How to Search a Grantor-Grantee Index

A grantor grantee index is the name index that county land records are searched through, and searching it well is half method and half reading. This article sets out the two-direction search as practiced in title work, how to build a name set that survives clerk spellings, and what to do when the index page itself is a dense hand you cannot resolve.

A grantor/grantee index is a county's official name index to recorded land instruments: the grantor index lists the parties conveying out, the grantee index the parties taking an interest. The standard search runs in two directions — backward through the grantee index from the present owner to a base or source deed, then forward through the grantor index across each owner's window of ownership, catching mortgages, liens, easements, agreements, and partial conveyances the backward deed sequence never shows. The index is a finding aid, not the record. Every hit must be resolved to the instrument itself and its legal description. Where searches actually fail is rarely the method. It is the reading — clerk hands, ditto marks, bare initials, and phonetic spellings that no exact-name query will ever reach.

What the index is, and what it is not

The distinction is directional, not qualitative. The grantor index is organized by the party conveying out — the seller or conveyor. The grantee index is organized by the party receiving. Georgia's title training material puts it plainly: grantors are the sellers, grantees the buyers. Local terminology varies. Orleans Parish's conveyance indexes, running back to 1827, are divided into "Vendor (seller)" and "Purchaser" rather than grantor and grantee, and you will meet "direct" and "inverted" or "reverse" for the same two directions.

Cornell's Legal Information Institute notes that the grantor-grantee index is used by most U.S. counties to record transfers of ownership, that it was historically kept in physical volumes grouped by year and name, and that digital versions are now organized in a variety of ways. That last clause carries more weight than it appears to. "Organized in a variety of ways" is why a search protocol that works in one county produces a false negative in the next.

A tract, geographic, or plat index is a different instrument entirely: it is organized around the land, not the parties. Oakland County, Michigan describes its tract index unit as a service for following ownership history of acreage under metes-and-bounds descriptions, subdivision lots, and condominium units. The two indexes are not interchangeable. Illinois guidance from Attorneys' Title Guaranty Fund treats the grantor-grantee index as the official index providing record notice; an accurate tract index may be used, but an unreliable one must be supplemented by the name index. The same guidance makes the point every examiner eventually learns the hard way: the legal description, not the PIN, controls what a document conveyed, so searching by parcel number alone can lead you to ignore documents that affect title.

Behind all of this sits the applicable recording act, which determines priority among competing claims and falls into race, notice, and race-notice types depending on the jurisdiction's text and exceptions. Recording date and notice, in other words, are substantive facts. That is one more reason the index entry alone will not do: the dates you rely on have to come off the instrument and the recording data, verified.

An abstract of title is a summary of recorded facts used as evidence of ownership; a title or abstract plant is a geographically filed assemblage of title information built to expedite examination. Neither makes the underlying index complete. What follows deals with the reading problem inside that structure — the same problem that runs through all deed and land-records transcription work.

The two-direction search

The procedure documented in the Georgia and North Carolina practice materials is not a national mandate, but it is the method most examiners recognize.

Backward, in the grantee index

Start with the present owner and run the name back through each index period until you find the conveyance into that owner. Take the name of the grantor in that deed and repeat. Georgia's title standards material describes continuing until you reach a deed at least fifty years old — the source or base deed. North Carolina's practice guide describes the same opening move: begin with the current owner in the grantee index to locate the first deed in the chain.

Forward, in the grantor index

Switch indexes at the base deed and run each owner's name forward from the date of acquisition to the date of conveyance out — the date of the conveyance, not the date of recording. This pass is where the adverse instruments live: mortgages, liens, easements, restrictions, agreements, partial conveyances. The North Carolina guide's assessment is worth keeping in front of you: there is no substitute for grinding through these names to verify the property was not mortgaged, conveyed away, or otherwise encumbered. Where a platted subdivision is involved, the guide also has you run the subdivision name itself through the grantor index.

Search period is local

Illinois guidance sets twenty years for residential platted land and back to 1900 for other Illinois property. Georgia's material uses the fifty-year base-deed step. North Carolina's guide observes that the scope of a "full" search is somewhat subjective and generally ranges from thirty to sixty years — and that a stated period is a minimum, often exceeded in practice. There is no uniform national rule. Check the statute, the county's indexing start dates, whether a Torrens or registration system applies, and your underwriter's standards.

Build the name set before you open the volume

The North Carolina guide warns that a minor keystroke error can cause a missed conveyance and, in the worst case, a total title failure, and insists the search terms be broad enough to capture common variations and nicknames. Treat query construction as its own step.

FamilySearch's reference on name variations in indexes and records separates two error sources that behave very differently: the indexer misread or mistyped the original, or the record's creator misspelled the name in the original. The first is recoverable by reading the image. The second is not — the wrong form is the record, and only searching that wrong form, or working from parcel and adjacent-entry clues, will surface it.

Variant classes worth generating for every principal name:

Spelling and phonetic variants. German surnames in particular were recorded inconsistently through the eighteenth and nineteenth centuries, and many immigrant names were anglicized after arrival.
Initials and abbreviations. Index clerks compressed given names freely; a search for "William H. Carter" will not find "Wm H C."
Misread letters. FamilySearch's commonly-misread-letters approach applies directly to deed indexes: CARTER may sit under GARTER, EARTER, CASTER.
Alternate surname structures. Spanish naming customs may combine a given name with paternal and maternal surnames joined by particles — de, y, e. An English single-surname assumption files the same person in the wrong place, or fails to find him at all.
Married and maiden forms, and entities recorded variously as a firm, a trustee, or an individual.

On digital indexes, wildcards do real work: FamilySearch supports an asterisk for multiple characters and a question mark for one, usually requiring at least three letters of the name. Soundex is useful for candidate generation and nothing more — NARA's description makes its limits visible: the code retains the first letter and reduces the rest to three consonant-group digits, so it collides freely and fails outright when the initial letter is the thing in doubt. Daitch-Mokotoff, Metaphone, NYSIIS, and edit-distance measures are reasonable complements, but the general name-matching literature finds no single best algorithm across data types, and none has been validated specifically on deed indexes. Use them to widen the candidate list, never to certify completeness.

One caveat, stated honestly: no national or county-comparable dataset exists on how often these variant classes actually cause missed conveyances. Treat the above as method, not statistics.

When the handwriting fights back

At some point the digital index runs out — the coverage dates stop, the entry looks wrong, or the name you need is not there — and you are reading a photographed index page in a clerk's hand. Three questions, in order.

Is it a capture problem?

Before blaming the hand, check whether the image can support a reading at all: resolution, focus across the gutter, glare on a tight binding, show-through from the verso. Faded iron-gall ink, foxing, and bleed-through are distinct problems with distinct fixes, and no amount of squinting substitutes for a better photograph.

Is it a layout problem?

An index page is a grid, not prose. The risk is not misreading a character — it is binding a correctly read name to the wrong line. Read column by column and line by line, keeping grantor, grantee, instrument type, execution date, recording date, and book/page attached to the same entry. Ditto marks and carried-forward names are the classic trap: a column of dittos under a name means each line inherits it, and a transcription that silently drops or duplicates one shifts every field below it. This is the same structural hazard that governs ledger and tabular material generally, where cell assignment matters more than character accuracy.

Is it a hand problem?

Then work as a paleographer for ten minutes. Build a letterform key from names you already know from the chain — a grantor you have confirmed from a deed body gives you that clerk's capital letters, his terminal flourishes, his ampersand, his minim strokes. Approaching an unfamiliar hand systematically is faster than guessing at the disputed entry directly. And when the name still refuses to resolve, use FamilySearch's fourth step: look for relatives, co-grantors, spouses, and adjacent entries, then open the instrument to see whether your party appears inside a record the indexer filed under someone else. Transcribed discovery layers help here too — FamilySearch's full-text search over unindexed images can surface a correctly transcribed entry a human skipped — but it cannot find what the clerk himself wrote wrong.

Where machine transcription fits, and where it doesn't

Two different jobs hide under "transcribing deed records," and conflating them produces disappointment.

The first is the index volume itself: ruled, pre-printed columns filled with dense handwriting. This is the hardest layout in the building, and it is worth saying plainly that it is also Leo's weakest case — on pages where printed structure dominates, a model can favor the printed headers over the manuscript entries. Verify index transcriptions line by line against the image, always.

The second job is the instruments the index sends you to: deed bodies, mortgages, releases, assignments, metes-and-bounds descriptions — running text in a nineteenth-century clerk's hand. That is what specialist handwritten text recognition is built for, and it is where the volume of reading actually sits. Leo's ATR-1 is a zero-shot model: no per-office training on your county's hands before you can read a book. It reads Latin-script material whatever the language on the page, which matters if your chain crosses French or Spanish notarial records or a Mexican land grant. Transcription is not translation; translating a transcribed instrument is a separate one-click Transformation that writes to its own tab, leaving the base text untouched.

The property that matters most for title work is source integrity. ATR-1 is trained to transcribe what is on the page rather than normalize it, so an archaic or misspelled name survives as written instead of being quietly corrected into the modern form. That is the failure mode that makes fluent AI output dangerous in a chain of title — a plausible name is harder to catch than a garbled one, and the misspelling is often the only searchable form of the party. Leo also checks its own output for failure patterns, hides and retries suspect results, and refunds the credit if a page cannot be read; the errors that do get through are the recoverable kind, checked against the image shown beside the text. Around that, transcriptions of a record series become fuzzy-searchable across documents with per-document metadata fields — Archive, Collection, Book, Identifier — and export to Word, PDF, HTML, or TEI for the file.

For contrast, general OCR is worth one sentence: ABBYY's own documentation states its handprint recognition works on non-cursive text written character by character and shows a cursive sample as not working — which is most of what a nineteenth-century deed book contains. Whatever tool you choose, test it on your own volumes before it touches production work; published benchmarks on other collections do not transfer to your county's hands.

Keep a trail you can stand behind

Speed in this work comes from narrowing, and defensibility comes from documentation. For every link in the chain, record the index searched and its direction, the volume and page, the entry as written, the book and page it points to, the instrument type, both dates, and the name form as it appears in the instrument versus the index. Where those two forms differ, note the difference rather than reconciling it silently — that note is what lets a reviewer, or you in two years, see exactly what was read and why. Verifying a transcription against the image is not a formality; it is the step that converts a hit list into an opinion.

The index is a clerk's claim about the record. The instrument, with its legal description, is the record. Every examiner who has traced a chain through a book of dittoed names in a faded hand knows that the skill being exercised is not searching. It is reading, and knowing when your reading is not yet good enough.

Frequently Asked Questions

How do you search a grantor grantee index?

Search it in two directions. Start in the grantee index with the present owner and work backward, finding the conveyance into that owner, then repeating with each prior grantor's name until you reach a base or source deed. Then switch to the grantor index and run each owner's name forward, from the date of acquisition to the date of conveyance out, to catch mortgages, liens, easements, restrictions, agreements, and partial conveyances the backward pass never shows. Every hit is then resolved to the instrument itself and its legal description. The index is a finding aid, not the record.

What is the difference between a grantor index and a grantee index?

The difference is directional, not qualitative. The grantor index is organized by the party conveying out — the seller or conveyor. The grantee index is organized by the party receiving the interest. Terminology varies locally: Orleans Parish divides its conveyance indexes into "Vendor (seller)" and "Purchaser," and you will also meet "direct" and "inverted" or "reverse" for the same two directions. Both are name indexes, which distinguishes them from a tract, geographic, or plat index, which is organized around the land rather than the parties.

How far back should a title search go?

There is no uniform national rule; the period is set locally. Illinois guidance sets twenty years for residential platted land and back to 1900 for other Illinois property. Georgia's title standards material describes running the grantee index back until you reach a deed at least fifty years old — the source or base deed. North Carolina's practice guide observes that the scope of a "full" search is somewhat subjective and generally ranges from thirty to sixty years, and that a stated period is a minimum, often exceeded in practice. Check the statute, the county's indexing start dates, whether a Torrens or registration system applies, and your underwriter's standards.

Why can't I find a name in the grantor-grantee index?

Usually because the form you are searching is not the form that was recorded. Two error sources behave differently: the indexer misread or mistyped the original, which reading the image can recover, or the record's creator misspelled the name in the original, in which case the wrong form is the record and only searching that wrong form will surface it. Generate spelling and phonetic variants, initials and abbreviations, commonly misread letters, alternate surname structures such as Spanish paternal-and-maternal forms, and married and maiden names. Then look for relatives, co-grantors, spouses, and adjacent entries.

Can AI transcribe handwritten deed index books?

It can help, but index volumes are the hardest case. Ruled, pre-printed columns filled with dense handwriting are Leo's weakest layout: where printed structure dominates, a model can favor the printed headers over the manuscript entries. Verify index transcriptions line by line against the image. Machine transcription earns its keep on the instruments the index sends you to — deed bodies, mortgages, releases, assignments, metes-and-bounds descriptions — where running text in a clerk's hand is what ATR-1 is built for. It is zero-shot, so no per-office training is needed, and it reads Latin-script material whatever the language on the page.