Here’s “what changed this week” AI briefing tuned for working genealogists, based on releases and announcements visible through June 15–16, 2026.
Nothing in the public record indicates a brand‑new xAI/Grok or Perplexity model shipped in the last 72 hours; current changes are primarily around OpenAI’s lineup, Anthropic’s Opus 4.8 features, and the looming Gemini 3.5 Pro launch.
A. Named releases & features (last ~72 hours)
OpenAI – ChatGPT model lineup changes (GPT‑5.2 retirement, o3‑pro rollout)
As of June 12, GPT‑5.2 models (Instant, Thinking, Pro) are removed from ChatGPT, while o3‑pro appears in the model picker for Pro/Team and o1‑pro enters a 90‑day retirement path.[help.openai]OpenAI – Consolidated flagship GPT‑5 with “Thinking” option
GPT‑5 is now the default flagship in ChatGPT across tiers, with a separate GPT‑5 Thinking variant available via the model picker for paid plans, emphasizing deeper multi‑step reasoning.[help.openai]OpenAI – Upcoming removal of GPT‑4‑class models from ChatGPT
GPT‑4.5 (the last GPT‑4‑class model in ChatGPT) will be retired from the interface on June 27, ending in‑product access to GPT‑4‑series models.[instagram]Anthropic – Claude Opus 4.8 general release + cheaper fast mode
Claude Opus 4.8 is live with improved reasoning, an optional “fast mode” that’s roughly 2.5× faster and significantly cheaper than the prior Opus fast tier.[felloai]Anthropic – Effort control (“how hard Claude thinks”) in claude.ai
New “effort” settings let you trade speed vs depth on a per‑question basis, effectively giving you a quick mode for small tasks and a deep‑thinking mode for complex analysis.[anthropic]Anthropic – Dynamic workflows in Claude Code (research preview)
Claude can now spin up hundreds of parallel sub‑agents to tackle large, multi‑part tasks, plan their work, and verify outputs before returning a synthesized result.[anthropic]Google – Gemini 3.5 Pro final pre‑launch details
Gemini 3.5 Pro, due later this month, is confirmed to be coming with a 2‑million‑token context window and a “Deep Think” reasoning mode for harder, multi‑step problems.[buildfastwithai]Open‑weight – Llama 4, Gemma 4, GLM‑5.x, Qwen 3.6 and friends now “settled” options
This spring’s wave of open‑weight models (Llama 4 variants, Gemma 4, GLM‑5.x, Qwen 3.6‑Plus, etc.) has stabilized and is increasingly available in consumer front‑ends, giving power users million‑token contexts on self‑hosted or low‑cost platforms.[fazm]OpenAI – Open‑weight reasoning models (gpt‑oss‑120b, gpt‑oss‑20b)
OpenAI’s two open‑weight reasoning models are now part of the “standard stack” many platforms are adopting, making stronger local/offline workflows more accessible.[help.openai]
B. Implications for genealogists this week
First, if you rely on specific OpenAI models for research plans, citation help, or narrative writing, you’ll want to update any “use GPT‑5.2” or “use GPT‑4.5” prompts, because GPT‑5.2 is already gone and GPT‑4.5 disappears from ChatGPT on June 27. GPT‑5 (and GPT‑5 Thinking) plus o3‑pro will be your realistic “go‑to” model choices inside ChatGPT going forward.[instagram]
Second, Anthropic’s Claude Opus 4.8 with effort control gives you a clearer knob to turn between “fast checklist” and “slow, careful reasoning,” which is ideal for complex problem analysis like reconciling conflicting evidence, building timelines from long texts, or tackling thorny brick walls. If you subscribe to Claude, this week is a good time to standardize how you label those modes in your research workflow (for example, “Opus‑Quick” vs “Opus‑Deep”).[felloai]
Third, the confirmed 2‑million‑token context for Gemini 3.5 Pro means you should start designing workflows that take advantage of ultra‑long context: think entire county‑level timeline packets, multi‑generation dossier reviews, or full pension files plus a set of maps in one go. In parallel, the maturing ecosystem of open‑weight models (Llama 4, Gemma 4, GLM, Qwen, gpt‑oss) opens paths for private, on‑premise work with large record sets and custom corpora—especially attractive for sensitive family papers or private DNA correspondences.[facebook]
C. Plug‑and‑play AI micro‑workflows (tied to this week’s changes)
Below are “copy‑and‑adapt” workflows you can drop into your week. I’ll call out which new capability each one leans on.
1–5: Adapting to OpenAI’s model lineup changes
Update your personal “standard prompt” set from GPT‑5.2 → GPT‑5 Thinking
Use in ChatGPT: select GPT‑5 Thinking.
Workflow: Paste an old favorite genealogy prompt that referenced GPT‑5.2; ask GPT‑5 Thinking to (a) rewrite it for GPT‑5 Thinking, and (b) optimize it for multi‑step reasoning on a specific task, such as “evaluate a conflicting birth date across three sources.”[help.openai]
Brick‑wall review with o3‑pro’s reasoning focus
Use in ChatGPT Pro/Team: choose o3‑pro in the picker.[help.openai]
Workflow: Paste a concise problem statement plus your summary of evidence, then instruct o3‑pro: “Step through your reasoning in numbered stages; identify gaps, propose 3–5 specific record searches, and flag any logical leaps I may be making.”
Use this when you want more “thinking aloud” than GPT‑5 default, but don’t need a huge narrative output.
Transcription‑first census worksheet generator with GPT‑5
Use GPT‑5 (default in ChatGPT).[help.openai]
Workflow: Paste a line or two of a census you’ve already transcribed. Ask GPT‑5 to create a reusable column set (name, age, inferred birth year, occupation, dwelling number, relationship notes) plus a one‑page summary worksheet template you can reuse for that decade and country.
Pre‑sunset GPT‑4.5 sanity‑check pass on legacy projects
Use GPT‑4.5 until June 27 if it’s been your “trust baseline.”[instagram]
Workflow: Before the model disappears, run 2–3 of your most complex case studies through your existing GPT‑4.5 prompt to capture its analysis style. Then ask GPT‑5 to mimic that style on the same cases and compare outputs, adjusting your prompts accordingly.
Citation language refiner with GPT‑5 Thinking
Use GPT‑5 Thinking for nuanced wording.[help.openai]
Workflow: Paste a set of draft source citations (for example, from RootsMagic or Zotero notes) and ask GPT‑5 Thinking to: “Normalize tense and voice, keep all facts intact, and draft 2–3 alternate narrative wordings that maintain genealogical standards while improving readability.”
6–9: Taking advantage of Claude Opus 4.8 and effort control
Two‑pass research log review (Opus fast vs deep effort)
Tool: Claude Opus 4.8; first run with low effort, second with high effort.[anthropic]
Workflow:
Low effort: paste a month’s research log and ask for a bullet list of obvious next steps.
High effort: using the same log, ask Claude to produce (a) a narrative synthesis of what is known about the target person and (b) a prioritized research plan with reasoning notes tied to each recommendation.
Complex conflicting‑evidence matrix with deep effort
Tool: Claude Opus 4.8, high effort.[felloai]
Workflow: Paste all the conflicting statements you have about a specific event (multiple birth dates, variant surnames, etc.). Instruct Claude: “Construct a table with columns: source, claim, date recorded, reliability comments, conflicts, and provisional conclusion; explain your reasoning after the table.”
Timeline extraction from narrative county histories
Tool: Claude Opus 4.8, medium effort for speed, high if text is messy.[felloai]
Workflow: Paste a section of a county history that mentions your family and ask Claude to build a chronological timeline with entries for events, locations, named associates, and implied migration steps.
Dynamic workflows for multi‑branch problem clustering
Tool: Claude Code dynamic workflows (research preview).[anthropic]
Workflow concept:
Provide a large bundle of notes covering several related families in the same county.
Ask Claude Code to: “Partition these notes into family clusters; for each cluster, generate a summary, a list of unresolved questions, and a mini research plan.”
Dynamic workflows let Claude parallelize the work across “sub‑agents,” which is helpful when you have many small related problems at once.[anthropic]
10–13: Preparing for Gemini 3.5 Pro’s 2M‑token context
Design a “mega‑packet” county study template
Tool: Existing Gemini (Flash/3.0) now; future‑proofed for Gemini 3.5 Pro’s 2M context.[buildfastwithai]
Workflow: Ask current Gemini to help you design a structure for a single document that will later hold: county timeline, maps, tax lists, deed index extracts, local church lists, city directories, and your working hypotheses. When 3.5 Pro is available, you’ll be ready to paste the entire packet for analysis in one shot.
End‑to‑end pension file and map correlation
Tool: Plan now with Gemini; execute at full scale once 3.5 Pro ships.[buildfastwithai]
Workflow:
Create a standard prompt: “Given a complete pension file and a set of historical maps, extract every place name and date, then correlate them on a timeline and flag likely migration paths.”
Once 3.5 Pro is accessible, you can feed full‑length files and multiple map captions at once, rather than chunking them.
Mass newspaper clipping synthesis
Tool: Gemini 3.5 Pro’s large context and Deep Think mode (when available).[buildfastwithai]
Workflow: Batch OCR 50–100 newspaper clippings related to a surname and locality. In one Gemini 3.5 Pro session, ask for:
A deduplicated list of unique events,
A timeline by person, and
A list of research leads (people or places) that appear only once.
Cluster DNA match notes plus documentary evidence
Tool: Gemini 3.5 Pro (planned), leveraging the 2M tokens for long notes plus documents.[buildfastwithai]
Workflow: Combine correspondence and notes from a DNA cluster with key supporting records (census extracts, obits, etc.). Ask Gemini 3.5 Pro to propose 2–3 candidate trees explaining the shared segments and to list specific records that would confirm or refute each hypothesis.
14–17: Using open‑weight models and gpt‑oss for private / specialized tasks
Local, private family‑papers corpus with gpt‑oss‑20b
Tool: A local interface wired to gpt‑oss‑20b.[help.openai]
Workflow: Index scans or text of family letters, diaries, and private documents. Use prompts like: “List all entries mentioning [surname] in [county] before 1900, grouped by inferred household.” This leverages open‑weight reasoning while keeping data fully under your control.
Custom locality expert chatbot using Gemma 4 or Llama 4
Tool: Self‑hosted Gemma 4 or Llama 4 model with your curated locality notes and finding‑aid excerpts.[fazm]
Workflow: Fine‑tune or RAG‑connect the model to your files about a specific county or tribe. Use it as a “virtual locality handbook” to answer questions like “Where would I find probate packets for 1890s X County?” and “What records survive for Y Territory land allotments?”
Bulk index analysis with GLM‑5.x or Qwen 3.6‑Plus
Tool: Open‑weight model with 1M‑token context (for example Qwen 3.6‑Plus).[fazm]
Workflow: Paste large text exports of index entries (census, tax, deed abstracts) and have the model group entries by inferred person, flag likely duplicates, and output a CSV‑style table for import into your research database.
Long‑form writing with local safety/usage constraints
Tool: Any of the open‑weight models (Llama 4, Gemma 4, gpt‑oss‑120b) tuned to your tone.[fazm]
Workflow: Provide a sample of your own writing (for example, a finished family narrative) and ask the model to emulate that style when drafting new narrative summaries from structured notes, ensuring you can work offline or under specific privacy constraints.
18–20: Cross‑tool workflows and weekly habits
Standardize “Quick vs Deep” labels across platforms
Tools: GPT‑5 vs GPT‑5 Thinking, Claude Opus effort levels, future Gemini Deep Think.[buildfastwithai]
Workflow: Create a short legend in your research manual:
“Q” = quick responses (GPT‑5 default, Claude low effort),
“D” = deep analysis (GPT‑5 Thinking, Claude high effort, Gemini Deep Think).
Use these tags in your research log so you can see at a glance which tasks got superficial help and which got full analytical treatment.
Monthly “AI‑assisted audit” of your brick‑wall list
Tools: whichever mix you have (GPT‑5 Thinking, Claude Opus 4.8, Gemini).[anthropic]
Workflow: Once a month, paste your list of biggest open problems and ask the model to:
Rank them by how likely they are to be solvable with currently accessible online records,
Identify which problems would most benefit from deep‑mode reasoning, and
Suggest which could be delegated to a “quick mode” session.
“Next‑action” summaries after each research session
Tools: Any current frontier model (GPT‑5, Claude Opus, or your preferred open‑weight model).[felloai]
Workflow: At the end of a session, paste your raw notes and instruct the model: “Summarize what I accomplished today in 5 bullets, then list 5 concrete next actions with the specific repository, collection, and search terms I should use.” Save that directly into your research log.
Twenty‑plus concrete AI uses for genealogists (you can try these now)
Below are focused, practical patterns you can drop into a current project or lesson plan. All are drawn from real-world genealogical practice and AI‑in‑genealogy guidance.
Research planning, strategy, and logging
Draft a research plan from a problem statement.
Paste a concise research question (e.g., “Identify the parents of John Smith who appears in Pittsburg County, Oklahoma, in the 1910 and 1920 census”) and ask AI to propose prioritized record sets, repositories, and search tactics, then refine with your jurisdictional knowledge.Turn messy daily notes into a structured research log.
Copy scattered notes from a research session (sites checked, call numbers, negative results) and have AI normalize them into a source‑by‑source log with fields such as date, repository/site, collection, search terms, and outcome, ready to paste into a spreadsheet or research journal.Transform a long finding aid into an actionable to‑do list.
Paste a multi‑page archive finding aid or digital collection description and ask AI to identify specific boxes, series, or microfilm rolls most relevant to your research question, along with short rationales.Brainstorm “next‑step” records when you feel stuck.
After you’ve exhausted census, vitals, and obvious databases, provide a short summary of what you’ve done and ask AI to suggest lesser‑used record types—tax lists, local court dockets, poorhouse records, school censuses, occupational licenses, or tribal rolls—then you decide which are realistic.Document AI’s role in your research log.
Treat AI as a tool you must document: record the date, tool, prompt, and output summary, plus what you verified or rejected, so anyone reviewing your work later can see exactly how AI contributed.
Document reading, extraction, and evidence comparison
Summarize long probate, land, or court packets.
After you’ve transcribed or obtained OCR text, ask AI for a structured summary listing parties, explicit relationships, key dates, property descriptions, witnesses, and a timeline of events, with quotations that support each assertion.Extract names, dates, places, and relationships from transcriptions.
Give AI a transcription and request a table of all proper names, dates, places, and explicitly stated relationships, with the exact wording quoted beside each entry so you can verify in the image.Compare conflicting evidence in narrative form.
Provide AI with two or more conflicting abstracts (e.g., different ages or birthplaces in separate censuses) and ask it to write a neutral comparison that clearly lays out each conflict, evidence by evidence, without reaching genealogical conclusions.Help interpret AI‑indexed handwriting errors.
When a site’s AI indexing seems off, you can feed the index text plus your own transcription attempt and ask AI to list likely mis‑read letters or surnames to check manually, especially valuable with heavily stylized script.Assist with metes‑and‑bounds land descriptions.
Paste several deeds and ask AI to normalize recurring landmarks, neighbors, and waterways, and produce a checklist of calls you need to plot in DeedMapper or similar software, plus a prose summary of the tract’s context.Translate foreign‑language records with genealogical focus.
For records in languages such as German, Spanish, Italian, or Latin, ask for a literal translation that preserves personal names and place‑names as written, and request a short glossary of key genealogical terms appearing in the document.Summarize church or religious‑record entries.
Use AI to summarize transcribed entries from baptismal, marriage, burial, or congregation registers, extracting names, dates, places, sponsors, and any notes, while you retain control over interpretation and correlation.
Working with AI‑indexed and AI‑enriched collections
Exploit AI handwriting recognition in major platforms.
Many platforms, especially FamilySearch, increasingly rely on AI handwriting recognition and entity extraction to make images searchable. You can then download images or transcriptions and use a separate AI tool to structure the data into person/event tables for import into your genealogy software.Use AI to sanity‑check AI‑generated indexes.
When a platform’s AI indexing produces questionable names or places, use another AI to propose alternate readings, then manually compare against the image; this is particularly useful for clustered mis‑reads in the same register or township.Generate migration timelines from multiple record hits.
Provide a list or table of dates and locales from censuses, land records, city directories, and tribal rolls, then ask AI to produce a concise migration narrative and an ordered list of locations you can later map visually.
Organization, synthesis, and writing
Turn raw notes into a draft research summary or proof argument.
Paste structured notes (research question, sources consulted, key findings, conflicting evidence) and ask AI to produce a first‑draft narrative, which you then edit heavily, adding citations and your own reasoning.Generate locality or jurisdiction guides.
Ask AI for a high‑level overview of a county, parish, or tribal jurisdiction—civil boundaries, key record types, timeframes, and known repositories—then verify and expand with your own citations before using it as a handout or blog post.Convert workflows into step‑by‑step tutorials.
Describe a process you already use (for example, systematically searching a specific newspaper database or tribal enrollment records), and have AI reorganize it into a numbered, student‑friendly tutorial or checklist for classes or blog readers.Draft blog posts from completed cases.
Feed AI a concise case file: problem, sources, major findings, resolution, and limitations; request a 800–1,200‑word blog‑style narrative emphasizing story and process, then revise for voice and add source citations and images.Create multiple audience versions of the same content.
Take a research write‑up and ask AI to produce: a short version for a family newsletter, a more technical version emphasizing methodology, and a slide outline for a society presentation, ensuring you still handle all citations and factual checks.Organize scattered research notes into thematic files.
Paste a long “everything in one place” document and ask AI to sort notes into sections such as census, land, court, DNA, and locality background, making it easier to move that content into your evidence file or writing draft.
Teaching, presentations, and outreach
Design course or workshop outlines.
Give AI a session title (e.g., “Using AI Responsibly in Genealogy Research”) plus time length and audience level, and ask for objectives, an outline, and suggested in‑class exercises or case studies; then refine and insert your own examples and citations.Generate handout drafts from slide decks.
Paste your slide text and speaker notes, and have AI produce a concise, two‑page handout with key points, terminology, and space for participant notes, which you can then format and annotate with real citations.Create FAQ documents for society websites or classes.
Based on recurring student questions (e.g., about AI ethics, data privacy, or appropriate use in proofs), ask AI to propose a short Q&A handout, then adjust language and examples to align with your teaching style.Brainstorm genealogical blog series ideas.
Provide your research interests (for example, Five Tribes research, Oklahoma Territory records, or probate case studies) and ask AI for a list of 20–30 blog series or post ideas, grouped by theme, then select a few to outline manually.Draft short explanations of complex methods.
Use AI to create first‑draft explanations of topics like cluster research/FAN club, negative evidence, or record linkage, then refine so they match your preferred terminology and standard genealogical scholarship.
Workflow, tools, and responsible practice
Use AI to normalize your Zotero or citation notes.
Export a small batch of messy notes and ask AI to standardize fields like author, title, repository, and URL, and to propose consistent description phrasing you can then paste back into Zotero after verification.Set up an “AI‑assisted research day” template.
Have AI help you design a repeatable template for a focused research session: pre‑work questions, tasks for the first hour, check‑ins, and a post‑session summary section, including explicit prompts to log AI usage.Develop a personal AI usage policy for your genealogy practice.
Drawing on existing guidance that stresses AI as assistant rather than source, ask AI to help you draft a one‑page policy covering what you will and will not use AI for, how you’ll cite or log AI involvement, and how you’ll protect sensitive data.Compile a reading list on AI in genealogy.
Request a curated list of blog posts, webinars, and courses focused on AI for genealogists, especially from trusted organizations and experienced researchers, to share with your society or include in your newsletter
No comments:
Post a Comment