Here’s your “what actually changed and how it matters for us this week” briefing based on the last 48–72 hours (plus anything still rolling out but clearly live).
Be sure to check out the workflows and prompts below.
The through‑line this week is reasoning + scale + agents.
GPT‑5.5 Instant, Gemini 3.5 Flash, and Grok 4.3 all push toward faster, cheaper models that can still reason well while reading very large contexts (100k+ tokens). For genealogists, that means you can feed more of a project—dozens of documents, research logs, and notes—into a single conversation and actually expect the model to keep track of it, rather than chopping everything into tiny, isolated prompts.clickforest+1youtubeOn top of that, tooling around files and workflows is maturing. ChatGPT’s Library, large‑paste‑as‑attachment behavior, and multi‑file uploads make it much more realistic to treat ChatGPT as a working “project notebook,” not just a one‑off Q&A box. Deep research and Codex/agents (plus Perplexity Personal Computer and Gemini/Claude agents) mean we’re moving from “ask a question, get an answer” toward “set up a repeatable workflow that runs over your archives, logs, and online collections.”youtubedigitalappliedyoutube
The trade‑offs are changing too: older models are disappearing. GPT‑4o, GPT‑4.1, GPT‑5.1, and legacy deep research are being retired, and some free auto‑switching to “Thinking” is going away. For working genealogists, this nudges you toward either (a) learning to explicitly select “Thinking” / “Pro” / “agents” when you need them, or (b) leaning into mid‑tier “Instant / Flash” models for routine tasks and paying for high‑reasoning modes only when you hit a truly gnarly problem.youtube
Plug‑and‑play AI micro‑workflows for genealogists
1. Multi‑document research log auditor
Uses: GPT‑5.4 Thinking + 256k context + ChatGPT Library improvements.youtube
Upload a full research log plus up to 20 related record images or transcripts into a single ChatGPT conversation.
Prompt: “You have all my files in this chat. Using only these files, create: (1) a table of each research question and sources attached, (2) a list of conflicts or gaps, (3) specific next searches in record‑set language (jurisdiction, date range, record type). Quote each recommendation back to the specific file name and line.”
2. “One surname, one weekend” deep research plan
Uses: Deep research mode in ChatGPT + GPT‑5.4 Thinking’s updated planning.youtube
Start deep research with a narrowly scoped question: “What are the key record types and online collections for researching the MORGAN surname in Cherokee Nation, 1870–1907?”
Require: a research plan, prioritized repositories, and short justifications for each collection, with links where available and clear caveats about coverage.familysearch+1youtube
3. Spreadsheet‑driven FAN club analysis
Uses: GPT‑5.4 Thinking (improved spreadsheet work) + 256k context.youtube
Export your FAN (Friends/Associates/Neighbors) table from Excel/Google Sheets to CSV and upload.
Prompt: “Using this table only, identify clusters of people who appear together across multiple events and time periods, highlight probable kinship clusters, and generate hypotheses (not conclusions) I should test in land, probate, and church records.”
4. “Big stack of obits” extractor
Uses: GPT‑5.5 Instant (fast text extraction) + ChatGPT multi‑file upload.clickforestyoutube
Upload up to 20 obituary PDFs or images.
Prompt: “For each file, extract all people, dates, places, and relationships as a structured table (columns: source file, deceased, birth info, death info, spouse(s), children, siblings, parents, pallbearers, clergy, organizations, locations). Do not infer; transcribe exactly and mark unclear items.”
5. Side‑by‑side record comparison for identity
Uses: GPT‑5.4 mini fallback (Thinking) + large context.youtube
Paste or upload two candidate records for the same name (e.g., two John Morgans in Indian Territory).
Prompt: “Compare these two records strictly as evidence for whether they describe the same individual. List agreements, conflicts, and ‘unknown’ separately, and finish with three alternate hypotheses I should test, not a conclusion.”
6. Context‑heavy biographical sketch from many sources
Uses: GPT‑5.5 Instant + GPT‑5.4 Thinking as needed + ChatGPT Library.youtubeclickforestyoutube
Upload a bundle: census transcripts, land abstracts, probate summaries, a draft narrative, and a timeline.
Prompt: “Using only the uploaded files, draft a 1,500‑word biographical sketch for teaching, with footnote‑style in‑line source markers (FileName: page/line). Flag any inferred statements with ‘[interpretation]’ so I can review.”
7. Fast “where could records exist?” scan
Uses: GPT‑5.5 Instant (default) + improved web search quality.clickforestyoutube
Prompt: “You are assisting a genealogist. For a person who lived in [county, state] between [years], list jurisdiction‑appropriate record types and online portals (state archives, county offices, tribal archives, church bodies), focusing on land, probate, court, and school records. Provide URLs and note any paywalls or onsite‑only constraints.”
8. Gemini 3.5 Flash “mega‑timeline” builder
Uses: Gemini 3.5 Flash 1M context.digitalapplied+1
Paste or upload a very long chronological note (dozens of pages of notes + snippets).
Prompt: “From this entire corpus, build a single, source‑cited timeline for [target ancestor], with separate lines for them, spouse(s), each child, and key community or tribal events. Use only facts that appear explicitly in the text and cite file+line for each.”
9. Rapid multilingual newspaper scan
Uses: Gemini 3.5 Flash or GPT‑5.5 Instant for long, multilingual text.clickforest
Paste a long run of OCR’d newspaper snippets in multiple languages (e.g., English + Spanish).
Prompt: “Identify only items that mention [surname] in connection with [town/county]. Extract exact quotes with date, newspaper title, and page if present; then summarize each item in one sentence.”
10. Grok 4.3 migration route explorer
Uses: Grok 4.3 (improved reasoning and web search).digitalapplied+1
Prompt: “Assume I’m tracing a family that appears in [location A] in 1870 and [location B] by 1885. Based on period transportation routes, land policies, and major events, outline plausible migration paths and associated record sets (rail, river, wagon roads, tribal relocation routes, homestead offices) I should check, with sources.”
11. Perplexity Personal Computer “research dashboard”
Uses: Perplexity Personal Computer agent across local files + web.linkedinyoutube
Point Perplexity’s Mac app at a folder of PDFs (probate, deeds, compiled notes) plus your browser with open tabs to Ancestry/MyHeritage collections.
Ask: “Build a summary dashboard of all occurrences of the surname CLARK in these local files and open browser tabs, grouped by county and date, and output a CSV I can import into Excel.”
12. “What’s missing from this county file?” audit
Uses: GPT‑5.4 Thinking + 256k context.youtube
Upload a county‑level “to‑do” list plus a stack of abstracts you’ve already done.
Prompt: “Review my county checklist against the abstracts provided and identify: (1) record groups not yet searched, (2) time periods with no sources, and (3) recommended next three record groups with justification.”
13. Privacy‑conscious local AI on Mistral Medium 3.5
Uses: Mistral Medium 3.5 open‑weight (self‑hosted).clickforest
Run Mistral locally (or via a privacy‑respecting host) and feed it transcriptions of records involving living or recently living persons.
Prompt: “From this text, create a de‑identified research summary where all living individuals are assigned pseudonyms, but relationships and chronology remain intact.”
14. Claude “sandboxed” archive assistant
Uses: Anthropic self‑hosted sandboxes + Claude Opus/Sonnet.digitalapplied+1
Configure Claude in a sandbox that has access to your on‑premise document store (scanned deeds, tribal enrollment applications, etc.).
Ask: “Within this sandbox, locate all files that reference the Dawes Commission for the [Nation], 1893–1914, and produce a reading list with one‑sentence descriptions and reasons each might matter for the MORGAN family.”
15. “One image, full analysis” for hard‑to‑read deeds
Uses: GPT‑5.5 Instant or Gemini 3.5 Flash (multimodal) + ChatGPT Images tooling.indigenousmexico+2youtube
Upload a high‑resolution scan of a deed or patent in difficult handwriting.
Prompt: “Transcribe this image exactly, including line breaks and punctuation. Then below the transcription, extract a structured summary (grantor, grantee, land description, dates, witnesses, consideration). Flag any uncertain words and offer 2–3 possibilities.”
16. Mass‑translation of foreign‑language church registers
Uses: GPT‑5.5 Instant / Gemini 3.5 Flash long‑context translation.familysearch+2
Upload multiple pages of a baptism or marriage register.
Prompt: “Translate each entry into English line‑by‑line, keeping names and place names in original spelling. Then create a table (entry number, date, child, parents, residence, sponsors, notes). Do not infer relationships beyond what the text states.”
17. AI‑assisted instructional design for a workshop
Uses: GPT‑5.5 Instant + ChatGPT Projects/Library.clickforestyoutube
Upload an outline of an upcoming genealogy AI workshop and some example case studies.
Prompt: “Design a 90‑minute workshop called ‘Using AI for Indian Territory Land & Probate Research’ with learning objectives, segment timings, demo prompts, and one hands‑on exercise using record images (describe the exercise, not the answers).”
18. Automated extraction from Revolutionary/Civil War pensions
Uses: GPT‑5.4 Thinking + large file handling.indigenousmexico+1youtube
Upload a full pension file PDF.
Prompt: “From this single pension file, create: (1) a chronological list of events mentioned, (2) a table of all named people with roles (soldier, widow, child, witness, neighbor), and (3) a list of jurisdictions mentioned. Quote the exact page and snippet for each entry.”
19. Research‑log hygiene via Codex plugins
Uses: Codex app + Codex plugins for reusable workflows.youtube
Install or create a Codex plugin that takes a plain‑text research log and outputs it in your preferred RootsMagic or spreadsheet schema.
Use it whenever you finish a research session: paste your messy notes, run the plugin, and then copy the cleaned output back into your master log.
20. “AI use in my methodology” documentation helper
Uses: GPT‑5.5 Instant / GPT‑5.3 Instant Mini.indigenousmexico+1youtube
Paste an excerpt of your current research methodology or instructions to students.
Prompt: “Revise this document to (a) explicitly describe how I use AI as a research assistant (extracting, organizing, translating, summarizing), (b) include cautions that AI is not a source, and (c) add a short section on documenting AI use in research logs, in line with best practices.”indigenousmexico
21. Teaching “AI reading of a record” safely
Uses: GPT‑5.5 Instant + ChatGPT Images.familysearch+1youtube
Upload a sample record image (e.g., a Freedmen enrollment card, Dawes packet page, or pre‑statehood land record).
Prompt: “Pretend you are co‑teaching a beginner class. Generate: (1) a step‑by‑step outline of how to ask AI to help read this record without over‑interpreting it, and (2) three example prompts that force the AI to quote the text rather than speculate.”indigenousmexico
No comments:
Post a Comment