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

You open the discovery folder, and there it is: a 911 call transcript annotated with guilt markers. The prosecutor isn't offering expert testimony. They’re offering a detective’s "lay opinion." This is where 911 call analysis becomes a defense problem.

Retired Deputy Chief Tracy Harpster developed this method, claiming to identify deception through word choice and cadence. His proprietary "COPS" scale flags indicators like politeness, pronoun distancing, and specific vocabulary choices. The scale has expanded from 16 indicators in 2006 to 38 in his 2017 book. Yet, at least eight published studies have failed to validate these guilt indicators.

The usage is widespread. ProPublica documented the method in over 100 cases across 26 states. Police departments often pay up to $3,500 for an eight-hour training course. One co-author even admitted the original research wasn't ready for real-world application. Despite this, the technique persists because courts have rejected it as expert testimony. Prosecutors now use a workaround: they train detectives to present the analysis as lay testimony, bypassing Daubert and Frye standards.

This shift creates a specific screening challenge for your team. You need to identify these markers in the audio and transcripts before trial prep. The COPS scale looks for:

• Inappropriate politeness, such as saying "hi" to the dispatcher.
• Using "blood" instead of "bleeding" when describing the scene.
• Lack of a plea for help or immediate acknowledgment of death.
• Pronoun distancing, like referring to the victim as "they" rather than "I."

When you see these flags in a police report or a detective's notes, you’re looking at a method built on subjective interpretation. Human deception detection accuracy is only about 54%, barely better than chance. Legal analyses confirm the method fails admissibility standards when scrutinized.

For defense attorneys, the workflow starts with systematic review. You need to cross-reference the 911 audio against the detective’s annotations. Tools like Evidence Copilot allow you to upload the 911 audio and get a diarized transcript with word-level timestamps. This lets you verify exactly what was said versus what the detective claimed. You can also use 911 call transcript analysis to spot inconsistencies in the timeline. If the case involves pre-arrest communications, jail call transcription provides the same granular detail for post-arrest statements.

The institutional backlash is growing in 2026. NACDL has labeled the technique "junk science," and the Office of Indigent Defense Services in North Carolina is warning attorneys statewide. You are now defending against a method that is scientifically weak, legally contested, and institutionally condemned. Recognizing the COPS scale markers in your discovery is the first step in challenging their admissibility.

The Lay-Testimony Workaround: How Prosecutors Sneak It Past Daubert

The courts have already spoken on 911 call analysis as expert testimony, and the answer was no. Alabama rejected it in 2009, Nevada followed, and a Michigan conviction was overturned after the technique came in as expert opinion. So prosecutors adapted. Instead of qualifying a witness as an expert, they put a trained detective on the stand and let the same methodology walk in through the lay-testimony door.

The playbook isn't hidden. Harpster himself advises prosecutors to present the analysis without expert qualification, with witnesses "guiding juries through both the audio and transcript of the calls" to "create lasting impressions of this critical evidence." As the UC Law Review puts it, the work "has been mixed into in-court statements and lay testimonies to sneak it past the standards for expert testimony." Some prosecutors are candid about why: the method "will not hold up to scientific scrutiny... So they get creative," according to Innocence Project reporting. Notably, Harpster has never testified as an expert witness in a criminal trial.

Your objection framework is straightforward. Under FRE 701, a lay witness may opine only on things a normal person could perceive and rationalize. A detective who flags "pronoun distancing," past-tense references to the victim, or "inappropriate politeness" as signs of guilt is applying a specialized analytical method, and under FRE 702 that method must survive Daubert (or Frye in your jurisdiction). The framing: the lay testimony is a Trojan horse for inadmissible expert opinion. Object, move to strike, and make the court choose which standard applies.

Before you can object, you have to spot it. Screen your 911 call transcripts and detective testimony for the method's telltale markers:

• Commenting on the victim's eyes or asking whether to touch the body
• Saying "blood" instead of "bleeding" (the studies treat "bleeding" as an innocence indicator)
• Lack of a plea for help, or evasion in responses
• Pronoun distancing ("they" instead of "I") and past tense when discussing the victim
• Extraneous information or excessive politeness during a crisis call

When the witness claims special insight, hit the research record. Harpster's original 2006 thesis rested on 100 homicide calls; the 2017 book expanded to 200 calls and a proprietary 38-indicator "COPS" Scale, and the underlying data has never been released. At least eight published studies failed to validate the guilt indicators, and no one has replicated the findings. Add that humans, including trained ones, detect deception at only about 54%, barely above chance, per peer-reviewed research. An eight-hour course doesn't change that baseline.

If you're systematically screening call audio, bodycam, and transcripts for these indicators, a 911 call transcript analysis workflow with word-level timestamps makes the flagging consistent across the whole discovery dump, and jail call transcription review can catch the same coaching language surfacing in custody recordings. The cross-examination writes itself once the indicators are on paper: ask the detective who trained him, how many calls the study used, and whether the data was ever published.

Building the Cross-Examination: Indicators, Baselines, and the Low-Emotion Trap

The cross-examination starts with the transcript itself. Before you argue admissibility, you need to know exactly which "guilt indicators" the detective will claim to see, because the Harpster method runs on a fixed checklist of word choices that you can screen for in advance. A disciplined 911 call analysis review means pulling the transcript and marking each claimed indicator against what the caller actually said.

The indicators themselves are surprisingly mundane. Per the method's own materials, analysts flag callers who are "too polite," who say "blood" instead of "bleeding," who use "they" instead of "I" (pronoun distancing), or who refer to the victim in past tense. One mother took an Alford plea after a detective flagged her saying "hi" as suspicious politeness. When you screen your transcripts, watch for:

• Politeness framing: the detective characterizing courtesy or a greeting as evasive
• Word substitutions: "blood" treated as guilt, "bleeding" treated as innocence
• Pronoun and tense claims: distancing language or past-tense victim references read as deception
• The missing plea: lack of an explicit plea for help framed as suspicious

Then attack the witness's claimed ability. The training behind these opinions is an 8-hour course that departments pay up to $3,500 for, often with taxpayer funds. Set that against the peer-reviewed baseline: humans, including trained ones, detect deception at only about 54%, barely better than a coin flip. A witness who claims to read guilt from word choice after a day of instruction is claiming an ability the research says doesn't exist. Ask them to explain the validation studies, because at least eight published studies failed to validate the method's indicators, and no one has replicated Harpster's findings.

The low-emotion trap deserves its own line of questioning. Dr. Jessica Salerno's research found that both civilians and police officers become significantly more suspicious of callers who express low emotion, and for officers, even moderate emotion wasn't enough to shift perceptions. Follow-up studies using real 911 calls showed the same pattern regardless of whether the caller was later convicted or exonerated. In other words, the emotional cue isn't diagnostic of anything. Huwe Burton is the concrete example: he served more than 20 years after an officer testified he "did not seem emotional enough," and he was exonerated in 2019.

When you prepare these questions, having both sides' exact quotes with timestamps matters. Tools like Evidence Copilot surface candidate inconsistencies between a detective's testimony and the call audio, and draft suggested cross-examination questions tied to quoted sources, saved as attorney-review notes. Whether you use a workspace or do it manually, the goal is the same: every question anchored to what the caller actually said, not what the analyst claims to hear.

A Practical Review Workflow for 911 Audio, Bodycam, and Jail Calls

Start by pulling the 911 audio out of the discovery dump and getting a diarized transcript with word-level timestamps. You can't flag indicator language or cross-examine a detective on what the caller "actually said" if your transcript is a rough summary. Word-level timing matters because the whole Harpster method turns on micro-details: whether the caller said "blood" instead of "bleeding," used past tense about the victim, or never pleaded for help. Per the method's own checklist, those are the markers analysts treat as guilt signals, along with inappropriate politeness, evasion, and pronoun distancing. Eight published studies have failed to validate these indicators, so your job is to document exactly what the caller said and when, then show why the inference doesn't hold.

Once you have a grounded transcript, run a structured screen:

• Flag every COPS-scale indicator phrase in the call transcript, with timestamps, so you can quote the audio directly if a witness characterizes it differently.
• Log the dispatcher's triage decisions against the call-processing protocol, since PSAPs route calls through formal decision trees you can cross-examine on misclassification or misrouting.
• Note who trained the detective or dispatcher in call analysis. Departments pay up to $3,500 for an 8-hour course, and that training record is impeachment gold.
• Check whether the analyzing witness testifies as an expert or slips the method in as lay opinion "guiding the jury through the audio and transcript," the workaround courts have let stand when the expert door closed.

Next, map the cascade. Dr. Jessica Salerno's research shows suspicion of low-emotion callers forms regardless of whether the person was later convicted or exonerated, and that mistaken suspicion shapes the investigation that follows: which interrogations happen, which leads get prioritized, how forensic findings get read. Build a timeline from the call forward. If the detective listened to the call, formed suspicion, then pushed the interrogation in a particular direction, that sequence supports both a suppression motion and a post-conviction record. The Huwe Burton case, where a man served more than 20 years after an officer testified he "did not seem emotional enough" and was exonerated in 2019, is the concrete illustration to cite.

Finally, connect the 911 audio to the rest of the media. Bodycam of the first response, the interrogation footage, and the jail calls should sit on one timeline with the call transcript, because inconsistencies between a detective's testimony and the actual audio are where these cases break open. In practice, that means uploading the whole discovery set to a workspace that produces diarized transcripts with word-level timestamps and answers questions with citations to the exact second of audio. Evidence Copilot's 911 call transcript analysis handles the call-audio side, and the same workflow extends to jail call transcription when you need to trace what the state's witnesses said to each other later. Every AI output is marked for human review; the judgment calls, admissibility objections, cross-exam framing, stay yours.

If you want to see the workflow on your own discovery, you can book a 20-minute walkthrough at evidencecopilot.com/book, or start on the Solo plan at evidencecopilot.com/signup for $39/user/month with 15 media-processing hours included.