Searchable PDF Scanning Service: What to Put in the Order
By Schema LawPublished Updated
When you order from a searchable PDF scanning service, put five things in writing: where each document starts and stops, how files are named, what "searchable" has to mean on a test you will run, what happens to pages the scanner cannot handle, and the terms for redoing a batch that fails. "Searchable PDFs, please" covers none of them.
A vendor can meet a bare request to the letter and still hand back something you cannot use: one enormous file per box, named by a counter, with text behind the typed pages and nothing behind the faxes. The order is where you prevent that. Everything after delivery is repair.
In the sample you open, search a pleading title, a docket number, and a witness name you can read on the sheet. Search for subpoena when the box contains one. Match the hit to the page image, and include a page near the end of the file. An affidavit line missing from a clean typed page is enough to send that file back.
What should you ask a scanning service for besides searchable PDFs?
Ask for document boundaries, a file naming rule, a stated treatment for every awkward kind of page, an index of what was scanned, a definition of searchable that you can test, and written rework terms. A bare request for searchable PDFs leaves each of those to the vendor's default, which you will learn only when the delivery arrives.
The gaps, and what each one costs you if it stays open:
| Left unsaid | What can come back | What to ask for |
|---|---|---|
| Where documents begin and end | One PDF per box or per folder, hundreds of pages long | One PDF per document, with the rule for deciding what a document is |
| File names | SCAN_000417.pdf | A naming pattern built from box, folder and sequence |
| What searchable means | Text on clean pages, none on hard ones, and no way to tell which | A test: named words found on sampled pages, including late and poor ones |
| Double-sided pages | Backs skipped, or blank backs kept for every sheet | Both sides scanned; blank backs dropped under a stated rule |
| Odd items | Sticky notes removed and lost, photos skipped, oversize plans folded and cut off | A written treatment for each kind |
| Page order and orientation | Sideways and upside-down pages | Pages upright and in the order found |
| Pages the scanner cannot take | Silent gaps | An exceptions log listing each item not scanned and why |
| A record of the job | Nothing | A spreadsheet index: file name, box, folder, page count |
| Failure terms | An argument | A rework clause with a time limit |
| Custody of the paper | Uncertainty about where the originals are | Chain of custody, and return or destruction only on your written instruction |
Before the boxes leave
A scanning vendor handles client files, so the usual questions about an outside provider apply: who has access, where the images are held, how long, and what is deleted and when. The post on whether it is safe to upload client documents to an OCR tool lists the questions to put to any vendor that touches client material. Whether your professional obligations require more is for the lawyer responsible for the files to decide.
Ask the vendor one more thing: whether recognition is done by the scanning equipment as it scans or as a separate step afterwards, and with what software. You are not choosing their tools. You want to know whether a page can be recognized again without being rescanned, because that decides how painful a failure will be.
A searchable PDF specification to hand a scanning vendor
A specification is one page the vendor signs off before work begins. It states the output in terms you can check, so that acceptance is a matter of comparing the delivery with the page, not of opinion. Fill in the blanks, delete what does not apply, and attach it to the order.
SCANNING SPECIFICATION
Job: ______________________ Boxes: ____ Date: __________
1. OUTPUT
Format: PDF with a text layer on every page that bears
printed or typed text. Page images are not to be replaced
with synthetic type.
Resolution: ____ dpi. Color mode: ____________________.
Image clean-up permitted: straighten / none / ___________.
2. UNITIZATION
One PDF per document. A document ends at:
[ ] a staple, clip or binder tab
[ ] a separator sheet we insert
[ ] other: ___________________________________________
Attachments stay with their parent: yes / no.
3. FILE NAMES AND FOLDERS
Pattern: [box]_[folder]_[sequence].pdf
Example: B014_F03_0007.pdf
Folder tree on delivery mirrors boxes and folders.
4. PAGES
Both sides scanned. Blank backs: drop / keep.
Orientation: upright for reading.
Order: as found in the box.
Sticky notes: scan page with note, then page without.
Oversize, bound, fragile or photographic items: ________.
5. RECOGNITION
Language(s): ____________.
Handwritten pages: scan; no recognition expected.
6. DELIVERABLES
a. The PDFs.
b. Index (CSV or Excel): file name, box, folder,
page count.
c. Exceptions log: every item not scanned or scanned
with a problem, with box, folder and reason.
7. ACCEPTANCE
We will test a sample of files within ____ business days
of each delivery. Tests: page count against index;
search for words visible on sampled pages; visual check
of image legibility. Paper is not returned or destroyed
until we accept in writing.
8. REWORK
Batches that fail acceptance are corrected at no charge
within ____ business days of notice.
Agreed: ____________________ (vendor) __________ (date)Notes on filling it in
Resolution and color. These decide how well recognition works and how large the files are. The guide to scanner settings that make OCR work explains the choices, so pick your values from there and write them into section 1.
Synthetic type. Some recognition software can replace the scanned letters with clean computer type to shrink the file. For legal files you want the picture of the page as it is, stamps and handwriting included. Say so.
Clean-up. Straightening a crooked page is harmless for reading. Speck removal and background whitening can also remove faint pencil, light stamps and highlighter. Decide per job and write it down.
Handwriting. Be honest in the order about what you expect. Asking for recognition of handwritten pages invites a text layer full of guesses that passes a quick test and misleads search later.
One file per document: telling the service where documents begin and end
A scanning operator sees paper, not documents. Unless you mark or define the breaks, the service will choose its own, and you have no say in what it picks. Decide the rule yourself, make it physical where you can, and put it in the order.
Three ways to mark the breaks
- Physical rule. "Each stapled or clipped set is one document; each loose sheet is its own document." This costs you nothing in preparation and is only as good as the stapling in the box.
- Separator sheets. You insert a printed sheet before each document. The operator, or the scanner software reading a barcode on the sheet, starts a new file there. This is the most exact method and the most work for you. Ask whether separator sheets appear in the output or are dropped, and whether they are billed as pages.
- A box inventory. You list each document with its box, folder and position, and the vendor breaks to match. This suits a small number of high-value boxes.
The decisions hiding inside "a document"
- Attachments. Does a letter with three enclosures become one file or four? If four, how will you know they belong together? A shared prefix in the file name answers that, and the post on document unitization covers parents and attachments in a scanned set.
- Envelopes and cover sheets. Scanned with the document, scanned separately, or skipped.
- Folders within folders. Whether a file folder's label is captured, as a scanned image of the tab or as part of the file name.
- Duplicates. Scanned as found. A scanning operator should not be deciding which copies matter.
- Bound volumes. Whether binding may be cut, and if not, how the volume is scanned and at what cost.
A worked example
A small firm sends 22 boxes of closed estate files to a vendor. The order says "searchable PDF, one file per document". Nobody defines a document. The delivery holds 22 PDFs, one per box, each over 2,000 pages. The vendor's position is that a box is a unit.
On the second attempt, with a different matter's boxes, the firm's paralegal, Inés, spends two days inserting barcode separator sheets and writes the staple rule in as a fallback. She specifies B[box]_F[folder]_[sequence].pdf and an index. The delivery holds 6,140 files. The index total matches the vendor's page count. A search for the testator's surname from the top folder finds the will, the codicils and the correspondence, each as its own file. The two days of preparation bought the result, and the order made it enforceable.
How do you acceptance-test a scanning service's searchable PDFs?
Test each delivery before you accept it: reconcile the counts, then open a sample of files and check each one for completeness, legibility and search. Do it while the vendor still holds the paper, because a rescan needs the paper.
Step one: reconcile without opening anything
- Count the files delivered and compare with the index.
- Add up the page counts in the index and compare with the invoice.
- Sort the index by page count. Look at the largest files, which may be several documents run together, and at any one-page files in a box of long documents, which may be a document split in error.
- Read the exceptions log. Every item on it is a decision you now have to make.
- Look for boxes or folders with no files at all.
Step two: sample and open
How many files to open depends on the size of the set and how much rests on it. The guide on how many pages to test before trusting a searchable set covers sample size. Choose the sample to include the hard material on purpose: faxes, carbon copies, small print, colored paper, double-sided originals.
For each sampled file, record the result of these checks:
| Check | How | Pass |
|---|---|---|
| Page count | Compare the PDF with the paper or the index | Equal |
| First and last page | Look at both | The document starts and ends where it should |
| Both sides | Find a double-sided original | Backs are present |
| Legibility | Zoom to 200% on the smallest print | Readable, not blocky |
| Orientation | Page through quickly | Upright |
| Search, late page | Search for a word visible on a page near the end | Found on that page |
| Search, name | Search for a surname you can see | Found |
| Search, number | Search for an amount or date exactly as printed | Found |
| Copy and paste | Paste one page into a text editor | Readable sentences in reading order |
Step three: decide
Set the pass mark before you test, not after. A simple version: every sampled file must have the right pages, and search must work on every sampled page of clean print. Failures on handwriting and badly degraded originals are recorded but do not fail the batch, since the order did not promise them.
Keep the test sheet. It is your evidence if you reject, and your record of what was checked if you accept.
A second reading can help you tell a recognition problem from an image problem. Take two or three pages that failed, run them through another OCR tool, and read what it returns against the page. Schema Law's Free plan takes one document a day of up to three pages and shows you the result page by page on screen, which is enough for that comparison; the how it works section of the homepage shows the steps. If the second tool reads the page, the vendor's image is fine and its recognition is not. If neither reads it, look at the image.
Which line items on a scanning quote relate to searchability?
On a scanning quote, the lines that bear on searchability are recognition itself, document preparation, unitization, indexing, quality control and rework. Check that each is either priced or stated as included. A line that is missing from the quote is a service you have not been promised.
Vendors structure quotes differently, so read yours for these items under whatever names it uses:
| Line item | What it covers | Question to ask |
|---|---|---|
| Scanning, per image | Capturing each side of each sheet | Is a blank back billed as an image? |
| OCR or text recognition | Adding the text layer | Is it included in the per-image rate or separate? Which languages? |
| Document preparation | Removing staples and clips, unfolding, repairing tears | Billed by the hour or by the box? Who decides how long it took? |
| Unitization | Breaking the stream into documents | By your separator sheets, by physical rule, or by the operator's judgment? |
| Indexing or data entry | Keying file names or index fields | How many fields, taken from where? |
| Oversize and special handling | Plans, bound books, photographs, fragile paper | Per item rate, and who identifies the items? |
| Quality control | The vendor's own checking | What do they check, and on what share of pages? |
| Rework | Correcting failures | At whose cost, and within what time? |
| Delivery | Download, drive or upload to your system | Format, encryption, and how long they keep a copy |
| Paper | Storage, return or destruction | Triggered only by your written instruction? |
Ask for the quote and your specification to refer to each other by name and date. A quote that prices "scanning with OCR" against a specification that requires unitized, indexed, tested output leaves room for a dispute about which document governs. Price is for you and the vendor; this post states none, and the vendor's own quote is the only reliable source for it.
Sending a failed batch back and getting it rescanned
Reject a failed batch in writing, by file, with the test that failed, and ask for a specific remedy by a specific date. A list of file names with page numbers and failed searches tells the vendor exactly what to redo, and a general complaint does not.
What the rejection notice contains
- The delivery it refers to, by date and batch name.
- The clause of the specification that was not met.
- A table of failures: file name, page, what was tested, what happened.
- The scope you want redone: those files, that box, or the whole batch.
- The remedy for each kind of failure.
- The date by which you expect the corrected delivery.
- A statement that the paper is to be held and nothing destroyed.
Matching the remedy to the failure
| Failure | Remedy to request |
|---|---|
| Good images, no text or poor text | Recognition run again on the existing images; no rescan needed |
| Illegible, cropped or skewed images | Rescan from paper |
| Missing pages or skipped backs | Rescan the document from paper |
| Documents run together or split wrongly | Re-unitize; rescan only if separator information was lost |
| Wrong file names | Rename from the index; no rescan |
| Pages out of order or sideways | Correct in the files; rescan if order cannot be established from the images |
Ask only for what the failure needs. A request to run recognition again on good images is a smaller job than a rescan, provided the vendor can recognize text again without touching the paper, which is a point to establish when you place the order.
After the redo
Test the corrected batch with the same checks, on the files that failed and on a fresh sample, since a rework pass can introduce new errors. Replace files only after the new ones pass, and keep the rejected delivery in a separate folder until the job is closed. Then, and only then, give the written instruction about the paper.
If a failure pattern repeats across batches, stop sending boxes until it is resolved. One bad batch is an error. The same fault in three is the vendor's process, and no amount of testing on your side will fix it.