What 911 Call Analysis Is and Why It Matters in 2026

A 911 call is often the first piece of evidence in a homicide case. Now prosecutors and investigators in at least 26 states use a technique called 911 call analysis to argue that the caller's own words reveal guilt, and defense attorneys are running into it in discovery whether they asked for it or not.

The technique traces back to Detective Tracy Harpster's 2006 master's thesis, which examined 100 homicide calls (50 guilty callers, 50 innocent) and identified 16 behavioral factors that supposedly signaled caller guilt. The proprietary framework that grew out of it is the COPS Scale, "Considering Offender Probability in Statements," and it scores a call across six categories: who the call is about, what the call is about, how the call is made, passive defenses, resistance to the dispatcher, and additional guilty indicators.

By 2017 the indicator list had grown to 38, based on an expanded sample of 200 calls. Police departments pay up to $3,500, often with public funds, for an eight-hour training course that promises to teach officers how to spot a guilty caller.

What the indicators actually look like is worth knowing, because you'll hear them echoed back to you in a police report or on cross:

• Referring to "bleeding" is scored innocent; referring to "blood" is scored guilty
• Asking whether to touch the body is read as an indication of guilt
• Saying "I need help" is scored 80% innocent
• Saying "hi" to the dispatcher counts as "too polite," a guilty indicator
• Saying "Huh?" or "I don't know" falls under "Defendant Mentality"

The scientific foundation is thin. The original study was explicitly exploratory, Harpster and co-author Susan Adams have never released their data or methodology, and at least eight subsequent studies have failed to replicate the guilt indicators. As one defense attorney puts it, no one has been able to reproduce the findings.

Here's the part that matters most for your case: courts keep rejecting 911 call analysis as expert testimony. Harpster himself was excluded in Alabama in 2009 and has never testified as an expert in a criminal trial. But the technique keeps entering cases anyway through a workaround the University of Cincinnati Law Review calls "sneaking past": prosecutors embed the analysis in lay testimony from trained detectives and dispatchers, or narrate the call audio and transcript to the jury themselves, so the Daubert or Frye gate never opens.

That's why you're seeing it now even though the caselaw is on your side. A detective who sat through an eight-hour course can take the stand and walk a jury through "guilty" phrasing without anyone ever qualifying him as an expert. Preparing to meet that means reviewing the 911 audio against the police report and bodycam with timestamped, citable transcripts, so you can quote the exact second a detective's story diverges from what the caller actually said. 911 call transcript analysis is one way to get that grounded record built before the suppression hearing or cross, and the same approach applies to jail call transcription when the state starts building its narrative out of your client's post-arrest calls.

The stakes are documented. Huwe Burton was 16 when he called 911 after finding his mother murdered; he spent more than 20 years in prison before exoneration in 2019. Robert Roberson has spent over 20 years on Texas death row, with his flat affect, later understood as autism spectrum disorder, read as evidence of guilt. In Michigan, a murder conviction was overturned because 911 call analysis should never have been admitted. The NACDL's May 2026 framing of the technique as forensic junk science tells you where the defense bar has landed.

The 'Sneaking Past' Problem: How Analysis Enters Through Lay Testimony

You won your Daubert motion. The court ruled 911 call analysis inadmissible as expert testimony. And then the detective takes the stand anyway, walks the jury through the call, and tells them exactly which phrases "struck him as odd." The analysis just came in through the back door.

That's the dominant workaround, and it's deliberate. Because courts have repeatedly rejected 911 call analysis as expert testimony, Harpster and prosecutors instead embed the analysis into lay testimony, in-court statements, and guided playback of call audio and transcripts, bypassing Daubert and Frye scrutiny entirely. Prosecutors have acknowledged the method "will not hold up to scientific scrutiny," so, as ProPublica's Brett Murphy reported, "they get creative." Harpster's own advice is that prosecutors can "guide juries through both the audio and transcript of the calls" and create "lasting impressions" without ever qualifying an expert.

The typical pattern looks like this:

• A detective or dispatcher who completed the training (departments pay up to $3,500 for the course) testifies as a lay witness about the caller's "behavior" in the call.
• The witness describes reactions in loaded terms: the caller "never asked for help for the victim," "was too calm," "didn't sound right."
• The prosecutor plays the call audio, pausing to highlight the same phrases, so jurors absorb the guilt inference without anyone naming the methodology.
• Closing argument then strings those observations together as if they were facts the witness simply perceived.

Each of these is objectionable under the Federal Rules of Evidence, and the objection points are specific. Under FRE 701, a lay witness may only offer opinions that are rationally based on perception, helpful to the jury, and not based on scientific, technical, or other specialized knowledge. The moment a trained detective explains that a caller's politeness or flat affect indicates guilt, he's smuggling in specialized knowledge acquired through the Harpster course. That's FRE 702 territory, which means it needs Daubert scrutiny the proponent is trying to avoid. Object on the ground that the "observation" is actually the COPS Scale wearing plain clothes.

Ask the witness directly about the training at sidebar or in a pretrial hearing. If he name-checks 911 call analysis or the training handout, you have your record for a motion in limine excluding any testimony that tracks the method's guilt indicators. The University of Cincinnati Law Review documents how the technique has been "mixed into in-court statements and lay testimonies to sneak it past the standards for expert testimony," and Dr. Jessica Salerno explicitly recommends that defense attorneys learn to identify and challenge exactly these lay-testimony attempts.

Preparation is what makes the objection timely rather than reactive. You need to know the call cold: where the dispatcher led, which words the detective will flag, how the trained narrative diverges from the raw audio. Reviewing the 911 audio alongside the police report and bodycam, with timestamped transcripts and citations into the exact moment, is the practical way to build that record. Tools built for this workflow, like grounded 911 call transcript analysis in a workspace such as Evidence Copilot, let you quote the call second by second when you argue the detective is reciting a checklist, not describing what he heard. The same discipline applies to other recorded statements in the case, including jail call transcription review, because the "sneaking past" playbook rarely stops at one piece of audio.

The Replication Record and Emotion-Bias Research: Your Cross-Examination Payload

The scientific case against 911 call analysis is short and brutal: the whole method rests on an exploratory study of 100 calls, and at least eight subsequent studies have failed to replicate its guilt indicators. No one has been able to replicate Harpster's findings, and Harpster has never testified as an expert in a criminal trial. That combination is your cross-examination payload.

Start with the foundation. The 2006 master's thesis sampled 100 homicide calls (50 guilty, 50 innocent) and identified 16 factors, which grew to 18 indicators in 2009 and 38 by the 2017 book, based on an expanded sample of just 200 calls. As defense-side commentary notes, it's unclear whether the calls were randomly selected, whether analysts knew case outcomes before listening, or whether confounds like anxiety, education, cultural factors, speech impediments, and mental health were controlled. Only English speakers were included.

Then layer in the transparency problem. According to a University of Cincinnati Law Review analysis, Harpster and Dr. Susan Adams have never released their data or explained their methodology, contrary to standard research practice. A technique with no public data, no error rate, and no replication isn't science a jury should weigh.

Ready-made cross sequence for the trained detective or dispatcher:

• "The study behind your training analyzed 100 calls, correct?"
• "You're aware at least eight follow-up studies failed to replicate the guilt indicators?"
• "The authors never published their underlying data, did they?"
• "Harpest... excuse me, Mr. Harpster has never qualified as an expert in a criminal trial, has he?"
• "Your eight-hour course cost the department up to $3,500, often taxpayer-funded?"

Pair that with Dr. Jessica Salerno's emotion-bias research as your affirmative alternative explanation. Her studies found that both civilians and police became significantly more suspicious of callers expressing low emotion, and in follow-up work using real 911 calls with known outcomes, low emotion predicted suspicion regardless of whether the caller was later convicted or exonerated. For officers, even moderate emotion didn't move the needle; callers had to sound highly emotional. In other words, the "guilt indicator" is really a suspicion bias with no diagnostic value. Robert Roberson, later diagnosed with autism spectrum disorder, spent 20+ years on Texas death row partly because his flat affect was read as guilt.

So offer the court a competing story: trauma, shock, autism, and cultural differences all produce flat or "inappropriate" affect in innocent callers. Innocence has no predictable emotional script. Salerno herself recommends that defense attorneys learn to spot and challenge these inferences arriving through lay testimony, which is exactly where courts have pushed back, including Michigan's reversal of Riley Spitler's murder conviction.

Practically, this fight lives in the quotes. You need the exact words of the call, the report, and the officer's testimony lined up side by side. A grounded review workspace like Evidence Copilot gives you timestamped, diarized transcripts with citations into the exact second of audio, so every suggested cross question ties to a quoted source, and every output stays marked for human review. If the case also involves calls from custody, the same workflow covers jail call transcription, keeping every impeachment quote pinned to its source before you ever stand up in court.

Real Cases, Real Stakes: Burton, Roberson, Spitler, and the 'Too Polite' Reclassification

Four documented cases show exactly how far 911 call analysis reaches before a defense attorney ever sees it, and where courts have drawn the line. The outcomes range from a wrongful conviction spanning three decades to a murder conviction tossed on appeal.

Huwe Burton: 30 years for calling 911. Burton was 16 when he called 911 after finding his mother murdered. Investigators read his behavior on that call as suspicious. He was wrongfully convicted and spent more than 20 years in prison before his exoneration in 2019, roughly 30 years after the call.

Robert Roberson: flat affect read as guilt. Roberson spent more than 20 years on Texas death row. After conviction, he was diagnosed with autism spectrum disorder, and his flat affect on the call had been interpreted as evidence of guilt. He narrowly avoided execution three times.

The Washington "too polite" reclassification. A Washington mother's baby died of undetermined causes. Detective Marty Garland's analysis contributed to reclassifying the death as homicide, partly because she said "hi" on the call. Being too polite counts as a guilt indicator in the COPS Scale checklist. She took an Alford plea to manslaughter in 2018.

Riley Spitler: the Michigan reversal. This is the case where a court pushed back. Michigan overturned Spitler's murder conviction because 911 call analysis and other statement analysis should never have been admitted.

What these cases have in common is the entry point. None of them turned on a Daubert hearing where the technique was properly vetted. The analysis shaped the investigation first, then arrived at trial through other doors.

• Death classification changed before charges were filed
• Interrogations shaped by the caller's vocal affect
• Confessions elicited using 911 call analysis handouts (Fremont, Ohio, 2012)
• Lay testimony from trained detectives carrying the inferences

That last entry matters most for your motion practice. When a detective who took an eight-hour training course testifies about what the caller's tone "meant," that's 911 call analysis wearing a lay-witness costume. Object on FRE 701 grounds.

For the record, the technique's creator has never testified as an expert in a criminal trial. Alabama rejected him in 2009. Nevada rejected a trained detective. Michigan reversed Spitler.

When you're prepping these challenges, you need the call audio, the transcript, and the police report side by side with citations to the exact second. That's the workflow behind 911 call transcript analysis and jail call transcription at Evidence Copilot: timestamped, diarized transcripts with every AI answer tied to a clickable source moment, marked pending your review. Published pricing starts at $39/user/mo on the Solo plan. If you want to see it against your own case files, book a 20-minute walkthrough.

Building the Challenge: Reviewing 911 Audio Against the Rest of Discovery

A challenge to 911 call analysis lives or dies on precision. The technique has been used in prosecutors' offices in at least 26 states, and because courts have repeatedly rejected it as expert testimony, it usually enters through lay testimony from a trained detective. To object effectively, you need to quote the exact words the detective is relying on, in context, and set them against everything else in discovery.

That means your review workflow has to treat the 911 audio as a first-class document, not a loose file. Practical steps:

• Get a timestamped, diarized transcript of the call so every phrase ("bleeding" vs. "blood," saying "hi," asking whether to touch the body) is pinned to the exact second it was spoken.
• Line the call up against the police report and bodycam. Report-vs-call discrepancies are where the state's narrative starts to fray.
• Track which "guilt indicators" the detective's report or testimony actually invokes, and match each one to its source moment in the audio.
• Preserve the full call context. A snippet played for the jury often omits the dispatcher's leading questions that produced the "suspicious" answer.

This is where grounded review tools earn their keep. When you upload the discovery dump, 911 call transcript analysis in Evidence Copilot produces timestamped transcripts with citations into the exact second of audio and the exact page of the PDF, so you can quote the call, the report, and the detective's testimony side by side. Every answer carries a clickable citation, and every output is marked AI-generated pending human review, which keeps you inside your ethical obligations. If your case also involves inmate calls, the same workflow extends to jail call transcription.

The payoff shows up in two places. At the motion stage, a Daubert or Frye challenge is easier when you can show the court that the "methodology" rests on an exploratory study of just 100 calls and that at least eight subsequent studies failed to replicate its guilt indicators, per the Innocence Project. At trial, cross-examination of the detective who "noticed" suspicious caller behavior becomes surgical: you play the exact eleven seconds, show the dispatcher's question that prompted the phrase, and ask whether the detective's eight-hour training course covered any of the replication failures.

Precision is also your protection against the sneak-past strategy. When a prosecutor narrates call audio for the jury without calling an expert, your objection has to identify precisely which statement crosses from lay observation into expert opinion. A transcript with word-level timestamps makes that objection quotable rather than vague.

Build the record once, cite it everywhere, and make the state defend every second of audio it wants the jury to hear.