What 911 Call Analysis Is (and Why It's in Your Discovery)

When a 911 call lands in your discovery, you're looking at more than audio. You're looking at a document that a prosecutor may walk a jury through, line by line, while implying the caller's guilt through tone, pauses, and word choice. That technique has a name: 911 call analysis, and it traces back to a single master's thesis.

In 2006, former Ohio police lieutenant Tracy Harpster built the method from just 100 homicide 911 calls collected from investigating officers. Half the callers were convicted or confessed; the other half were never suspects. From that sample he extracted 16 behavioral "factors." No data was ever released. No methodology was made public.

The method spread fast, and the FBI helped. A bureau bulletin carrying the initial study reached roughly 200,000 readers without labeling the research as exploratory. Law enforcement treated it as settled practice. Harpster then taught nationwide classes and consulted on homicide cases, while the indicator list kept growing:

• 2006 thesis: 16 factors drawn from 100 calls
• 2009 paper with FBI analyst Susan Adams: 18 "guilt indicators"
• 2017 book: 38 indicators organized into the proprietary "COPS" scale across six categories

By then the sample had doubled to 200 calls, and the underlying data still wasn't public. The creators have never released how they obtained the calls or how the COPS scale was validated.

The scale of adoption is what makes this a discovery problem for you, even if your jurisdiction hasn't seen it yet. ProPublica's investigation documents the technique playing a role in more than 100 cases across 26 states. Adoption is uneven, though. One Virginia public defender told reporters he'd never encountered it. That gap is closing as training spreads.

Practically, this means you may face 911 call analysis before you've ever heard the term. It shows up as a detective's "statement analysis" during an interrogation, a prosecutor's guided transcript walk-through at trial, or a charging memo citing "Defendant Mentality." Getting a clean, timestamped 911 call transcript analysis alongside your police report and interrogation footage is how you spot where that framing crept in. The same applies to jail call transcription, where offhand comments get read through the same suspicious lens.

Treat every 911 call in your file as contested evidence until you've reviewed it yourself. The technique's origin story, its missing data, and its failure to replicate are all part of your motion record.

Why the Science Falls Apart: The Daubert and Frye Attack File

The replication record on 911 call analysis is brutal, and it maps almost perfectly onto every prong of the Daubert test.

Start with what the FBI itself found. According to a ProPublica investigation, two FBI Behavioral Analysis Unit studies failed to corroborate the technique's indicators. In the 2020 study, results were so inconsistent that some indicators pointed "in the opposite direction of what was previously found." A later BAU study concluded that applying 911 call analysis in real life "may exacerbate bias." Even John Jarvis, the FBI behavioral scientist who co-authored the original exploratory paper, told ProPublica: "I don't see where that work rises to the level of success by the scientific community."

The independent replication record is just as bad. Cornell researchers report that at least six peer-reviewed studies have failed to replicate the behavioral-checklist methodology. Across four studies involving roughly 1,800 civilians and 300 law enforcement officers, 80% of participants cited emotion as the factor most affecting their suspicion of a caller. That is a damning finding for a method built on detecting "deception cues" in tone and word choice.

Then there is the data problem. Harpster and Adams built the proprietary COPS scale from a 2006 master's thesis sample of just 100 calls, half from cases where the caller confessed or was convicted, half from cases where the caller was eliminated. They have never released the calls, never published the methodology in a form that allows independent verification, and never disclosed an error rate. As the UC Law Review puts it, the technique fails admissibility under both Daubert and Frye.

Map those failures onto the Daubert five factors and the case writes itself:

• Testability: No released data, no transparent methodology, no independent verification possible.
• Peer review and publication: The foundational thesis was exploratory; subsequent peer-reviewed replications contradict it.
• Error rate: None published. The FBI's own replication found indicators pointing the wrong way.
• Standards and acceptance: Even the FBI's BAU concluded the method "may exacerbate bias."
• General acceptance: Rejected as expert testimony in Alabama (2009) and Nevada; a Michigan conviction (Spitler) was overturned because the technique should never have been admitted.

The defense playbook is straightforward. Object under FRE 702 and Daubert when a prosecutor tries to qualify a 911 call analyst. Object under FRE 701 when a detective or lay witness is "guided" through the transcript to deliver the same specialized opinion dressed up as common sense. The replication record, the FBI's own findings, and the absence of any published error rate give you the record you need. For defense teams building the challenge, the first step is getting the 911 audio and transcript into a reviewable workspace alongside the police report and interrogation footage, so demeanor-based suspicion can be traced against the actual investigation timeline. If your case involves jail calls as well, the same jail call transcription discipline applies: timestamped, diarized, and cross-checked against the report. Evidence Copilot's grounded review workspace is built for exactly this kind of cross-document inconsistency work, with every AI output marked for human review and cited to the exact second of audio.

The science is not there. The case law is starting to catch up.

How Prosecutors Sneak It In as Lay Testimony

Prosecutors rarely try to qualify a 911 call analysis expert anymore. They don't have to. Harpster's own published guidance tells them to "guide juries through both the audio and transcript of the calls," a strategy that presents behavioral interpretation as ordinary lay observation rather than a contested forensic method. A University of Cincinnati Law Review analysis describes this as an effort to sneak the technique past expert-admissibility standards entirely.

The rule they're working around is FRE 701. Lay opinion testimony is limited to opinions rationally based on the witness's perception, helpful to a clear understanding of the testimony or a fact in issue, and not based on specialized knowledge. When a detective trained in the COPS scale narrates a 911 call and labels pauses, word choices, or flat affect as "guilt indicators," that's specialized knowledge wearing a lay witness's clothes.

Object early and with specificity. Ask for a limiting instruction. Move in limine to bar any testimony using the vocabulary of the method: "defendant mentality," "guilt indicator," "deceptive call." Then draw the line with a few pointed questions:

• Did you receive training in 911 call analysis? Who taught it, and how long was it?
• Does the phrase "guilt indicator" come from that training, or from your own experience?
• Are you telling the jury that a caller's tone or word choice reveals whether they committed a crime?
• Have you ever been qualified as an expert in this method?

The demeanor problem sits underneath all of it. Cornell research found that 80% of participants cited emotion as the factor most affecting their suspicion, and at least six peer-reviewed studies have failed to replicate the behavioral checklists. Lead author Jessica Salerno put it plainly: "There is no 'one size fits all' for responses to traumatic events." Robert Roberson, a Texas man with severe autism whose flat demeanor read as guilt, is the clearest illustration of where this leads.

For your client, that means documenting neurodivergence, disability, language background, or cultural communication style early, before the State characterizes a "hi" or a delayed plea for help as consciousness of guilt. Review the call audio against the police report, dispatch records, and interrogation footage to trace whether demeanor-based suspicion shaped the investigation. Teams running that cross-document review on a full discovery dump, including 911 call transcript analysis and jail call transcription, catch the moment a "hunch" hardened into a charge.

Your Case-File Workflow: Trace the Contamination and Build the Cross

You cannot build a suppression motion on vibes. You need a documented timeline where the detective’s "gut feeling" diverges from the actual audio.

Start by requesting the raw 911 call audio, Computer-Aided Dispatch (CAD) records, and the detective’s training materials. If the officer attended a class on 911 call analysis, that training material is discoverable.

Transcribe the call with word-level timestamps. Compare this transcript against the police report and interrogation footage. You are looking for the moment demeanor-based suspicion shaped the investigation.

Cornell researchers found that 80% of participants cited emotion as the factor most affecting suspicion. This metric is scientifically fragile, especially for neurodivergent callers. If the report describes your client as "unemotional" or "guarded," but the audio shows normal stress responses, you have a discrepancy.

This contamination often drives confirmation bias. Investigators may ignore exculpatory evidence because the initial "analysis" flagged the caller as guilty. The FBI’s own Behavioral Analysis Unit concluded that applying these methods may exacerbate bias. This is a critical point for your Daubert motion.

A grounded review workspace accelerates this process. Evidence Copilot processes the discovery dump into a searchable format.

• Diarized transcripts with word-level timestamps allow you to isolate specific phrases.
• Auto-signals flag inconsistencies between the police report and the call audio.
• Cross-exam drafts tie suggested questions directly to quoted sources.
• Suppression-motion playbooks outline the legal arguments with cited evidence.

You can start this process with 911 call transcript analysis. For related cases, jail call transcription helps track post-arrest statements.

Every AI output is marked as pending human review. The tool does not make legal conclusions. It surfaces the evidence so you can make them.

The goal is not just to transcribe. It is to prove the investigation was contaminated. When you show the jury that the detective relied on a debunked checklist rather than the facts, the credibility of the entire case erodes.

Document every inconsistency. Keep your citations tight. Let the evidence speak.