How to Search for Responsive Records Across Agency Systems
Identify the custodians likely to hold material, then searcheach system they use: mail, document libraries, collaboration tools, casesystems, and older storage. Scanned material needs optical characterrecognition to be searchable at all, and email threads are best handled assingle documents rather than dozens.
The order in that sentence matters more than it looks. Mostsearches that miss something were not short on effort. They were run in thewrong direction, starting from the systems the records office can reach andstopping there, rather than from the people who did the work.
The instinct is to start with the tools you have access to.Search the mail system, then the document library, then the case system, and callit done. The problem with that order is that it can only ever reach the systemsyou already remembered. Nothing in the method surfaces the ones you did not.
Starting from custodians inverts it. Ask who was involved inthe thing being requested, then ask what each of those people actually used.That reaches the licensing database nobody thinks of as a records system andthe shared folder a department built around one project in 2021.
Building the custodian list is usually quicker than itsounds. The request itself often names a program, a contract, or a decision,and two or three questions to the owning department will name the people. Workoutward from there: who ran it, who approved it, who was copied, and who hassince left. Departed employees are the most commonly missed custodians, becausetheir accounts are handled by a different team on a different schedule andnobody thinks to ask.
Every agency has an official answer to where records liveand a real one. The official answer is the document management system. The realanswer is longer.
Mail, including calendar entries and the attachments peoplenever filed anywhere else. Document libraries and the shared drives that grewup beside them. Collaboration and chat tools, where a lot of the discussionthat used to be email now happens. Case management, permitting, licensing, andcomplaint intake, none of which anyone calls a records system. Whatever financeuses. HR platforms. The ticket queue IT runs. Physical files. And materialinherited from a system that was decommissioned but never emptied.
Writing that list down once, with a named custodian againsteach entry, is the highest-value hour a records office can spend, and most havenever spent it. It converts the hardest recurring question, which is where tolook, into something closer to a checklist. It also makes exclusions visible: asystem deliberately left out of scope is a defensible decision, while a systemnobody remembered is not, and months later the only difference between them iswhether somebody wrote it down.
A single keyword lifted from the request is almost neverenough. Members of the public describe what they want in ordinary language.Agencies use acronyms, project code names, contract numbers, and internalshorthand that nobody outside the building would guess.
A request about a shelter contract may need the contractor'slegal name, the name everyone actually calls them, the contract number, thedepartment's abbreviation for the project, the name of the program it satunder, and two or three misspellings that appear in real correspondence.Assembling that cluster takes about ten minutes and is usually the differencebetween a search that finds the material and one that returns a handful ofobvious documents and stops.
Date ranges deserve the same care. The period a requesternames is the period they care about, not necessarily the period the recordswere created in. A decision made in March is often discussed the previousNovember.
Scanned binders, photographs of a whiteboard, PDFs producedby a scanner with no text layer, audio, and video all return nothing from akeyword search.
A search that cannotread a document reports the same result as a search across a document that wasnever there. The empty result gives you no way to tell the two apart.
Optical character recognition solves this for most scannedand image-based documents, and it needs to run before the search, not after.Where it cannot be applied, and audio and video are the usual cases, thematerial has to be listed and reviewed by hand rather than passed over insilence. Listing it is the important part. A file you reviewed manually and afile you never opened look identical in the case record unless one of them waswritten down.
An email chain is one conversation. Treated as fortyseparate messages it produces forty review decisions, most of them repeats, anda release package that is unpleasant to read. Handling threads as singledocuments cuts review time noticeably and produces a better release.
The same applies to near-duplicates. The identicalattachment forwarded six times is one document that appeared six times, and itshould be reviewed once and released once, with the appearances noted.
In most jurisdictions the legal standard is whether thesearch was reasonably calculated to uncover responsive records. That is astandard about method rather than outcome, which means missing a document doesnot automatically make a search inadequate. It also means the description hasto exist: the systems queried, the custodians covered, the terms run, and theperiod searched.
That description is easy to write while you are searchingand unpleasant to reconstruct a year later from memory. It is also the thing anappeal actually asks for, so treating it as part of the search rather than aspaperwork afterward is the whole trick.
Govflo treats thesearch itself as a record rather than as a step that produces one. Each searchwrites down its date, the terms used, the sources queried, how many documentscame back, how many were responsive, and what was ultimately withheld orredacted. That is the same set of facts a statutory response letter has tostate, so the letter is populated from the search rather than written frommemory afterward.
On the collection side it connects to the systems agenciesalready run, including Microsoft 365, SharePoint, and Purview for mail search,so material reaches the case without anyone exporting and re-uploading it.Scale is worth naming because it is usually the objection: single documentsrunning to five thousand pages are handled, as are uploads into the hundreds ofgigabytes for video.
What it does not do is decide responsiveness for you.Whether a document falls inside the scope of a request is a judgment thatbelongs to the agency, and it should stay there.
