Using AI to Transcribe Old Cursive Wills, Deeds, and Letters

If you have ever uploaded a 1780 county probate will, an 1840 metes-and-bounds land deed, or a faded Civil War letter into a standard Optical Character Recognition (OCR) program, you already know the result: a wall of garbled punctuation marks. Conversely, if you paste that same manuscript image into a modern multimodal AI assistant with a casual prompt like “Transcribe this old letter for me,” you face a far more dangerous problem: confident paleographic hallucination. Whenever an ink blot, fold crease, or microfilm scratch obscures a given name or date, an unconstrained language model silently guesses a smooth, modern-sounding word and presents it as fact.

Getting archival-grade transcriptions from AI requires scanning the manuscript at sufficient optical resolution (400–600 PPI unfiltered) and constraining the AI with NARA diplomatic transcription rules, historical abbreviation glossaries, and colonial land and currency math checks. You can build customized, copy-ready paleography prompts in Tab C of our Tombstone, Census & Kinship Calculator and plan your document scan settings with our Photo, Slide & Home Movie Digitization Planner.

---

Why 17th–19th Century Cursive Hands Baffle Standard OCR

Traditional OCR engines segment pages into isolated character boxes trained on modern printed fonts. Historical manuscripts violate those assumptions across three distinct eras of American and British penmanship:

Historical Script Era	Primary Date Range	Common Genealogical Records	Key Paleographic Features & Letterform Traps
Secretary Hand	1500 – 1680 (into 1700s in courts)	Early colonial town books, Plymouth/Virginia/Massachusetts deeds, 17th-century wills	Reverse-looped lowercase e (resembles o), two-stroke c and r, descending h and s, double-f capitals (ff = F), and scribal brevigraphs (p with a barred tail for per/par).
Copperplate / English Round Hand	1690 – 1840	Revolutionary War pensions, 1790–1840 censuses, parish registers, 18th-century probate	Shaded downstrokes paired with thin hairline upstrokes that vanish on high-contrast microfilm; pervasive use of the medial long s (ſ).
Spencerian & Palmer Cursive	1840 – 1920	1850–1900 US Federal Censuses, Civil War letters, county deed and mortgage books	Rapid running loops where lowercase m, n, u, i, v, r collapse into identical vertical strokes (minims), making strings like munn or inurn visually identical.

Compounding these scripts is iron-gall ink bleed-through (verso show-through), where acidic oak-gall ink ate through rag paper so backward writing from the reverse side appears between the front lines.

---

The NARA Diplomatic Transcription Standard: Stopping AI Hallucinations

Following the National Archives (NARA) Citizen Archivist transcription guidelines, a diplomatic transcription preserves every original spelling, capital letter, abbreviation, and line break without editorial modernization. Enforce these four rules in every AI paleography prompt:

1. Preserve Line-by-Line Numbering (Line 01:, Line 02:): Require one numbered line per physical line on the scan. This prevents eye-skip (homeoteleuton, where the transcriber jumps between two lines ending in the same word, dropping an entire clause or heir from a will) and lets you audit every word against the image.
2. Verbatim Spelling and Capitalization: Keep phonetic spellings ("recieved", "widdow", "payd", "yeoman", "ffebruary") and archaic capitalization exactly as written without inserting [sic].
3. Strict Uncertainty Brackets ([illegible] and [conjectural?]): Forbid the AI from guessing over torn paper or ink blots. Require [illegible — approx. 5 chars] or ranked paleographic alternatives in brackets, such as [Samuel? / Lemuel?] or [14s? / 18s?].
4. Separate Diplomatic Text from Expanded Reading: Output the verbatim diplomatic transcript first, followed by a separate plain-English reading that expands contractions in brackets (th[e], s[ai]d).

---

Decoding Archaic Abbreviations, Legal Terms & the Long s (ſ)

Manuscript Writing	Literal Expansion / Meaning	Paleographic & Genealogical Explanation
ye, yt, ym, yn	the, that, them, then	Here y substitutes for Old English thorn (þ = th). ye was always pronounced “the”, never “yee”.
ſ (Long s) / ſs / fs	s / ss	Used at the start or middle of words (ſame, Teſtament, Miſsouri, Eſſex), never at the end. Lacks the right-side crossbar of f. Double ſs (Dorcheſter, Preſton) often looks like fs or ß.
ff (Initial Double f)	Capital F	In 17th- and 18th-century Secretary and Round hand, capital F was written as a doubled lowercase ff (ffebruary = February, ffrancis = Francis).
sd / s'd / afsd	said / aforesaid	Refers back to the person, parcel, or sum named earlier ("the sd John Carter").
vizt / viz.	videlicet (“namely / to wit”)	Introduces an itemized list of heirs, boundary trees, or inventory goods (z represents the medieval contraction for -et).
inst. (instant)	of the current calendar month	In a newspaper dated May 20, 1874, “Died on the 12th inst.” means May 12, 1874.
ult. (ultimo)	of the previous calendar month	In that same May 20, 1874 paper, “Died on the 28th ult.” means April 28, 1874.
prox. (proximo)	of the next calendar month	“Auction on the 10th prox.” in a May paper means June 10.
Do. / do. / "	ditto (“the same as above”)	Repeats the surname, birthplace, or unit of measure from the line above.
Relict vs. Consort	Widow vs. Living Spouse	Relict of John Smith means she died a widow after John’s death; Consort means her husband was still living when she died.
et ux. (et uxor)	and wife	Standard in county land deeds ("Samuel Vance et ux.") when a wife joins a conveyance to release her common-law one-third dower right.
Jno., Jas., Jos., Wm., Thos.	John, James, Joseph, William, Thomas	Standard clerks’ given-name contractions (Jno. is John, never Jonathan [Jona.]).

---

Colonial Land Surveying & Probate Currency Math

Because colonial and 19th-century clerks used non-decimal British units and summed every column at the bottom of the page, checking their arithmetic proves which digit is hidden under an ink smudge:

Colonial Land Surveying (Edmund Gunter's 1620 Chain):
1 Gunter's Chain = 4 Rods (Poles / Perches) = 66 feet = 100 Links (1 Link = 7.92 in)
1 Rod / Pole / Perch = 16.5 feet = 25 Links
1 Acre = 10 Square Chains = 4 Roods = 160 Square Perches = 43,560 sq. ft.
1 Rood = 1/4 Acre = 40 Square Perches = 10,890 sq. ft. (1 Sq. Perch = 272.25 sq. ft.)

British & Colonial Probate Currency (£ / s / d):
£1 (Pound)     = 20 Shillings (s.) = 240 Pence (d., from Roman denarius)
1 Shilling (s) = 12 Pence (d.)     | 1 Penny (d) = 4 Farthings (1/4d, 1/2d, 3/4d)

• Worked Deed Check (Acres, Roods, Perches): Adding two inherited tracts of 42 A. 3 R. 28 P. and 19 A. 2 R. 34 P.: first sum the Perches (28 + 34 = 62 Perches = 1 Rood + 22 Perches, carrying 1 R at 40); next sum the Roods (3 + 2 + 1 = 6 Roods = 1 Acre + 2 Roods, carrying 1 A at 4); finally sum the Acres (42 + 19 + 1 = 62 Acres, 2 Roods, 22 Perches, or 62.6375 Acres).
• Worked Probate Check (£ / s / d): Adding three inventory lines of £14 12s 6d, £8 [18s? / 13s?] 8d, and £3 7s 10d against a bottom-of-page clerk total of £26 19s 0d: sum the Pence (6d + 8d + 10d = 24d = 2s 0d, carrying 2s at 12); to reach 19s with a £1 carry (39s total), the smudged shilling entry must be 39 − 12 − 7 − 2 = 18s, proving the ink-blotted figure is 18s rather than 13s.

---

Worked 4-Part AI Paleography Prompts

Before uploading your manuscript image, scan the original page at 400–600 PPI in 16-bit grayscale or 48-bit color without aggressive sharpening (see Best PPI Resolution for Scanning Old Family Photos and Slides). Then paste the prompt below alongside your image:

ROLE & CONTEXT:
You are a senior archival paleographer following the US National Archives (NARA) diplomatic transcription standard. Attached is a scan of a [17th/18th/19th]-century [Will / Land Deed / Census / Newspaper] from [County/State, Year].

STRICT DIPLOMATIC RULES (ANTI-HALLUCINATION):
1. Transcribe line-by-line (Line 01, Line 02, ...) matching the physical manuscript lines.
2. Preserve original spelling, capitalization, punctuation, and archaic contractions (ye, yt, sd, afsd, vizt, inst., ult., Do., Relict, et ux., ff, and medial long s [ſ]). Do NOT silently modernize.
3. NEVER guess over ink blots, tears, or ambiguous minims. Mark unreadable text as [illegible — approx. N chars] and give ranked alternatives in brackets, e.g., [Jno.? / Jas.?] or [18s? / 13s?].

COLONIAL MATH VERIFICATION:
4. Audit all monetary columns (£1 = 20s = 240d) and land parcels (1 Acre = 4 Roods = 160 Perches; 1 Chain = 4 Rods = 66 ft = 100 Links) against page totals.

OUTPUT FORMAT:
- Section 1: Line-Numbered Verbatim Diplomatic Transcript
- Section 2: Plain-English Expanded Reading (expanding contractions in brackets)
- Section 3: Extracted Genealogical Index (Names, Kinship Roles, Dates, Places, Totals, and Flagged Uncertainties)

Worked Example: Decoding an 1848 Newspaper Obituary (ult., inst., Relict)

Suppose a weekly county newspaper printed on Wednesday, March 8, 1848 carries this notice:

“DIED—In Dorcheſter, on the 27th ult., Mrs. Hannah Preſton, Relict of the late Capt. Jno. Preſton, aged 74 yrs, 6 mos, 11 ds. Funeral services were held on the 2d inst.”

Applying our paleography and calendar rules:

• Dorcheſter and Preſton: Medial long s (ſ) before t → Dorchester and Preston.
• Relict of the late Capt. Jno. Preſton: Hannah survived her husband Captain John Preston and died a widow.
• on the 27th ult. (ultimo): The 27th day of the month preceding March 1848 → February 27, 1848 (and because 1848 was a leap year, February had 29 days). Subtracting 74 years, 6 months, 11 days from February 27, 1848 places her birth date on August 16, 1773.
• on the 2d inst. (instant): The 2nd day of the current publication month → March 2, 1848.

Once transcribed, convert pre-1752 dates using Julian to Gregorian Calendar 1752 Shift and Colonial Double Dating and cross-check census ages with US Federal Census Dates (1790–1950) and Birth Year Math.