Few topics have dominated recent eDiscovery discourse as artificial intelligence. Conference agendas, vendor roadmaps, and professional commentary are all saturated with promises of document reviews, issue spotting, and fact development powered by AI. Across the industry, AI is increasingly seen as the solution to rising data volumes, tightening timelines, and mounting cost pressures.
Yet while AI has captured the industry’s attention, within its shadows a more fundamental reality has quietly been reshaping the evidentiary landscape: we are now operating in a post-email world.
Email is not disappearing, but its role is different today. Just like letters before the computer revolution, it is increasingly serving as a channel for external, formal, or business-to-business communication. Internally, organizations now rely on short-form collaboration platforms such as Slack, Microsoft Teams, and Discord for day-to-day coordination, decision-making, and escalation. These platforms have become the primary venue where intent is expressed, reactions occur, and critical decisions take shape—often informally and at speed.
For eDiscovery practitioners, this shift introduces a set of practical challenges that AI doesn’t address and, in some cases, makes even more challenging.
Conversations Treated as Documents: A Structural Mismatch
Modern chats are fluid conversations, but discovery workflows still treat them like stagnant documents.
Most review environments rely on breaking apart a conversation thread and converting them into exports broken down into 24-hour document-like records. It is like trying to fit a square peg in a round hole.
This approach exists largely to accommodate legacy review tools built for emails and e-documents rather than how short message communications actually unfold. As new forms of corporate communication continue to proliferate—from chat platforms, to Corporate knowledge bases and ticketing systems, to ephemeral messaging—legal teams are increasingly forced to confront workflows that were never designed for these types of data[1] (Casey, 2025) From an evidentiary perspective, this conversion imposes artificial boundaries that disrupt context, continuity, and meaning.
Search Limitations in Segmented Conversations
Traditional Boolean and proximity search operators assume a document context fits within its four corners. This assumption breaks down when conversations are arbitrarily segmented by a specific time period like a calendar day. Messages that are logically connected—but separated by an arbitrary time stamp—are no longer searchable together.
As a result, common constructs like AND or proximity searches can fail when analyzing short message communications. Terms existing in close conversational proximity and fall into separate 24-hour segments, can prevent a search from identifying them as related. This is not a theoretical concern; it is a genuine structural limitation directly affecting recall and completeness.
Practical Implication
Search results appearing complete with the potential of silently omitting relevant content creating a false sense of confidence, which can undermine defensibility, and increases the risk of late-stage discovery surprises.
Practical Solutions
- Search terms should be applied to full conversation threads prior to partitioning.
- To limit over capture of AND operators, the allowed distance between terms should be constrained to multiple days.
- Workflows automatically expand search hits to include surrounding messages, preserving conversational context without over-collecting irrelevant content.
Review at Scale: Noise, Redundancy, and Cognitive Overload
Short-message communications can generate volume and variability at a scale that email never did. Conversations often drift across topics, participants come and go, and are often interspersed with reactions, emojis, system notifications, and logistical chatter.
When these conversations are promoted into review platforms without proper analysis, several problems can emerge:
- High Non-Responsive Density
The majority of messages in enterprise chat environments are non-responsive in most legal matters. Reviewers are forced to wade through vast quantities of irrelevant content to locate the few messages that matter, creating vast amounts of unnecessary waste in time and money. - Limited De-Duplication Opportunities
Chat data resists traditional de-duplication. Conversations replicated across devices, exports, or platforms are rarely identical at the RSMF or document level, producing overlapping records where much of the resulting data must be reviewed independently which can be duplicative. - Increased Redaction and Inconsistency Risk
Fragmented conversations generate repeated and overlapping review decisions. This increases the likelihood of inconsistent relevance and privilege calls, excessive redactions, and reconciliation challenges during production.
Practical Implication
These factors can cause reviews to become slower, more expensive, and less consistent. Cognitive fatigue also increases, while the likelihood of inconsistent decisions and excessive redactions grows materially.
Practical Solution
- Conduct deeper analysis upstream, before content reaches final review or production
- Filter, normalize, and reduce conversations prior to document conversion for production (e.g., RSMF creation)
- Exclude system-generated noise and irrelevant messages at the earliest stage possible to allow review teams to focus on substantive content rather than raw volume
Data Density: When Scale Breaks the Model
Larger collaboration platforms are introducing an additional and growing challenge: data density.
It is increasingly common for a single conversation—such as a broad internal channel or incident-response thread—to contain thousands of messages and/or participants in a single day. The current practice of flattening this activity into 24-hour segments often produces files that are unwieldy, exceed technical limits, or lose meaningful conversational structure.
At this scale, the document metaphor collapses entirely. A single day of conversation no longer resembles a document in any functional sense—it resembles a live, evolving system of interaction across many different topics and conversing parties.
Practical Implication
Oversized, dense records reduce usability, hinder review, and obscure conversational flow. Instead of clarifying events, the documents themselves become a barrier to understanding.
Practical Solution
- Decouple conversations from documents — discovery workflows must treat these as fundamentally distinct concepts rather than forcing one into the shape of the other
- Reconstruct review documents dynamically using auto-extended context windows (e.g., ±5, 10, 25 messages around relevant content) rather than locking content into static, monolithic, and outdated artifacts
- Preserve message-level metadata natively, including dynamic contextual details such as which participants were active or had access to messages within the resulting relevant message window.
The Smallest Signal: Reactions as Evidence
The most abbreviated form of modern communication is the reaction. These can be a thumbs-up, checkmark, hearts, and other emoji-based images. They are currently often treated as peripheral noise—if captured at all. Most eDiscovery platforms do not give users the power to search for, filter on, or easily discern specific short message communication reactions or gauge their meaning. This results in reactions frequently getting overlooked, flattened into meaningless Unicode characters, or entirely omitted from analysis.
Yet in short message communications, reactions increasingly communicate decisions, an agreement, a dissenting view, acknowledgment, or someone’s intent as clearly—and often more decisively—than the classic message in written form. In fact, emojis and reaction indicators have already begun appearing regularly in litigation[2] (Austin, 2025), where they can influence how messages are interpreted by courts and juries. A single reaction has the power to denote approval for a course of action, signal alignment of views, or close a discussion without someone typing a single word.
Practical Implication
When a reaction gets ignored or treated as non-substantive by default, discovery workflows risk missing key indicators that can be highly relevant to a litigation or investigation. This is especially problematic in environments where brevity is the norm and reactions replace written responses altogether.
Practical Solution
- Treat reactions as a first-class communication element by capturing them with clear attribution to the reacting user, precise timestamps, and explicit linkage to the parent message.
- Ensure reactions remain associated with their underlying messages so reviewers can understand sentiment, agreement, escalation, or dissent within the full conversational thread.
- Modern review environments should support filtering, searching, and clear visual differentiation of reactions so they can be efficiently identified and analyzed rather than dismissed as noise.
Hyperlinked Attachments: An Unsettled Evidentiary Question
One of the most unresolved issues in modern discovery is how to treat hyperlinked attachments.
Before web-based cloud computing became the business norm, attachments were static and could be memorialized at a fixed point in time. In modern collaboration platforms, attachments are often linked to shared documents that continue to change after, for example, a message referencing them is sent.
This raises a difficult and still-unsettled question: First, should these links be treated as a) traditional attachments, even if the produced version may not reflect the state of the document at the time of the communication, or b) standalone documents, with relational references back to the messages that shared them?
Both sides of the question have rational foundations, but each introduces trade-offs. Regardless of the approach taken, the bottom line is does the produced document preserve enough information to allow reviewers an author’s original intent. Additionally, courts have also emphasized that these issues should be addressed early. In In re StubHub Refund Litigation, the court underscored that once parties agree to an ESI protocol governing linked documents, those terms will be enforced—even if the technical realities later prove more complicated than anticipated[3] (Exterro, 2024).
Practical Implication
Productions that fail to clearly explain how hyperlinked documents relate to a message’s content that include them can sow confusion and create ambiguity as to what was known, shared, or relied upon at a given moment.
Practical Solution
- Whether cloud-linked files are treated as attachments or standalone documents, the underlying file must be identified and preserved with core metadata such as filename, source path, version identifiers, and modification timestamps.
- Hyperlinked files should retain explicit linkage to the message, post, or chat where they were shared so reviewers can understand how the document entered the conversation. Productions should include the metadata and structural references necessary to trace the origin, timing, and version of the linked content, enabling accurate interpretation during investigation and review.
Refocusing on the Ground Shift Beneath the Technology
AI will undoubtedly play an important role in assisting legal teams in navigating modern data. While it is an improvement over stagnant search terms alone, enhancing natural language search, contextual classification, and summarization, it cannot fix persistent fundamental structural issues in what is increasingly the most important evidentiary evidence, short message communication.
Before we leap headlong into AI to interpret or summarize communications, practitioners must address these foundational modern data workflow issues:
- Using searches that take an entire conversation into context.
- At what point in the discovery lifecycle should short message communications get committed into an artifact for review and production?
- Providing enough information in productions that allows someone to understand the complete context of a communication.
AI can be transformative—but only if it’s fed the right inputs through the right pipes. Right now, much of eDiscovery is still running modern corporate ESI through legacy plumbing built for email and paper-shaped “documents,” forcing conversations, reactions, and cloud-linked content into brittle artifacts that lose context and distort meaning. Before we turn up the AI pressure, we need to modernize the system: treat communications as structured data, preserve relationships and message-level metadata, and build workflows that keep conversations intact until the moment they must become production artifacts. Do that, and AI becomes a force multiplier rather than a glossy overlay—because it can finally operate on evidence that reflects how organizations actually communicate.
[1] Cat Casey, eDiscovery in the Brave New Digital World (Reveal – July 21st, 2025)
[2] Doug Austin, Emojis and How to Tame Them in Discovery: eDiscovery Best Practices (eDiscovery Today – October 14th, 2025)
[3] Team Exterro, Key Lessons on Modern Attachments and Chat Applications for E-Discovery Case Law (EDRM – October 3rd, 2024)
