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

A 911 call placed in the worst moment of someone's life can now be used to convict them, and the technique behind that inference is sitting in your discovery file.

The practice is called 911 call analysis, and it trains detectives and dispatchers to flag a caller's word choice, cadence, and grammar as indicators of guilt or innocence. The dominant framework is the COPS Scale (Considering Offender Probability in Statements), built by retired Deputy Chief Tracy Harpster. His 2006 master's thesis drew on 100 calls, his 2017 book on 200 calls, and the indicator list has grown from 16 factors to 38 across six categories covering who the call is about, what it describes, how it is made, passive defenses, resistance to the dispatcher, and additional guilt markers.

The reach is wider than most defenders realize. ProPublica documented more than 100 cases in 26 states where Harpster's methods were used, and as of March 2024 he had consulted on at least 2,500 homicide investigations. Departments pay up to $3,500 for eight hours of training, often with taxpayer funds. The FBI is cited as one of the agencies that helped the technique spread.

The indicators themselves can sound absurd until you see them applied to a real transcript. A 2018 Washington case had a detective flagging a mother's first word "hi" as too polite. In a 2012 Fremont, Ohio investigation, Detective Jason Kiddey used call analysis to elicit a confession that led to a manslaughter plea and an 11-year sentence. Huwe Burton was 16 when he called 911 to report his mother's murder; an officer testified Burton "did not seem emotional enough." He served more than 20 years before exoneration in 2019.

The COPS Scale's core indicators, the ones your client's call may already contain, include:

• The call is about someone other than the caller, or centers on the caller as a victim
• The caller fails to identify themselves by full name, or identifies a victim before saying their own name
• The caller displays low or "inappropriate" emotional expression, or shows none at all
• The caller resists or redirects the dispatcher's questions, or volunteers unsolicited information
• The caller uses passive voice, distancing language, or omits first-person pronouns

These markers are not science. An FBI Behavioral Analysis Unit evaluation (O'Donnell et al., 2025) found the COPS Scale misclassified calls over 40% of the time and concluded it "should not be used in actual investigations to evaluate callers' potential innocence or guilt." A Villanova replication (Markey et al., 2025) confirmed the scale failed to differentiate guilty from innocent callers, and at least eight subsequent studies have failed to replicate its claimed indicators. Harpster and Adams have never released their underlying data, contrary to standard research practice.

The deeper problem is how the technique arrives in court. Cornell Law Review (2026) labels the pattern "shadow forensics": the methods are never disclosed and the founders are never proffered as experts, so the analysis largely escapes Daubert or Frye challenge. Instead, trained detectives and dispatchers deliver it as lay opinion, a workaround Harpster's own training materials reportedly coach prosecutors to use.

That is why your discovery review matters. A grounded, word-level transcript of the 911 call lets you test every claimed indicator against the actual second it occurred, build the impeachment record on the witness's own training, and surface the taint that flows from a "guilty" call analysis into the interrogation that followed. For defenders reviewing the full evidence stack, that same timestamped workflow extends to jail call transcription and the bodycam and interrogation footage that frame the call. The technique is already in your file; the question is whether you see it before it sees the jury.

The 2025 Science That Kills the COPS Scale

If you're facing a prosecutor who wants to use 911 call analysis at trial, 2025 handed you the best ammunition this technique has ever faced: the government's own researchers took the COPS Scale apart, and the results were brutal.

An evaluation by the FBI's Behavioral Analysis Unit (O'Donnell et al., 2025) found the scale misclassified calls over 40% of the time and concluded it should not be used in actual investigations to evaluate callers' potential innocence or guilt. That's the FBI, testing a method built by a retired deputy chief, and telling law enforcement to stop using it. Both 2025 studies appeared in Psychology, Public Policy, and Law, so they're citable, peer-reviewed, and on the record.

The independent replication failed too. A Villanova team (Markey et al., 2025) found the COPS Scale failed to differentiate between guilty and innocent callers, and that core assumptions (like the idea that deceptive callers avoid urgency or sound unemotional) simply didn't hold up. When researchers compared effect sizes, the pattern was textbook: the scale's own authors reported much stronger results than independent replications ever could. And that's on top of at least eight prior studies that failed to replicate the method's claimed indicators, according to the Innocence Project.

Dr. Jessica Salerno's research explains why the technique feels persuasive to officers anyway. In her experiments, both civilians and police became more suspicious when callers expressed low emotion, regardless of whether the caller was actually guilty. In follow-up studies using real 911 calls with known outcomes, the same suspicion pattern emerged whether the caller was later convicted or later exonerated. In other words, the "indicators" track emotional expression, not deception. For officers, even moderate emotion wasn't enough; callers had to sound highly emotional before suspicion dropped. That's your mechanism argument, and it maps directly onto the Daubert factors: error rate above 40%, no published data from the founders, no general acceptance among independent researchers.

When you're building the challenge, pull the key findings into one place so you can drop citations into a motion in limine or a Daubert voir dire without hunting:

• FBI BAU (2025): over 40% misclassification; explicit recommendation against investigative use.
• Villanova replication (2025): scale failed to distinguish guilty from innocent callers.
• At least eight failed replication studies of the claimed indicators.
• Founders' data never released: no sample sourcing, no methodology, no public dataset.

This is also where your own transcript work pays off. Every claimed "indicator" is a statement about specific words at specific moments, so you need a word-level, timestamped transcript to test each one against the audio. Grounded 911 call transcript analysis lets you pin every claimed indicator to the exact second it did or didn't occur, and the same workflow covers jail call transcription when the state tries the same word-choice tricks downstream. The government built your reliability record for you. Your job is to make sure the judge actually reads it.

Shadow Forensics: How It Slips Past Daubert as Lay Testimony

Most defenders expect a junk-science fight to happen at the bench under Daubert. With 911 call analysis, it almost never does. The technique is engineered to skip that step entirely.

A 2026 Cornell Law Review article by Wisconsin Innocence Project Co-Director Christopher Lau names the pattern "shadow forensics." The architecture is simple. The underlying methods stay undisclosed. The founders never get proffered. The analysis arrives in court as the opinion of a trained detective or dispatcher, and it walks in under FRE 701 as lay testimony.

That structural choice is not accidental. Tracy Harpster, the retired deputy chief whose 2006 master's thesis and 2017 book produced the COPS Scale, has not taken the stand as an expert in a criminal trial, according to UC Law Review. His work has been "mixed into in-court statements and lay testimonies to sneak it past the standards for expert testimony."

Once you see it, you see it everywhere. A detective testifies the caller "seemed too calm." A dispatcher says the caller "kept avoiding the question." Both phrases come straight from the COPS Scale indicator list. Neither witness is offered as an expert. Neither is subjected to a Rule 702 voir dire. The reliability fight never happens.

The technique's own training materials reportedly walk prosecutors through this. Harpster's "ten tips" include guiding juries through the audio and transcript to "create lasting impressions of this critical evidence," per the same UC Law Review analysis. The goal is to get the indicators in front of the jury without ever triggering expert-qualification scrutiny.

Judicial pushback is real but inconsistent:

• A 2009 Alabama trial court rejected Harpster as an expert.
• A Nevada judge excluded a detective trained in the method.
• Michigan's Riley Spitler conviction was reversed after the trial court admitted a trained detective as an expert based on the training being recognized by the Michigan Commission on Law Enforcement Standards.

Each ruling matters, but each one came after the technique had already been admitted and argued to a jury. The Spitler reversal is the cleanest defense precedent because it ties the admissibility question directly to the training-pipeline workaround.

The empirical case against the COPS Scale is unusually strong. An FBI Behavioral Analysis Unit evaluation (O'Donnell et al., 2025) found calls misclassified over 40% of the time and concluded the scale "should not be used in actual investigations to evaluate callers' potential innocence or guilt," as summarized by the APA. A Villanova replication (Markey et al., 2025) found the scale failed to differentiate guilty from innocent callers, and effect-size comparisons showed the COPS Scale's own authors reported much stronger results than independent replications.

That record maps cleanly onto the Daubert factors and gives you the record you need the moment a trained detective opens their mouth on the stand.

Your first move is rarely a Daubert motion. It is a standing FRE 701(c) objection the second a lay witness starts opining on caller culpability, paired with a demand for a Rule 702 voir dire. Cornell's Dr. Jessica Salerno has explicitly told defense attorneys to "learn to identify and challenge attempts to introduce 911 call analysis through lay testimony," per the Innocence Project. Build the objection early, cite the Spitler reversal and the Alabama and Nevada rejections, and put the FBI misclassification rate on the record before any "indicator" reaches the jury.

You also need your own clean, timestamped record of the call so you can test every claimed indicator against the actual audio. A grounded workflow with 911 call transcript analysis and jail call transcription gives you word-level timing you can walk a witness through on cross, and a citation trail that survives the post-conviction record.

Your Three-Front Attack: 701 Objections, Daubert Record, and Taint Litigation

Because the technique almost never arrives as qualified expert testimony, your challenge has to be built before trial and aimed at three separate fronts at once: the lay-opinion door, the reliability record, and the tainted investigation behind it.

Front one: the standing motion in limine. Harpster has never taken the stand as an expert in a criminal trial, according to UCLA Law Review; instead, his method gets "mixed into in-court statements and lay testimonies to sneak it past the standards for expert testimony." Dr. Jessica Salerno's advice to defenders is blunt: learn to identify and challenge attempts to introduce 911 call analysis through lay testimony. Your template should do two things. First, demand a Rule 702/Daubert voir dire the moment a trained detective or dispatcher starts opining on caller culpability. Second, object under FRE 701(c) on the ground that the opinion rests on scientific, technical, or specialized knowledge that belongs within the scope of FRE 702. File it in every case where the call audio shows up in discovery, so the objection is already on the record when the witness testifies.

Front two: build the Daubert/Frye record now. The 2025 empirical studies are your exhibit list. An FBI Behavioral Analysis Unit evaluation found the COPS Scale misclassified calls over 40% of the time and concluded it should not be used in actual investigations to evaluate callers' guilt or innocence. A Villanova replication found the scale failed to differentiate guilty from innocent callers, and the effect sizes reported by the method's own authors dwarfed those in independent replications. Map those findings onto the five Daubert factors: testability, peer review, error rate, standards, and general acceptance. Then add the persuasive authority: the Riley Spitler reversal in Michigan, the 2009 Alabama rejection of Harpster as an expert, and the Nevada judge who refused a trained detective.

Front three: taint litigation. The strongest structural argument is that the "analysis" poisoned everything downstream. The Innocence Project reports that mistaken suspicion is rarely corrected by the investigation that follows; more often, it shapes the investigation itself. The FBI study found that judgments of caller involvement influenced the nature of subsequent interview questions. That gives you a confirmation-bias theory tying the call "analysis" to tainted interrogations, narrowed forensic interpretation, and selective lead pursuit, which supports both a motion to suppress and a request for a taint hearing.

Because the founders never proffer themselves as experts and never publish their data, discovery is where you do damage. Serve Brady/Giglio and Rule 16 demands for:

• Training certificates showing the detective or dispatcher attended a Harpster-affiliated or COPS Scale course (departments pay up to $3,500 for eight hours of instruction)
• Consultation invoices, emails, or call notes involving Harpster, whose methods ProPublica documented in more than 100 cases across 26 states
• Any internal policy or directive on the use of 911 call analysis

Then cross the witness on the eight failed replication studies and the FBI's misclassification rate. To run that cross, you need your own word-level, timestamped transcript of the call, so every claimed "indicator" can be tested against the exact second it occurred. That's the workflow behind Evidence Copilot's 911 call transcript analysis: diarized, timestamped transcripts with citations into the audio itself. The same tooling covers jail call transcription when the taint trail runs into post-arrest calls. Every answer stays marked AI-generated and pending attorney review, so the record you build is yours to verify.

Build Your Own Word-Level Record of the Call

The technique operates on word choice, cadence, and the grammar of a caller in crisis. So the defense needs a record that can test every claimed "indicator" against the actual second it occurred. A flat transcript won't cut it.

A word-level, diarized transcript is the foundation. You need to know exactly when the caller's first sentence lands, how long the dispatcher waited before responding, and where pauses cluster. Without that granularity, you cannot rebut a detective's claim that the caller "delayed" or "avoided" specific words. The audio itself stays untouched; the transcript becomes the working copy you mark up, cite, and attach to motions.

The review workflow runs in three passes. First, read the transcript cold for content. Second, build a timeline that pairs each flagged indicator with its timestamp. Third, lift the exact quotes you want for cross-exam into a separate file with citations locked to the audio. This is also where jail calls enter the picture. A defendant's later statements from the jail often mirror or contradict the 911 call's emotional register, and pairing the two transcripts side by side exposes whether the prosecution's "guilt indicators" are actually consistent with how the same person talks in a non-emergency recording. A grounded workspace that handles 911 call transcript analysis and jail call transcription on the same case clock makes that comparison straightforward rather than a manual slog.

Preserve the record with the same rigor you bring to impeachment prep. Save the diarized transcript, the timeline, and the audio hash to the case file. If a conviction happens, that package becomes the spine of post-conviction review, where the FBI BAU's finding that the COPS Scale misclassified calls over 40% of the time, and the eight failed replication studies, can be paired with the caller's actual words at the actual seconds the prosecution relied on.

The tactical payoff is concrete: a timestamped record lets you object under FRE 701(c) when a trained detective starts narrating "indicators" as if they were lay observations, and it gives the appellate record everything it needs if the trial judge lets the testimony in anyway.