export).
Let Claude’s workflow flag inconsistent formats, missing elements (e.g., page numbers, series), and potential mismatches between citation and description.[anthropic]
9–11: Gemini 3.x / 3.2 Flash for images & quick tasks
Cemetery photo triage (Gemini 3.2 Flash if available)
In the Gemini app, upload a folder of gravestone photos and ask: “Create a table of names, dates, and inscription snippets; flag stones that likely belong to the same family group.”[youtube]
Form‑style census extraction (Gemini 3.x Flash/Pro)
Map‑based migration summary (Gemini 3.x)
Give Gemini a list of places and dates for one family line and ask for a concise narrative of migration patterns plus a bullet list of likely record sets at each location/time.[ai.google]
12–15: Perplexity Computer as a research assistant
Always‑on locality‑record watcher (Perplexity Computer)
Point Perplexity Computer at a set of bookmarked archives and library catalog pages for a county or tribe.
Instruct it to check weekly for new digitized collections or finding aids and email you a short bulletin of changes.[perplexity]
Cross‑site ancestor dossier builder (Perplexity Computer)
Have Computer open your working Google Doc, Ancestry profile page for the ancestor, and relevant web articles.
Ask it to compile an integrated dossier: brief biography, attached sources with links, and notes about conflicting data; then run an “audit” pass for logical consistency.[perplexity]
Blog‑post fact‑check before publishing (Perplexity independent audit)
Draft an article about a family or locality in Google Docs.
Let Perplexity Computer perform an audit pass to flag factual inconsistencies, repeated claims, or unsupported assertions, without altering your prose.[perplexity]
Session‑to‑session continuity for a case study (Perplexity Computer)
Use Computer as the “memory” for a complex project by keeping your case study docs, notes, and last session open.
Each new day, prompt: “Review our last session’s steps and propose the next three concrete record searches, with URLs where possible.”[perplexity]
16–18: Open‑weight Mistral for privacy‑sensitive work
Local‑only surname extraction (Mistral‑Small 3.2 via Ollama)
Run Mistral‑Small 3.2 locally and feed it OCR’d PDFs of sensitive adoption or recent‑generation records that you don’t want leaving your machine.[simonwillison]
Ask it to output only names, dates, and locations for internal tracking.
Institution‑scale analysis with Mistral Large 2 (Snowflake)
If you or a partner society has research data warehoused in Snowflake, use Mistral Large 2 via Cortex AI to query: “Find all instances where two unrelated families share the same township, occupation, and witnesses; rank by strength of overlap.”[docs.snowflake]
Bulk note‑cleaning for imported trees (Mistral‑Small 3.2)
Send messy GEDCOM note fields into the local model and ask: “Standardize date formats, normalize place names, and split this into structured facts vs speculation.”[simonwillison]
19–22: Responding to model retirements and lineup shifts
ChatGPT workflow refresh after GPT‑5.1 removal
Open your saved prompts, templates, or instructions that referenced “GPT‑5.1 Thinking/Pro.”
Update them to target the currently strongest reasoning model (e.g., GPT‑5.5 Pro or o‑series) and add explicit instructions like: “Use step‑by‑step reasoning but keep your final answer under 800 words, suitable for a research log entry.”[openai]
Claude model selector clean‑up
In Claude, remove any project documentation that still refers to “Opus 4” or “Opus 4.1,” and standardize on Opus 4.8 for deep work and a cheaper model (e.g., Sonnet) for quick‑turn tasks.[support.claude]
Gemini 2.0 → 3.x migration checklist
Grok 4.20 limits reality check
If you’ve been experimenting with Grok for historical text or social‑media‑sourced obituaries, adjust expectations: 4.20 Beta 2 is the current public model and Grok 5 is not yet live.[nxcode]
Frame it as an auxiliary tool for quick web‑aware lookups, not your primary analysis engine.
Twenty‑plus practical AI uses for genealogists (immediately try‑able)
Below are concrete, tool‑agnostic ideas you can do today with a general LLM plus your usual genealogy platforms. Several echo proven patterns discussed in recent AI‑for‑genealogy guidance and workshops.
Research planning, analysis, and evidence work
Draft a targeted research plan from a brick‑wall summaryPaste your current problem statement (e.g., identifying a woman before marriage in Oklahoma Territory) and have the model outline a step‑by‑step plan with record types (probate, land, territorial court, tribal rolls), jurisdictions, and priority order to investigate.
Map conflicting hypotheses for one identityPresent two candidate identities for the same person, summarize your evidence table, and ask the AI to argue for and against each hypothesis and suggest specific new records that would best discriminate between them.
Turn a long deed or probate packet into an evidence summaryPaste a cleaned transcription or abstract of a deed or multi‑document probate file and ask for: key parties, stated relationships, property descriptions, and a chronological timeline you can verify against the original images.
Generate a problem‑focused timeline with gaps highlightedProvide a list of dated events from your notes and ask AI to normalize the dates, sort chronologically, and annotate likely gaps or inconsistencies (e.g., missing censuses, unexplained geographic jumps) for further research.
Ask for alternative search strategies when you “run out” of ideasPaste the section of your research log where you’ve listed negative searches, and have the model propose additional search strategies: variant name spellings, nearby counties, associated families, or underused record types.
Working with documents and languages
Summarize and extract facts from long narrative documentsUse AI to summarize lengthy court cases, guardianship files, or estate settlements, asking it to list every person mentioned with role (heir, neighbor, surety), date, and place in a small table you then double‑check.
Translate short segments from non‑English parish or civil recordsPaste a line or two from German, Latin, Spanish, or Scandinavian church registers and ask for both a plain translation and an extraction of the genealogical facts: names, dates, places, and relationships.
Refine OCR/handwriting outputs from separate toolsAfter running a will book or territorial court docket through a handwriting/OCR tool, paste the raw transcription into an LLM and ask it to normalize capitalization, repair obvious misread names and places, and flag words it is uncertain about for your manual review.
Standardize place names while preserving historical jurisdictionsPaste a column of inconsistent place strings (e.g., “Muskogee I.T.”, “Muskogee, Indian Terr.”, “Muskogee Co., OK”) and have AI split each into fields for standardized naming (modern), historical jurisdiction (e.g., Indian Territory), and notes on boundary changes.
Extract structured data from narrative parish or civil entriesPaste several baptism, marriage, or burial entries and ask the model to output a simple CSV‑style table with columns for name, event type, date, parents/spouse, witnesses, and locality, ready for import into a spreadsheet or research log.
Pattern finding and locality/context work
Scan for naming and migration patterns in a localityProvide a set of transcribed baptisms, marriages, or probate entries from one town or county and ask AI to summarize recurring surnames, common given‑name patterns across decades, and evidence of in‑ or out‑migration.
Draft a locality guide for a county or tribal jurisdictionGive the name of a county, reservation, or district (e.g., Muskogee County or Cherokee Nation) and ask the model to outline a locality guide: main repositories, core record sets and coverage dates, known record losses, and online vs on‑site access; then you fact‑check and expand.
Generate short historical context boxes for reports or blog postsAsk for a 150‑word, source‑checkable overview on topics like Oklahoma land runs, allotment policies, coal‑mining communities, or World War I draft registration in a specific county, to drop into research reports or blog articles.
Create quick glossaries for unfamiliar legal or land termsPaste a handful of phrases from probate files, allotment records, or land patents and ask the model to explain each term succinctly in genealogist‑friendly language so you can better interpret the original records.
DNA‑related uses (conceptual rather than platform‑integrated)
Conceptually cluster DNA matches for research prioritizationExport or copy a simplified match list (shared cM, known relationships, notes) and have AI group matches into likely clusters—e.g., “maternal‑grandmother line,” “paternal Creek line”—and suggest which cluster to target to test a specific hypothesis.
Draft plain‑language explanations of DNA resultsProvide a summary of your match distribution and ethnicity regions and ask the model for a concise explanation suitable for non‑technical relatives, focusing on what the results can and cannot tell them.
Help narrate how DNA supports a documentary conclusionFeed a short written explanation of how several triangulated matches point to a shared ancestor and ask AI to improve clarity, structure, and caveat language while keeping your evidentiary logic intact.
Teaching, blogging, and client‑facing work
Turn a research log into a client‑ready or cousin‑friendly summaryPaste a portion of your research log and ask the model to reorganize it into a short narrative report: objective, methods, sources consulted (with your citations preserved), negative findings, and next steps.
Generate lesson outlines or handouts for genealogy classesProvide your topic (e.g., “Using probate records in territorial Oklahoma” or “Intro to Native American genealogy using the Five Tribes”) and ask for a 45‑minute lesson outline with learning objectives, key examples, and suggested homework exercises; then you adjust details and add citations.
Brainstorm series topics and post titles for a family history blogDescribe your typical audience and focus (e.g., Muskogee‑area research, Five Tribes records, or probate case studies), then have AI list blog‑series concepts and draft working titles and short synopses for each.
Create comparison tables for tools, record sets, or repositoriesAsk the model to help you build structured comparison tables—for example, differences among major AI transcription tools, coverage of Oklahoma probate records across sites, or pros/cons of key DNA vendors—based on information you provide or point it to.
Draft ethical‑use notes and disclaimers for AI‑assisted workUsing guidelines that emphasize “AI assists but never replaces the genealogist,” have the model help you phrase short statements you can include in reports or blog posts documenting how you used AI and how you verified its outputs.
Organization, productivity, and experimental workflows
Normalize and tag messy research notesCopy a set of free‑form notes and ask AI to identify and tag people, places, record types, and tasks, then rewrite the notes into a structured format compatible with your Zotero or research‑log templates.
Generate checklists for specific record setsFor a record type like “Oklahoma Territory probate” or “Dawes enrollment packets,” ask for a checklist of typical documents, key fields to capture, and common pitfalls, to use as a data‑entry guide while you work through image sets.
Summarize webinars or long articles into actionable stepsPaste your own notes from a genealogy webinar or article on AI in research and ask the model to distill them into a short list of concrete actions you can take in your next research session.
No comments:
Post a Comment