E-Discovery
What Is ESI? Electronically Stored Information Explained
A screenshot is ESI; the metadata that ships with it is ESI; and the rule that pulls both into discovery is broader than most attorneys remember. This guide walks the FRCP Rule 34 definition, the modern data sources courts already treat as discoverable, and the ten ESI-protocol terms small firms can win at the Rule 26(f) conference.
Legal technology researcher and data scientist specializing in AI governance for litigation teams. Expertise in NLP and AI-assisted document review.
What Is ESI? The FRCP Rule 34 Definition
ESI is electronically stored information: every byte your client generates that lands in discovery, from a Slack message to a Fitbit ping.
A plaintiff in Rossbach v. Montefiore Medical Center produced a screenshot of a harassing text message that she said had been captured on her old iPhone 5. The message carried a heart-eyes emoji rendered in the iOS 13 style. iOS 13 didn’t ship until 2019. iPhone 5 can’t run iOS 13. The Southern District of New York sanctioned her for fabrication and spoliation, and the Second Circuit affirmed in part. Rossbach v. Montefiore Med. Ctr., 2021 WL 3421569 (S.D.N.Y. Aug. 5, 2021), aff’d in part, vacated in part, 81 F.4th 124 (2d Cir. 2023).
One emoji unwound the case.
The Federal Rules of Civil Procedure adopted the term in the 2006 amendments to Rule 34. That rule lets a party request electronically stored information in any medium, with metadata, in a reasonably usable form. The text on the page is ESI. So is the device fingerprint underneath it.
Rule 34(a)(1)(A) is the operative text. A party may request “any designated documents or electronically stored information.” The rule reaches anything “stored in any medium from which information can be obtained either directly or, if necessary, after translation by the responding party into a reasonably usable form.” The 2006 Advisory Committee Note calls the phrase “expansive” and puts ESI “on equal footing with discovery of paper documents.”
Rule 34 does two things at once. It treats every byte your client generates as a discovery target. And it forces the analog instinct out of the lawyer’s head: a Slack message is a document, a smartwatch reading is a document, a heart-eyes emoji on a screenshot is a document. For the broader workflow ESI sits inside, see the e-discovery process.
How Electronically Stored Information Got Its Modern Definition
The 2006 amendments drafted around technology the drafters could not yet name. The Advisory Committee Note said “[t]he wide variety of computer systems currently in use, and the rapidity of technological change, counsel against a limiting or precise definition.” The rule should be “broad enough to cover all current types of computer-based information, and flexible enough to encompass future changes and developments.”
Read that as a design choice, not a hedge. The Committee saw email replacing paper memos and refused to rewrite the rule next time a new format arrived. So they wrote “any medium from which information can be obtained.” That phrase is the operative one. It absorbed Slack. It absorbed Teams. It now absorbs Loom recordings, Notion pages, and generative-AI prompt logs.
The breadth has been tested at the edge. In Columbia Pictures Industries v. Bunnell, the Central District of California held that server log data temporarily in random-access memory was within Rule 34’s scope on the facts. 2007 WL 2080419 (C.D. Cal. May 29, 2007). Bunnell is a district-court holding; later courts have read “stored” more narrowly when the data lives in transient memory.
Rule 34 isn’t a categorization rule. Lawyers had spent fifteen years arguing whether email was a “document” under the pre-2006 rule. The Committee resolved the argument by putting ESI on equal footing with paper.
And that is the practical result an attorney needs to walk into a Rule 26(f) conference with.
Types of ESI in 2026: Slack, Teams, Mobile, and IoT
ESI is the category. The 2026 data stack is a list. Slack threads, Microsoft Teams channels, and Notion pages all live under the same rule. None is optional in discovery.
Collaboration Platforms
Slack and Microsoft Teams are email-equivalent for discovery. The Central District of California said so in Benebone LLC v. Pet Qwerks, Inc.: producing Slack is “generally comparable to requiring search and production of emails.” 2021 WL 831025, at *4 (C.D. Cal. Feb. 18, 2021). When a Notion page sits in the same evidentiary spot as the email it replaced, “any medium” covers the page.
The cost of getting it wrong isn’t theoretical. In Drips Holdings, LLC v. Teledrip LLC, the Northern District of Ohio entered a mandatory adverse-inference instruction because defendants “did not change their Slack retention settings for ten months after receiving the litigation hold.” 2022 WL 4545233 (N.D. Ohio Sept. 29, 2022).
The District of Massachusetts went further in Red Wolf Energy Trading, LLC v. BIA Capital Management, LLC, entering default judgment under Rule 37(b)(2). 2022 WL 4112081 (D. Mass. Sept. 8, 2022). The defendants used “an unpaid novice in Kazakhstan” to search Slack. The court wasn’t impressed. See the small-firm e-discovery workflow for a five-attorney walk-through.
The non-Slack stack does the same work. A Notion page is ESI the moment a custodian uses it as the operating record; a Loom recording of a deal walkthrough is ESI on the same logic. A responding party that didn’t preserve them answers the same Rule 37(e) questions Drips and Red Wolf answered on Slack.
Modern attachments are the next fight. A hyperlinked Google Doc isn’t technically an attachment but functions as one. The Southern District of New York held in Nichols v. Noom, Inc. that “there was no meeting of the minds on whether hyperlinks were attachments.” 2021 WL 948646, at *3 (S.D.N.Y. Mar. 11, 2021). And subsequent rulings in Meta Pixel, 2023 WL 4361131, at *4–*5 (N.D. Cal. June 2, 2023), and Uber, 2024 WL 1772832 (N.D. Cal. Apr. 23, 2024), updated Noom for the cloud era.
Three rulings, one trajectory.
Ephemeral Messaging
Auto-deleting messages give opposing counsel a Rule 37(e) framing that is hard to rebut. Choosing the tool after a hold is evidence of intent under Rule 37(e)(2). In WeRide Corp. v. Kun Huang, the Northern District of California issued terminating sanctions against three defendants where “[t]he amount of spoliation that AllRide concedes is staggering.” 2020 WL 1967209 (N.D. Cal. Apr. 24, 2020). Judge Alsup’s in limine order in Waymo v. Uber, No. C 17-00939 WHA (N.D. Cal. Jan. 30, 2018), let plaintiff argue that defendant’s Wickr use explained the missing evidence.
Two satellite rulings:
- Fast v. GoDaddy.com LLC on Facebook Messenger unsend and Telegram, 340 F.R.D. 326 (D. Ariz. 2022).
- FTC v. Noland on Signal and ProtonMail, No. CV-20-00047-PHX-DWL (D. Ariz. Aug. 30, 2021).
Mobile and IoT
The phone in the witness’s pocket is a minicomputer, and the Supreme Court has said so. Riley v. California, 573 U.S. 373, 393 (2014). Civil discovery doesn’t get Fourth Amendment protection, but mobile volume and intimacy are why every ESI protocol now negotiates mobile collection separately. Carpenter v. United States extended that intimacy concern to cell-site location records, echoing Riley’s “privacies of life” framing. 585 U.S. 296, 311 (2018). On IoT, State v. Dabate, No. SC 20749, 2025 WL 730530 (Conn. Mar. 10, 2025), affirmed a murder conviction where Fitbit data showed the victim moving an hour after the defendant said she had been killed. Civil parallels: activity trackers in personal-injury, location pings in employment, smart-speaker logs in wrongful-death.
Authentication of mobile data has become its own specialty. Rossbach is the ESI thesis in one fact pattern.
FRCP ESI Rules: Rule 26(b)(1), Rule 26(f), and Rule 37(e)
Rule 26(b)(1) sets the scope. Discovery reaches “any nonprivileged matter that is relevant to any party’s claim or defense and proportional to the needs of the case,” with proportionality assessed by six factors: the importance of the issues at stake, the amount in controversy, the parties’ relative access to relevant information, the parties’ resources, the importance of the discovery in resolving the issues, and whether the burden or expense of the proposed discovery outweighs its likely benefit. Fed. R. Civ. P. 26(b)(1). Six factors, all on the record.
Rule 26(b)(2)(B) is the companion lever: a responding party may identify sources as not reasonably accessible due to undue burden or cost, shifting the burden to the requesting party to show good cause. If the court orders production, it may impose conditions, including cost-shifting under Sedona Principle 13.
The 2015 amendments returned proportionality under Rule 26 to the body of the rule. The 2015 Advisory Committee Note read the change to forbid “a boilerplate objection that [discovery] is not proportional.” Specific objections, factor by factor, are the standard.
Boilerplate is dead.
The ESI protocol itself gets born under Rule 26(f)(3)(C). Together with Rule 16(b)(3)(B)(iii), those rules require the discovery plan to address ESI preservation, search, and production form, and let the scheduling order do the same. But skip the conference, and you’ll give the opposing party the form fight, the custodian fight, the search-term fight, and the privilege-log fight in a single motion at month four.
Run the conference; you’ll save the budget.
Rule 37(e) is the spoliation provision, and it has two tiers. The 2015 rewrite was meant to end the circuit split around adverse-inference instructions. The (e)(1)/(e)(2) line carries two different sanctions on two different findings.
- (e)(1) trigger: prejudice from a preservation lapse. On a finding of prejudice, “measures no greater than necessary to cure the prejudice.” Curative discovery, depositions, or fee-shifting are the typical menu.
- (e)(2) trigger: intent to deprive. Only on a finding that the party “acted with the intent to deprive another party of the information’s use in the litigation,” the court may presume the lost information was unfavorable, instruct the jury to presume the same, or dismiss or enter default. No separate prejudice finding is required at this tier. In Drips the intent finding rested on the ten-month retention delay; in WeRide, on the volume of conceded spoliation. Intent rarely turns on a single act; courts read it from a pattern.
Don’t conflate the two. The 2015 Note expressly rejected pre-2015 cases like Residential Funding Corp. v. DeGeorge Financial Corp., 306 F.3d 99 (2d Cir. 2002), that allowed adverse-inference instructions on a finding of negligence or gross negligence. Negligence or gross negligence triggers (e)(1) measures. Intent triggers (e)(2). And the gap between them, in dollar terms, is the difference between a curative order and a mandatory adverse-inference instruction or worse.
The ESI Protocol: 8 Points Small Firms Should Negotiate
An ESI protocol is a stipulated case-management order, usually entered under Rule 16(b)(3)(B)(iii), that fixes how parties will identify, preserve, search, and produce ESI. Sedona Principle 6 places primary methodology responsibility on the responding party. Principle 8 directs parties to primary-source data first, then to less accessible sources only as long as the request remains proportional. The Sedona Cooperation Proclamation, favorably cited by Judge Grimm in Mancia v. Mayflower Textile Services Co., 253 F.R.D. 354, 358–65 (D. Md. 2008), reframes cooperation as zealous advocacy.
Eight points to negotiate.
- Custodians and date ranges. Rule 26(b)(1); Pippins. Responding: a named list capped to actual decisionmakers. Requesting: a wider list with a written justification per name.
- Search terms, TAR, GenAI methodology. Da Silva Moore; Rio Tinto; Hyles, 2016 WL 4077114, at *2–*3 (S.D.N.Y. Aug. 1, 2016). Responding: keyword plus CAL (continuous active learning) with a sampling plan. Requesting: agreed seed sets and validation sampling. See GenAI versus TAR.
- Form of production. Rule 34(b)(2)(E)(ii); Sedona Principle 12. Responding: TIFF plus load file plus a defined metadata set. Requesting: native files for spreadsheets, presentations, and collaboration data.
- Modern attachments. Noom; Meta Pixel; Uber. Responding: hyperlinks treated as references unless contemporaneously embedded. Requesting: hyperlinks collected as the document the link resolved to at the time of the parent message.
- Privilege log format. Rule 26(b)(5). Responding: categorical logging (grouped by privilege type and custodian) for high-volume custodians. Requesting: document-by-document logging with metadata fields.
- FRE 502(d) clawback order. Specify the entering court (the court signing the order, which fixes cross-jurisdictional effect). Responding: a 502(d) order at the start of any non-trivial production. Requesting: clawback narrowed to attorney-client material.
- Collaboration platform handling. See Section III. Responding: Slack channels and DMs with reactions, edits, and deleted messages where retention permits. Requesting: per-channel custodian lists with date ranges.
- Mobile device collection. Riley. Responding: targeted text-message export by date and contact. Requesting: full forensic image when warranted by the claim.
“The primary source of ESI to be preserved and produced should be those readily accessible in the ordinary course. Only when ESI is not available through such primary sources should parties move down a continuum of less accessible sources until the information requested to be preserved or produced is no longer proportional.”
Add a modification clause. Michael Berman put it in his June 2023 commentary: a court-ordered ESI protocol is an interlocutory order subject to reconsideration, and good drafting provides an escape hatch. A poorly drafted protocol can override Rule 26 proportionality and saddle the responding party with obligations that exceed the rule.
On methodology, Rio Tinto remains the working standard: it is “inappropriate to hold TAR to a higher standard than keywords or manual review.” 306 F.R.D. at 129. The counterweight is Hyles, where the same court declined to compel TAR over the responding party’s objection.
How AI Changes ESI Review and Authentication
TAR received its first federal judicial approval in 2012, when Magistrate Judge Andrew Peck held that “computer-assisted review is an acceptable way to search for relevant ESI in appropriate cases.” Da Silva Moore v. Publicis Groupe, 287 F.R.D. 182, 183 (S.D.N.Y. 2012). Three years later, the same judge wrote that holding TAR “to a higher standard than keywords or manual review” was inappropriate. Rio Tinto, 306 F.R.D. 125, 129 (S.D.N.Y. 2015). That logic extends to GenAI-assisted review and vector-indexed retrieval under Sedona Principle 11 and the 2015 proportionality rewrite. The standard is reasonableness in context.
It isn’t perfection.
What changes is the lawyer’s competence duty. ABA Formal Opinion 512 (July 29, 2024) restated Model Rule 1.1 comment [8] for the GAI era. The Opinion warned that GAI tools “lack the ability to understand the meaning of the text they generate or evaluate its context,” and are “prone to ‘hallucinations.’” A small firm using AI on review now owes a competence duty in the AI tool itself. The Op. 512 duty interlocks with Rule 26(g), which requires counsel to certify after reasonable inquiry that responses are complete and correct. AI-assisted review without verification is a Rule 26(g) problem before it is an ethics problem.
Authentication is where AI bites back. Mendones v. Cushman & Wakefield, Inc., No. 23CV028772, 2025 WL 2613764 (Cal. Super. Ct. Alameda County Sept. 9, 2025), caught AI-fabricated evidence on a forensic-metadata theory: “Apple did not introduce Apple Intelligence until iOS18 and required an iPhone 16, iPhone 15 Pro, or iPhone 15 Pro Max.” The device couldn’t have generated the artifact. Same forensic-metadata method Rossbach used four years earlier. AI-fabricated evidence leaves device-and-OS fingerprints a forensic expert can surface on a budget a small firm can defend in a Rule 26(b)(1) proportionality argument.
An iOS-13 emoji once unwound a harassment screenshot the same way.
Modern ESI breaks the keyword-and-folder model. Vector indexing across email, Slack threads, exported Teams channels, and PDFs lets a single question pull cited passages from every format at once. Hintyr is one example: an agentic review system that returns answers with page-level citations the responding attorney still verifies before producing, the way Op. 512 expects.
The agent proposes; counsel signs.
This post is for informational purposes and does not constitute legal advice. Court rulings on collaboration platforms, ephemeral messaging, and AI-generated evidence vary by jurisdiction; consult counsel licensed in your jurisdiction before adopting an ESI protocol or AI-review workflow.
Run AI Document Review Across Your Modern ESI
Modern ESI lives in Slack threads, Teams exports, mobile messages, and AI chat logs, not just email and PDFs. Hintyr indexes the full collection and answers your case questions with cited excerpts you can verify on the source page, the way Op. 512 expects. Built for the firms that staff their own document review.
Frequently Asked Questions
What is ESI?
ESI stands for electronically stored information. The Federal Rules of Civil Procedure adopted the term in 2006 when Rule 34(a)(1)(A) put discovery of ESI on equal footing with paper documents. The Advisory Committee deliberately wrote the rule to be expansive, covering any information stored in any medium so it would absorb future formats. Today that includes email, Slack and Teams messages, mobile data, cloud documents, IoT logs, and AI chat histories.
When was ESI added to the Federal Rules of Civil Procedure?
The 2006 amendments to the Federal Rules added ESI as a category co-equal with documents. Rule 34(a)(1)(A) was rewritten to allow inspection, copying, testing, and sampling of any electronically stored information stored in any medium. The 2006 Advisory Committee Note explained the choice as deliberate: the rule should be broad enough to cover all current types of computer-based information, and flexible enough to encompass future changes and developments. The same Note confirmed ESI stands on equal footing with paper documents.
Are Slack and Microsoft Teams messages ESI?
Yes. Slack and Microsoft Teams messages are ESI under Rule 34(a)(1)(A). In Benebone LLC v. Pet Qwerks, Inc. (C.D. Cal. 2021), the court held producing Slack is generally comparable to producing email. Drips Holdings v. Teledrip (N.D. Ohio 2022) imposed a mandatory adverse-inference instruction after the defendant left Slack retention at seven days for ten months after a litigation hold, and Red Wolf Energy Trading v. BIA Capital (D. Mass. 2022) entered default judgment under Rule 37(b)(2).
Are text messages and ephemeral chat apps ESI?
Yes. Text messages, WhatsApp, Signal, Wickr, and Snapchat content are ESI subject to preservation under Rule 37(e). Courts treat the choice to use auto-deleting tools after litigation is anticipated as a Rule 37(e)(2) intent-to-deprive question, not a software detail. WeRide Corp. v. Kun Huang (N.D. Cal. 2020) imposed terminating sanctions after staggering Slack and ephemeral-messaging spoliation. Authentication of texts requires forensic metadata, as Rossbach v. Montefiore Medical Center (S.D.N.Y. 2021) showed when an iOS-13 emoji exposed a fabricated screenshot.
What is an ESI protocol?
An ESI protocol is a stipulated case-management order, usually entered under Rule 16(b)(3)(B)(iii), that fixes how parties will identify, preserve, search, and produce electronically stored information. Standard provisions cover custodians and date ranges, search terms or technology-assisted review methodology, form of production, treatment of modern attachments such as hyperlinked Google Docs, privilege log format, an FRE 502(d) clawback order, collaboration platform handling, mobile collection, source code, and cost allocation. Sedona Principle 6 places primary responsibility for methodology on the responding party.
How does AI change ESI review?
AI changes ESI review along three axes. Technology-assisted review has been judicially approved since Da Silva Moore v. Publicis Groupe (S.D.N.Y. 2012), and Rio Tinto v. Vale (S.D.N.Y. 2015) called TAR by a willing producing party black letter law. Vector indexing and semantic retrieval extend that logic to GenAI tools under Sedona Principle 11. ABA Formal Opinion 512 (July 2024) requires lawyers to be competent in the AI tools they use, and Mendones v. Cushman & Wakefield (Cal. Super. Ct. 2025) shows how device metadata can unmask AI-fabricated evidence.