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

If a prosecutor's theory of guilt rests partly on how your client answered a dispatcher, you need to know about 911 call analysis before the trial starts. The technique claims to detect guilt from a caller's word choice, cadence, and grammar, and it has quietly shaped homicide investigations for nearly two decades.

The method traces back to Detective Tracy Harpster's 2006 master's thesis analyzing 100 homicide 911 calls, which identified 16 factors. By the 2017 book version, the sample had grown to 200 calls and the proprietary COPS scale to 38 guilt indicators across six categories. Those indicators include things like no immediate plea for help, "Defendant Mentality," and even saying "Huh?" or "I don't know." A caller in Michigan was flagged as a suspect for saying "I need help."

Here's what makes it dangerous: Harpster has never testified as an expert in a criminal trial, and courts have repeatedly rejected the method under Daubert and Frye. So instead of defending the science, prosecutors embed the analysis into lay testimony. As a law review analysis puts it, the work has been "mixed into in-court statements and lay testimonies to sneak it past the standards for expert testimony."

The footprint is bigger than most defense bars realize. ProPublica documented the method in 100+ cases across 26 states, and Harpster consulted on at least 2,500 homicide investigations as of March 2024. The real number is likely higher: of roughly 50 agencies canvassed, only 11 responded to questions about their personnel's involvement.

• North Carolina's Office of Indigent Defense Services issued a statewide warning telling attorneys to watch for the technique.
• Fair and Just Prosecution called on member prosecutors to review past cases where it was used.
• The Supreme Court of Illinois agreed to review Jessica Logan's conviction after ProPublica's reporting on the Harpster-trained detective who testified against her.
• Prosecutors, judges, and defense attorneys nationwide are calling for bans and conviction reviews.

Don't assume this is a historical problem. Orange County, California prosecutors charged a woman with a 26-year-old murder after consulting Harpster; the lead detective credited him with identifying "indicators of guilt" and advising on presenting the calls to a jury. That means screening for call analysis belongs in your intake process for both new cases and cold ones.

The practical starting point is the same in either situation: pull the 911 audio, get a reliable transcript, and compare it against the police report line by line. That's exactly the workflow Evidence Copilot is built for, with timestamped, diarized transcripts and side-by-side report-versus-call comparisons, and it extends to related call evidence like jail call transcription. For a deeper look at the review process itself, see our guide to 911 call transcript analysis.

Why 911 Call Analysis Fails Daubert and Frye

A 911 call can put a defendant in prison before a jury ever hears a witness. That's the leverage prosecutors have learned to exploit with so-called "911 call analysis," and it's exactly why defense teams need to know how flimsy the method is when it walks into court.

The technique was built by Detective Tracy Harpster, a Moraine, Ohio police lieutenant, starting with a 2006 master's thesis on 100 homicide calls that identified 16 factors. By the time he and Dr. Susan Adams published their 2017 book, the proprietary COPS scale had grown to 38 guilt and innocence indicators across six categories. The underlying data, how those calls were obtained, and how the research was conducted have never been released, a University of Cincinnati Law Review analysis notes, contrary to standard research practice. Independent researchers who tried to corroborate the claims have failed.

Defense practitioners can build a Daubert challenge prong by prong. The five-factor test gives you a clean template:

• Testability: No published study shows the COPS scale reliably distinguishes guilty callers from innocent ones. Replication has failed.
• Peer review and publication: The 2009 paper appeared in an exploratory venue; the book is not peer-reviewed; the raw data remains sealed.
• Known or potential error rate: Never established, because the underlying dataset has never been released.
• Standards and controls: No published standards govern how an analyst applies the 38 indicators, which makes cross-analyst reliability untestable.
• General acceptance: The NACDL has characterized the technique as forensic junk, and a University of Michigan policy brief finds it "lacks robust empirical support for its scientific validity."

Frye is even simpler. The relevant community of forensic scientists and statisticians has not accepted 911 call analysis as valid. A 2009 Alabama court rejected Harpster as an expert witness, and a Nevada judge rejected a Harpster-trained detective in the same year.

This is the fact most prosecutors hope defense counsel never asks about. Harpster has consulted on at least 2,500 homicide investigations, yet he has never testified as an expert in a criminal trial, according to the Cincinnati Law Review. If the State tries to substitute a "trained detective" rather than Harpster himself, the same foundation problems travel with the testimony: the detective is parroting an untested scale with undisclosed data.

Because courts keep rejecting the method as expert evidence, Harpster advises prosecutors to "guide juries through both the audio and transcript of the calls" and embed analysis into lay testimony and in-court statements to bypass Daubert and Frye. The law review author puts it plainly: the work "has been mixed into in-court statements and lay testimonies to sneak it past the standards for expert testimony."

That disguise is detectable if you know where to look. Pull the 911 audio and transcript alongside the police report and the detective's interview notes. If a detective testifies about a caller's "defendant mentality," a "too polite" greeting, or a missing plea for help, that's expert opinion dressed as lay observation. Object on that basis and demand the underlying data, which the State cannot produce because it has never existed outside Harpster's files.

For teams building the suppression or motion in limine, getting the call evidence transcribed and citable is the first step. A grounded review workspace lets you upload the 911 audio and police report, surface report-versus-call inconsistencies with timestamps, and walk into the hearing with citations to the exact moment in the recording. If you're handling related call evidence in the same case, jail call transcription follows the same workflow, and 911 call transcript analysis is built for exactly this comparison. Book a 20-minute walkthrough at evidencecopilot.com/book to see how the citation chain holds up under cross-examination.

Spotting the Lay-Testimony Disguise

When a prosecutor can't get 911 call analysis past a Daubert hearing, the workaround is simpler than you'd think: put it in a detective's mouth as lay testimony. Harpster himself advises prosecutors to "guide juries through both the audio and transcript of the calls," embedding the analysis in in-court statements to bypass expert-evidence standards entirely, according to a University of Cincinnati Law Review analysis. One law review author put it bluntly: the method has been "mixed into in-court statements and lay testimonies to sneak it past the standards for expert testimony."

The tell is framing. A detective never says "the COPS scale says this caller is guilty." Instead, the testimony walks the jury through the call's wording and lets the conclusion land on its own. ProPublica's investigation found prosecutors knowingly disguised the analysis in trials even after courts rejected it as scientific evidence.

Pattern-match for these documented examples:

• The "too polite" hi: a Washington mother's 2018 call was flagged because her first word, "hi," struck a detective as too polite. Three pathologists couldn't determine homicide, yet the death was reclassified after call analysis, and she took an Alford plea to manslaughter.
• The "I need help" flag: a Michigan caller drew suspicion for saying "I need help," one of the scale's guilt indicators.
• The Fremont handout confession: in 2012, Det. Jason Kiddey used a 911 call analysis handout to elicit a confession. The man pled to manslaughter and got 11 years.
• The "Huh?" and "I don't know" markers: the COPS scale's 38 indicators include no immediate plea for help and what it calls "Defendant Mentality."

Once a detective narrates indicators from the stand, the jury has heard expert opinion without the foundation. Object on relevance and foundation the moment testimony shifts from what the caller said to what the caller's wording supposedly means. Demand the underlying data: Harpster and co-creator Dr. Susan Adams have never released how the calls were selected, how the research was run, or the data itself. Note for the record that Harpster has never testified as an expert in a criminal trial.

The defense workflow here is transcription work. Pull the 911 audio, get a timestamped transcript, and set it beside the police report and the detective's preliminary-hearing testimony. In 911 call transcript analysis, that means diarized transcripts you can cite to the exact second, with report-versus-call inconsistencies surfaced as review signals before the state's narrative hardens. Every AI output is marked for human review, and the workspace never makes legal conclusions; you do. For cases with related call evidence, the same setup handles jail call transcription under one case file.

The technique is still live, not historical: Orange County prosecutors charged a woman with a 26-year-old murder after consulting Harpster, and the lead detective credited him with identifying "indicators of guilt." Screen new cases and cold files both.

Book a 20-minute walkthrough at evidencecopilot.com/book, or start on the Solo plan at evidencecopilot.com/signup. Published pricing, no sales wall, and a 7-day full-access trial with no card required.

Your 2026 Screening and Challenge Workflow

Start every 911 call review the same way: pull the audio, pull the transcript, pull the police report, and put them side by side. The gap between what a caller actually said and what a report claims was said is where most challenges live, and in 2026 it's also where the fight against 911 call analysis begins.

The screening itself is straightforward. Request the 911 audio and any transcript the state intends to use, then compare each claimed "indicator" against the recording with timestamps. Harpster's COPS scale now runs 38 indicators across six categories, including things as thin as saying "Huh?" or "I don't know," so you need to know the exact moment each flagged phrase occurs and what surrounds it. If the state's transcript doesn't match the audio, that discrepancy is your first exhibit. When you're working through a stack of call evidence, a grounded workflow for 911 call transcript analysis keeps every observation pinned to the exact second of audio rather than a paraphrase.

Then screen your cold cases and post-conviction files. ProPublica documented the method in at least 100 cases across 26 states, and it's still surfacing in new prosecutions: Orange County, CA charged a woman with a 26-year-old murder after consulting Harpster. Check whether any detective, dispatcher, or expert in your file trained with Harpster, referenced his book, or testified about the caller's "word choice" or "priority of the victim." The Jessica Logan case is the live anchor here: the Supreme Court of Illinois agreed to review her conviction after a Harpster-trained detective testified about her 911 call, and the Exoneration Project and Center for Integrity in Forensic Sciences stepped in to represent her.

Use the institutional resources already on your side:

• North Carolina's Office of Indigent Defense Services issued a statewide warning telling attorneys to watch for 911 call analysis in their cases.
• Fair and Just Prosecution called on member prosecutors to review past cases, and is partnering with the Innocence Project on the issue.
• The University of Michigan Ford School policy brief documents that these techniques lack robust empirical support and can reinforce investigative confirmation bias.
• NACDL has published its own position on the technique as another example of forensic junk science.

Watch specifically for the lay-testimony disguise. Because courts rejected the method as expert evidence, Harpster advised prosecutors to "guide juries through both the audio and transcript of the calls" and embed the analysis in lay testimony. When a detective starts narrating what a caller "should have said," object: that's expert opinion wearing a lay witness's clothes.

Whatever you find, cite the moment, not the memory. Tools that produce timestamped, diarized transcripts with citations to the exact source, like those used in jail call transcription and call-audio review generally, make it possible to quote the caller verbatim and hold the state to its own recording. Every AI-generated observation stays marked for human review; the judgment call about what to do with it stays yours.

Post-Conviction and Institutional Momentum Defenders Can Use

The reckoning is already here, and it is moving faster than most defenders realize. North Carolina's Office of Indigent Defense Services issued a statewide warning telling attorneys to watch for 911 call analysis in their cases, while Fair and Just Prosecution called on its member prosecutors to review past convictions tainted by the method, presenting ProPublica's reporting at a prosecutors' summit in Austin and partnering with the Innocence Project on the issue, according to ProPublica's investigation.

That pressure creates openings you can use right now. A University of Michigan Ford School policy brief found that the technique lacks robust empirical support, reinforces investigative confirmation bias, and has contributed to wrongful convictions. The Supreme Court of Illinois agreed to review Jessica Logan's conviction after ProPublica's October story, with the Exoneration Project and the Center for Integrity in Forensic Sciences offering representation.

Where the pressure points sit:

• Conviction review: cite the Fair and Just Prosecution call and the Innocence Project partnership in post-conviction motions. Prosecutors facing sanction threats have reason to concede error rather than litigate it.
• Cold cases: the method stays in active use. Orange County prosecutors charged a woman with a 26-year-old murder after consulting Harpster, so screen old files too.
• New prosecutions: North Carolina's warning gives you a citable institutional flag to raise at preliminary hearing.
• Plea negotiations: an Ohio judge told ProPublica she "never anticipated that prosecutors, officers of the court, would engage in systematic organized frauds." That quote belongs in your correspondence.

Documented cases give you pattern-matching material: a Michigan caller flagged as a suspect for saying "I need help," a Washington mother whose "hi" was deemed "too polite" before she took an Alford plea to manslaughter, and 16-year-old Riley Spitler, whose murder conviction was overturned after admitted call-analysis testimony. The University of Cincinnati Law Review notes that agencies themselves undercount the footprint: of roughly 50 departments canvassed over a decade, only 11 responded, and some, like the Illinois State Police, learned of their own personnel's involvement only when ProPublica provided documentation.

The practical spine is comparison. Pull the 911 audio, the transcript, and the police report, then line them up side by side with citations to the exact moment. Evidence Copilot was built for that workflow: upload the call audio and reports, get timestamped diarized transcripts, and surface report-versus-call inconsistencies as review signals before anyone asks. Every AI answer cites the exact source moment, output stays marked for human review, and the tool never draws legal conclusions. For teams handling related call evidence, the same workspace covers jail call transcription, and for the call evidence itself, 911 call transcript analysis is the dedicated service page. Book a 20-minute walkthrough at evidencecopilot.com/book, or start on the Solo plan at evidencecopilot.com/signup: published pricing, no sales wall, and a 7-day full-access trial with no card required.