Make an Unsearchable Image Production Usable Under Rule 34

An image-only production looks like documents, yet nothing in it can be searched until text is restored.

A production of a little bit more than 180,000 documents, delivered as TIFF images with load files, was held sufficient by a federal court, and the motion to compel native files was denied. The plaintiffs, the court observed, "have not claimed, let alone shown, that the emails are unsearchable." That ruling, dissected in a 2024 episode of the Meet and Confer podcast, is the hollow center of every image-dump dispute.

What, then, can your own file prove about these images? The research section below walks through what such proof must contain.

So the question is not merely how to make dead pages speak. It is how to make them speak without conceding that their silence was acceptable. Optical character recognition (OCR) will lift text from an image; a written objection will keep the form dispute alive; and the two, done in the right order, are not at war.

There is a trade-off here that I will not pretend away. Every hour spent processing the other side's images is an hour they did not spend, and yet waiting for a motion to ripen can cost the deposition you needed those pages for. Which cost to bear depends on three things, and the first of them is the calendar.

What does Rule 34(b)(2)(E) actually give you, and what does it withhold?

Rule 34(b)(2)(E) sets three defaults and one hard limit, so a requester who named no form rarely wins a second one, and self-processing at cents per document becomes the rational fallback.

Before choosing any remedy, read your own requests against the rule in this order:

1. Check whether your requests for production, or an ESI stipulation, named a form such as native files or TIFF with text.
2. Check whether the images preserve what the producer ordinarily maintains, including extracted text and metadata.
3. Check whether the parties agreed on any form, because an agreed form can lift the production outside the rule altogether.

The rule itself is short. As Kelly Twigger read it on her Meet and Confer podcast in 2024, "unless otherwise stipulated or ordered by the court," a party must produce documents as kept in the usual course of business or organize and label them to the requests; where no form is specified, it must produce ESI "in a form or forms in which it is ordinarily maintained or in a reasonably usable form or forms"; and it "need not produce the same electronically stored information in more than one form." A reasonably usable form is one the receiving party can work with in roughly the way the producer could: searched, sorted, read in context.

That last clause is the trapdoor. Twigger put it bluntly: a requester who waits and receives something reasonable "will not be entitled to ask for it in a second form." Silence, in such a request, is not neutrality. It is a choice whose cost arrives later.

Contrary to what many producing parties assume, the reasonably usable option is no free substitute either. Craig Ball, speaking on the Bow Tie Law podcast in 2014, argued that a producer must honor a requested form or object and obtain relief from the court; it may swap in a reasonably usable form only where the requester specified none. In the Anderson Living Trust matter he discussed, Judge Browning went further and held that the parties' own agreement on form acted as a stipulation, removing the production from Rule 34(b)(2)(E) entirely.

Courts weigh usability unevenly, and therein lies the strange and dreadful uncertainty of the whole terrain. Ball described a split wherein some cases accept searchable PDFs from a silent requester, others hold that non-searchable PDFs are not reasonably usable, and many object to converting searchable ESI into static images. Images that arrive stripped of their text layer and load files also break downstream, as explained in why a production won't load into another review tool.

Set side by side, Rule 34(b)(2)(E) and its commentators leave one question for your own file: which of the three subparts did your requests actually invoke? Ask, then, what your own requests said. Everything afterward is salvage, and salvage has its own rules, which begin with how you weigh re-production against processing the images yourself.

Should you demand re-production or OCR the images yourself?

Choose by three factors, deadline distance, native availability and opponent cooperation; for a small firm on one messy matter, self-processing is affordable enough to start at once while the objection stands.

Apply the rule in this order, and stop at the first factor that decides it:

1. Deadline distance. If a deposition, expert report or dispositive motion falls before any re-production could plausibly arrive, process the images now and keep the objection alive in writing.
2. Native availability. If you specified a form, or the producer stripped text or metadata it ordinarily maintains, demand re-production; the rule is on your side and the record should show you used it.
3. Opponent cooperation. If opposing counsel will stipulate to a searchable form, take the stipulation; if they stonewall and time allows, move to compel with facts, not argument.

Cooperation is not a fantasy. In the Anderson Living Trust dispute, as a 2014 Bow Tie Law podcast episode recounted, the parties agreed that the defendants would scan 17,000 to 20,000 paper documents, run them through optical character recognition (OCR, the conversion of pictured letters into searchable text), and produce searchable PDFs instead of paper. One speaker, drawing on service-provider experience, added that searchable PDF output is merely a setting in the OCR engine. The burden of searchability, then, is slight. The fight over it is not.

That asymmetry is why I treat a motion as the exception rather than the reflex. Conventional wisdom says move to compel whenever images arrive dark and textless; yet where no form was named, the motion buys delay, fees and a doubtful result. Cost decides much of the rest. Our own economics are plain: AI-assisted review runs cents per document, against dollars per document and roughly $19K/GB for manual review, a delta a firm can measure on its own matter. For a solo or small firm, the segment enterprise platforms price out, that gap is the whole argument; the comparison of per-matter versus annual platform pricing shows how it lands on a single case.

Self-processing also yields something re-production never does. In our platform, every answer links back to the exact exhibit it came from, so an attorney can verify with one click before filing. A re-produced set is merely readable. A processed one is already half argued, provided the objection was preserved before the first page went through the engine.

How do you OCR the images without accepting the defective form?

Object in writing first, then process a copy while the produced originals stay immutable and content-hashed, with an append-only audit trail recording every step you took.

The objection is a paper act before it is a technical one. I would sequence the work in three moves:

1. Object and reserve. Send a written form-of-production objection that names the defect (no text layer, missing load files, stripped metadata), states that any processing is undertaken solely to meet case deadlines, and expressly reserves the right to seek re-production or costs.
2. Freeze what you received. Store the production exactly as delivered, hashed on intake, so the record can always show the form the other side chose.
3. Process a working copy. Run OCR and issue-coding on derivatives, never on the originals, and log each pass.

Why does the freezing matter so much? Because the danger in self-help is not the OCR itself; it is the quiet suggestion, made later by opposing counsel, that you found the images usable enough. A hashed original is the answer to that suggestion. It stands beneath all the derivative work like a sealed reliquary, unchanged, wherein the defect remains visible to any judge who asks to see it.

Our platform is built for exactly this posture. Originals are kept immutable with content hashing, the audit trail is append-only, chain of custody is documented, and a fail-closed privilege gate sits on production, so the process holds up if opposing counsel challenges it. The same discipline that keeps privileged material from leaking out, described in designing a privilege gate so privileged ESI never reaches review, also keeps your working copies from blurring into the record you received.

A second worry follows close behind. Running the other side's production through an AI tool raises the question of where those pages go. Processing in our environment runs single-tenant, or inside your own AWS account under your own keys, with no vendor retention and no model training. Nothing you feed it trains anyone's model. Nothing leaves your control.

The working copy, once coded, becomes more than a searchable heap. Enterprise platforms supply the defensible spine, meaning Bates numbering, privilege handling and chain of custody; newer AI tools supply cited questions and answers and chronologies. Combining both in one tool means the processed set is at once defensible and interrogable.

The proof is best seen on one's own documents. We invite firms to bring a messy collection and a hard question, then watch the system read, code and cite their own evidence rather than reading a case study about someone else's. An image dump that arrived dark and dumb is precisely such a collection, and the hard question is usually the one the producer hoped you would never get to ask.

Where is the image-production fight heading, and what should you do next?

The fight is moving earlier: name a searchable form in your first requests, and when images still arrive dark, object in writing and process them yourself without delay.

I expect the leverage to keep migrating toward the request and the meet-and-confer, because after production the rule offers so little. The research section above walks through those findings; what remains is to put the form clause, and the written objection, on your own calendar.

That is the whole of it. Keep a log of every search that fails on the images. Ask in writing whether extracted text already exists. Hash what you received, OCR a copy, code it, and let every finding point back to its page.

Whichever review platform you weigh, Relativity or one of its rivals, ask the same question of it: can it take a dead production, make it searchable, and still prove what the other side actually sent? The answer decides more than the price does, and it is the answer opposing counsel will test first.