What 911 Call Analysis Is and Why It's Still Landing in Your Cases

The call usually arrives first. Before the bodycam, before the report, before the interview, there's the 911 audio sitting at the top of your discovery dump, and if the state has its way, it won't just be evidence. It'll be an accusation.

That's the pitch behind 911 call analysis, a technique built by former police chief Tracy Harpster that claims to detect guilt from a caller's word choice, cadence, and grammar. Harpster's master's thesis used 100 recordings of 911 calls, from which he identified 20 guilt indicators, things like whether the caller asks for help for the victim or requests specific punishment for an attacker. Officers who complete his training, at up to $3,500 for an eight-hour class typically paid with tax dollars, then testify about a caller's culpability as a lay opinion.

The problem is that the science collapsed. Five independent studies, including two FBI Behavioral Analysis Unit studies and research from Villanova and James Madison, failed to corroborate Harpster's guilt indicators, and some indicators performed in the opposite direction of what was previously found. John Jarvis, Harpster's former FBI co-author, told ProPublica he doesn't see where the work "rises to the level of success by the scientific community." Susan Adams, the other co-author, has disavowed any single-indicator use, stating her study examined indicators only in combination.

Prosecutors keep using it anyway. A ProPublica investigation, built on more than 80 open records requests and roughly 120 interviews, documented over 100 cases across 26 states where law enforcement employed the methods. Some prosecutors knew the technique wouldn't be recognized as scientific evidence but used it at trial anyway. One Missouri prosecutor wrote in a 2016 email that the research was not "recognized" as a science in her state, but she was confident a jury would be convinced of the defendant's guilt if it could hear the information.

Here's why this keeps landing in your cases: the technique is structured to avoid the scrutiny that would kill it. As the Cornell Law Review documents, the founders rarely testify. Instead, a 911 operator or patrol officer who took the training offers culpability opinions, letting the method operate in the shadows and largely escape Daubert or Frye challenge. You're not facing a forensic expert you can cross on validation studies. You're facing a cop giving what sounds like common-sense observations about tone and word choice.

That's the core problem for the defense in 2026: junk science disguised as lay opinion. NACDL's president has called the practice "dangerous and insidious," and North Carolina's Office of Indigent Defense Services has issued a statewide warning to defenders to watch for it. When a 911 call shows up in discovery, treat it two ways at once: as an admissibility threat to challenge, and as a comparison opportunity. Discrepancies between the call audio, the officer's report, and the bodycam footage are often where impeachment material lives. A grounded review workspace like Evidence Copilot lets you pull timestamped transcripts of the call and line them up against the report and bodycam, with every AI answer citing the exact moment it came from, so you can verify each claimed "indicator" against what the caller actually said. For jail calls that surface later in the case, the same jail call transcription workflow applies. And if you need a deeper dive on the technique itself, a structured 911 call transcript analysis built for the defense side starts from the premise that the state's indicators are claims to be tested, not facts to accept.

Shadow Forensics: How the Technique Escapes Daubert and How to Stop It

The most dangerous thing about 911 call analysis isn't the science. It's the paperwork, or the lack of it. A 2026 Cornell Law Review article describes how the technique operates as "shadow forensics," largely escaping Daubert or Frye challenge because the underlying methods are never disclosed and the founders are rarely proffered as experts (Cornell Law Review).

Here's how the structure works in practice. Tracy Harpster, the technique's architect, charges up to $3,500 for an eight-hour class, typically paid with tax dollars (Forensic Magazine). Rather than putting Harpster on the stand, prosecutors call a 911 operator or patrol officer who attended that training and can testify about a caller's culpability. The witness has no independent scientific basis, the methodology never gets formally tested, and the technique "operates in the shadows."

ProPublica found that some prosecutors knew 911 call analysis would not be recognized as scientific evidence but still disguised it in trial against unwitting defendants anyway (ProPublica). That's your opening.

Your challenge playbook starts with the witness's training file:

• Demand the training materials, slides, and instructor credentials through discovery and a Daubert hearing. If the witness learned "20 guilt indicators" from a for-profit course rather than validated research, the admissibility argument writes itself.
• Cite the FBI co-author's disavowal. John Jarvis, former co-author of the foundational study, stated: "I don't see where that work rises to the level of success by the scientific community."
• Cite the five independent studies (FBI BAU 2020, FBI BAU missing-child cases, Villanova, and James Madison) that failed to corroborate the guilt indicators, some of which went "in the opposite direction of what was previously found."
• Point to institutional positions: NACDL president Nellie King called the business of 911 call analysis "dangerous and insidious," and North Carolina's Office of Indigent Defense Services issued a statewide warning telling defenders to watch for appearances of the technique.

The co-author problem matters too. Susan Adams has stated that no single indicator can determine innocence or guilt because the original study examined indicators in combination. If the testifying officer is applying indicators piecemeal, that's a cross-examination point before it's an admissibility one.

When you're preparing the challenge, the practical work is comparing the call audio against the officer's report and the witness's claimed indicators, which is exactly the kind of cross-evidence comparison a grounded 911 call transcript analysis workflow is built for. Every claimed "guilt indicator" in the officer's testimony should be traceable to a specific second of the call, and every discrepancy between that testimony and the recording should be quotable with a citation.

The shadow only works if nobody asks what's underneath it. Ask early, in writing, and on the record.

Turning the 911 Call Itself Into Impeachment Material

A 911 call is rarely just one piece of evidence. It's a control document you can hold every other piece of the State's case against. The same audio that prosecutors mine for guilt indicators is often the fastest route to impeachment, suppression, and reasonable doubt.

Start by building a timestamped, diarized transcript of the call, then line it up against the police report, bodycam footage, and any later complainant statements. Where the call audio and the report diverge, you've found a review signal worth running down. A 911 call transcript analysis workflow gives you the exact second of each discrepancy, so you can quote both sides instead of paraphrasing from memory.

The comparison itself is where cases turn. Work through it systematically:

• Did the caller report a fact the responding officer later wrote down differently?
• Does the bodycam capture words, timing, or a sequence the report leaves out?
• Did the complainant's later statement shift on location, timing, or who did what?
• Does the call's timeline undercut a stop-duration, consent, or search-scope argument?
• Was the call itself analyzed or characterized by an officer who attended a call-analysis training?

That last question matters because the technique behind much of this testimony has been formally disavowed. Five independent studies failed to corroborate the guilt indicators, and the University of Michigan Ford School's policy brief found the method "lacks robust empirical support." ProPublica documented more than 100 cases across 26 states where the methods were used, and NACDL has called the practice "dangerous and insidious." If an officer or operator testifies about a caller's culpability after a training rather than as the technique's architect, that's a disclosure and admissibility fight, not a fact.

None of this works if the transcript is wrong. Verify every AI output against the source audio before you cite it. Axon's own documentation states its Miranda Warning Detector "is not 100% accurate," and Sonix's FAQ describes its transcripts as "working drafts, not certified court records." Treat any automated transcript the same way you'd treat a rough draft from a clerk: useful, unverified, and never quoted in a filing until you've heard the words yourself.

The same discipline applies to jail call transcription, where call audio often surfaces months after the 911 call and contradicts what a witness told officers on scene. Evidence Copilot's workspace timestamps and diarizes each call, then flags candidate inconsistencies against earlier evidence as unreviewed signals, with every AI answer citing the exact moment it came from and marked for human review.

If you want to see how that comparison workflow runs on a real discovery dump, book a 20-minute walkthrough at evidencecopilot.com/book, or start on the Solo plan at $39/user/mo with 15 media-processing hours included.

A Working Review Workflow for 911 Calls, From Intake to Cross-Exam

The discovery dump hits your inbox with a mix of audio files, PDFs, and video. You need to process this 911 call analysis quickly to find the gaps before the state locks in their narrative.

Start by uploading the raw files. The system generates diarized transcripts with word-level timestamps, letting you jump to exact moments in the audio. This replaces the tedious scrubbing of play buttons and saves hours per case, as noted by Berkeley’s Criminal Law and Justice Center.

Next, look for the auto-signals. These flags highlight discrepancies between the police report and the actual call audio. If the officer’s report claims the caller was "calm" but the transcript shows a frantic cadence, that inconsistency surfaces immediately for your review.

Here is the core workflow for spotting impeachment material:

• Compare the 911 transcript against the police report for factual contradictions.
• Check if the caller’s statement matches the later jail call or interview footage.
• Identify any "shadow forensics" claims made by officers in the discovery.

When you find a discrepancy, the platform can draft cross-exam questions tied to those specific quoted moments. Every suggestion is marked as AI-generated and pending human review. You remain the final authority on what goes into your motion or trial prep.

Be prepared for the state to proffer call-analysis testimony. A Cornell Law Review article notes that these techniques often escape Daubert scrutiny because the underlying methods are rarely disclosed. Your prepared inconsistencies help you challenge the admissibility of such testimony effectively.

For deeper dives into audio evidence, consider 911 call transcript analysis services that integrate with your existing case management. Similarly, jail call transcription helps track consistency between the initial call and later statements.

Every output cites the exact source. You verify the audio, confirm the context, and build your argument. The tool organizes the evidence; you make the legal judgment. This workflow ensures no detail is missed while keeping your review process grounded and efficient.