Why 911 Call Analysis Keeps Landing in Your Cases (and Why That's Good News)

You probably have a 911 call in your case right now, and someone on the other side already has a theory about what it means. 911 call analysis, the practice of reading guilt from a caller's words and demeanor, has quietly spread into more than 100 cases across 26 states. That is the good news buried in a bad trend: because the state's playbook is documented, you can prepare before the charging decision even lands.

The technique rarely arrives as expert testimony. Christopher Lau, co-director of the Wisconsin Innocence Project, calls it a "shadow forensic": methods stay undisclosed, the founders never take the stand, and instead a 911 operator or officer who sat through an 8-hour commercial course testifies that the caller sounded guilty. That pathway lets the state smuggle expert conclusions past Daubert and Frye without ever qualifying an expert.

The underlying science is thin. Tracy Harpster's method grew from a master's thesis built on 100 homicide calls, later expanded to roughly 200 cases and 38 "guilt indicators" in his 2017 book. No one has replicated his findings, and at least six peer-reviewed studies have failed to reproduce them.

Read the indicator list and the problem becomes obvious. According to a defense practitioner's account of the method, the theory treats these as signs of guilt:

• Commenting on the victim's eyes
• Asking whether to touch the body
• Saying "blood" rather than "bleeding"
• Immediately acknowledging death
• Saying "I need help," or even opening with "hi"

Roughly 80% of callers who said they needed help were classified innocent, so the same phrase cuts both ways depending on who is scoring it. Cornell's Jessica Salerno warns that expecting a single "right" demeanor on a 911 call is dangerous, because culture, neurodivergence, disability, and raw stress all shape how people sound. Robert Roberson's Texas capital case turned partly on his flat affect; he has severe autism.

The stakes show up in the case names you already know. Jessica Logan's conviction is now before the Illinois Supreme Court, and Michigan overturned Spitler because call analysis should never have been admitted. Build your file early: pull the call audio, get a timestamped 911 call transcript analysis, and compare every word against the report and bodycam. When calls span both sources, jail call transcription in the same workspace keeps the record citable. Evidence Copilot gives you grounded citations into the exact second of audio, so you can show the court what the caller said and what the trained officer claims it means.

What the 20/38 'Guilt Indicators' Actually Claim and Where They Come From

The 20/38 "guilt indicators" you'll hear quoted in a suppression hearing didn't arrive from a research lab. They trace to one man's master's thesis, then grew with each retelling. Tracy Harpster's method started in 2006 as a thesis built on 100 homicide calls and 16 indicators, expanded to 18 guilt indicators in a 2009 paper, and reached 38 indicators across the proprietary COPS Scale in a 2017 book covering roughly 200 calls, as the UC Law Review documents.

The training that spreads these claims is an 8-hour commercial course now taught to police departments nationwide, according to a defense-practitioner account of the method. The indicators themselves read like folk wisdom dressed as science. Commenting on the victim's eyes suggests guilt. Asking whether to touch the body suggests guilt. Saying "bleeding" suggests innocence, while saying "blood" suggests guilt. A caller who immediately acknowledges death, or who discusses prior conversations with the victim, gets flagged. So does one who simply says "I need help."

Memorize the specific ones for cross. When an officer testifies that your client's word choice signaled guilt, you can name the exact indicator, its origin, and the sample size behind it:

• "Bleeding" versus "blood": a single word choice treated as a guilt signal, drawn from a thesis of 100 calls.
• Commenting on the victim's eyes, or asking whether to touch the body, as supposed admissions.
• "I need help" or an opening "hi" read as suspiciously polite.
• Immediate acknowledgment of death treated as consciousness of guilt.

The scale is real. ProPublica has documented law enforcement use of Harpster's methods in 100+ cases across 26 states, with the FBI among agencies helping spread it. Meanwhile, at least six peer-reviewed studies have failed to replicate the underlying research, and no one has replicated Harpster's findings at all. When you're building a 911 call transcript analysis, the point isn't to debunk the theory in the abstract. It's to pin down exactly what your caller said, at what second, so you can show the jury the gap between the words and the conclusion an officer hung on them.

How to Attack It: Daubert/FRE 702, Replication Failure, and the Lay-Witness End-Run

The prosecution's 911 call analysis arrives in court dressed as a lay-witness opinion. Your job is to strip the costume off, and there are three reliable ways to do it.

Prong 1: Recharacterize the testimony as expert opinion under FRE 702. When a 911 operator or officer testifies that a caller's phrasing, cadence, or word choice reveals guilt, that's not personal knowledge. It's a conclusion drawn from a proprietary framework, and the framework is exactly what Christopher Lau's Cornell Law Review article calls "shadow forensics." Run the five Daubert factors and the technique collapses on every one of them. The method isn't testable because Harpster and Adams have never released their data or methodology. There's no peer-reviewed validation, no known error rate, no published standards, and the broader forensic community has rejected it. Courts in Alabama (2009) and Nevada already refused to qualify Harpster as an expert, which is the clearest evidence that what the operator is offering is, in substance, expert testimony. Argue the witness is an undisclosed expert and the gatekeeping hearing is mandatory.

Prong 2: Hit the sample size and the replication failure. The underlying research rests on roughly 200 calls across two small studies, and at least six peer-reviewed studies have failed to replicate the findings. That's not a methodological quibble. It's the core of a Daubert attack. A technique that can't be reproduced by independent researchers has no place in a courtroom, no matter how many times it's been used in investigative work. Drive this point home in cross: ask the witness to name a single published replication. There isn't one.

Prong 3: Use the Cornell demeanor science to rebut officer hunches. When the witness describes "suspicious" behavior, you're no longer limited to arguing methodology. You have peer-reviewed research on your side. The Cornell team identified five behaviors that trigger caller suspicion: poor communication, guarded information, impression management, lack of emotion, and lack of urgency. Lead with Professor Jessica Salerno's point that expectations about demeanor on 911 calls are dangerous because people communicate differently based on culture, personality, neurodivergence, disability, or stress. The Robert Roberson case, where severe autism masked emotional expression and contributed to a death sentence now widely viewed as a wrongful conviction, makes the stakes concrete. A grounded 911 call transcript analysis workflow lets you lock down exactly what the caller said and when, so you can cross-reference the officer's "indicators" against the actual audio. When the same case includes jail calls, the same workspace handles jail call transcription so your demeanor rebuttal covers every call the prosecution might surface.

The 2026 Defense Workflow: Screening, Transcripts, and the Cross-Exam Script

You get the discovery dump. Somewhere in the stack sits a 911 call, maybe a jail call, maybe a police report that paraphrases the call. Your job is to figure out whether anyone in the prosecution's chain of custody was trained to "read" the caller's words for guilt, and then pin down exactly what the caller said before anyone else's interpretation gets layered on top.

The first question is who in the agency attended the training. ProPublica's investigation found that departments often didn't know their own personnel had been trained. Illinois State Police learned only from ProPublica's documentation. That means a general discovery request won't catch it.

Send a targeted request asking whether any officer, detective, 911 operator, communications supervisor, or training coordinator attended an 8-hour commercial course on 911 call analysis, used the COPS Scale, consulted with Tracy Harpster or anyone using his methodology, or received any instruction on caller "guilt indicators" or "stress indicators." Ask for the dates, the trainer, and any handouts. Ask the prosecutor's office the same questions under Brady. The North Carolina Office of Indigent Defense Services has already issued a statewide warning to defenders to screen for this, per the same ProPublica reporting.

Once you have the audio, you need a transcript that ties every word to a second on the recording. A grounded 911 call transcript analysis workflow gives you word-level timestamps and speaker diarization, so you can isolate the caller's statements from the dispatcher's prompts. That matters because the prosecution's theory often depends on fragments.

The 20 guilt indicators in the original theory include things like commenting on the victim's eyes, asking whether to touch the body, and saying "bleeding" versus "blood," according to a practitioner breakdown. Pull each of those alleged indicators and check the transcript. Did the caller actually say it, or did the officer paraphrase? Was the question prompted by the dispatcher? Was the caller cut off mid-sentence? Pin the exact timestamp for every quote you intend to use.

When the call evidence spans both 911 and jail calls, the analysis has to span both. A defendant who sounds "flat" on the 911 call and "evasive" on a jail call to a relative is being judged twice on the same demeanor. Run jail call transcription through the same timestamped, diarized pipeline so you can compare statements side by side and challenge the officer's interpretation with the caller's own words in context.

Build the cross-exam around replication failure and sample size. The underlying studies used roughly 200 calls total, according to UC Law Review's analysis. No one has replicated the findings. At least six peer-reviewed studies have failed to replicate, per Cornell research. Harpster has never been proffered as an expert, was rejected in Alabama (2009) and Nevada, and never released his data or methodology.

Sample questions:

• "How many calls were in the underlying study you relied on?"
• "Has anyone replicated those findings in a peer-reviewed journal?"
• "What is the published error rate of this methodology?"
• "Did you account for the caller's neurodivergence, disability, or cultural communication style?"
• "Are you testifying as an expert, or as a lay witness applying training you received?"

Frame the testimony as expert opinion under FRE 702 and argue it fails Daubert's five factors, per the Cornell Law Review's "shadow forensics" framing. Cite the overturned Spitler conviction in Michigan. Cite NACDL's characterization of the technique as "another example of forensic junk." Cite the University of Michigan STPP policy brief on investigative confirmation bias. If you're in Illinois, note that the state Supreme Court agreed to review Jessica Logan's conviction, with representation from the Exoneration Project and Center for Integrity in Forensic Sciences.