The Junk Science Hiding in Plain Sight: How Prosecutors Use 911 Call Analysis

A Missouri detective built a method to spot guilt in 911 calls. Five independent studies have now demolished it. Prosecutors keep using it anyway.

Det. Tracy Harpster's approach claims that callers who commit the crime sound different from callers who didn't. Polite phrases like "please" and "thank you," unusual pauses, and certain word choices are flagged as "guilt indicators". His 2006 master's thesis analyzed 100 homicide calls; a 2017 book expanded the sample to 200. Yet a Forensic Magazine investigation documents that five studies have contradicted his findings, including two from the FBI's own Behavioral Analysis Unit.

The FBI's 2020 study found the indicators were inconsistent, with some running "in the opposite direction of what was previously found." A 2022 FBI study went further, warning that applying the method "may exacerbate bias." FBI behavioral scientist John Jarvis, a co-author, put it plainly: "I don't see where that work rises to the level of success by the scientific community." The method fails all five Daubert factors and the Frye general-acceptance test, according to University of Cincinnati Law Review analysis.

So why does it persist? Prosecutors in at least 26 states have deployed the method in more than 100 documented cases. They bypass expert admissibility hearings by smuggling the analysis in as lay testimony under FRE 701. A dispatch supervisor takes the stand and says the call "sounded staged." No Daubert challenge. No qualification hearing. No disclosure of the underlying methodology.

Russ Faria learned how that loophole works the hard way. A dispatch supervisor testified that his 911 call didn't sound right. Faria was convicted of murdering his wife. The conviction was later overturned. Missouri prosecutor Leah Askey later admitted she got "creative … without calling it 'science'" to avoid the hearings the method could not survive.

You can spot the workaround in your own cases. Watch for these patterns:

• A lay witness (dispatch supervisor, detective, officer) offering opinions on caller tone, phrasing, or "emotional content"
• No pretrial disclosure of an expert witness or methodology
• The prosecution framing the testimony as "based on the witness's training and experience" rather than a validated method
• Specific words or pauses being highlighted as suspicious without any stated scientific basis

The defense counter is straightforward. File a motion in limine challenging any testimony that applies guilt-indicator analysis, whether the witness is labeled expert or lay. Cite the FBI's own contradictory studies and the Daubert failure. Build the record with a precise 911 call transcript analysis that maps every caller statement against the dispatcher's standardized protocol, the same protocol documented at 911.gov. When the caller's "please" and "thank you" turn out to be responses to a dispatcher who asked them to stay on the line, the prosecution's narrative collapses.

Know the Protocol Before You Fight the Narrative: What a 911 Call Actually Contains

Before you can rebut a prosecutor's "gut feeling" reading of a 911 call, you need to know what a call actually contains. Every 911 call follows a standardized structure, and that structure gives you an objective baseline for 911 call analysis that no subjective interpretation can override.

Per the National 911 Program, call-takers ask a standardized set of questions in a predictable order:

• The caller's location and a callback phone number
• The nature of the emergency
• Details: suspect descriptions, injuries, what the caller can see
• Real-time instructions, such as CPR or first aid, while the caller stays on the line until told to hang up

That protocol matters because the prosecution's favorite "guilt indicators" map directly onto normal behavior within it. Det. Tracy Harpster's method flags saying "please" or "thank you" as signs of guilt, along with tone and pauses. But per the ProPublica investigation published in Forensic Magazine, five independent studies, including two from the FBI's own Behavioral Analysis Unit, have contradicted Harpster's findings. The FBI's 2020 study found the indicators were inconsistent, with some going "in the opposite direction of what was previously found," and its 2022 study warned the method "may exacerbate bias."

So when a dispatcher asks a scripted question and your client answers politely, follows instructions, or stays on the line until told to hang up, that's not deception. That's compliance with a national standard. A caller who says "please" is a stressed person being coached through an emergency by a trained call-taker.

The baseline also gives you something prosecutors' lay witnesses can't offer: measurability. Once you have a timestamped transcript, you can show exactly where the dispatcher's scripted questions end and the caller's free narrative begins, and compare each answer against the police report and bodycam. That's the work behind proper 911 call transcript analysis, and it's why we built Evidence Copilot around citations to the exact second of audio rather than impressions of tone.

When a dispatch supervisor testifies that a call "sounded staged," as one did in the Russ Faria case before that conviction was overturned, your cross-examination writes itself: Which protocol question triggered that behavior? Which study validated that reading? The answer to both is none, and the standardized record proves it.

Your Challenge Playbook: Motions, Objections, and the FBI Reversal

You’ve likely seen it in discovery: a dispatch supervisor testifying that a 911 call "sounded staged." This is the "lay testimony" workaround prosecutors use to bypass rigorous scientific scrutiny. They embed unvalidated 911 call analysis methods into routine witness accounts to avoid Daubert hearings.

The method, developed by Det. Tracy Harpster, fails all five Daubert factors and the Frye general acceptance test. Yet, it continues to influence outcomes in at least 26 states. You must spot when subjective "gut feeling" masquerades as objective fact. The stakes are high, with over 100 documented cases where this flawed analysis played a pivotal role in convictions.

Your first move is a motion in limine. Cite the method’s failure of scientific standards and the FBI’s own contradictory findings. The FBI’s 2020 study found that Harpster’s indicators were inconsistent, with some pointing in the opposite direction of previous claims. By 2022, the agency warned that applying the method "may exacerbate bias." These admissions from the originating agency are powerful tools for your suppression motions.

Consider the testimony of John Jarvis, an FBI behavioral scientist and co-author of the reversal studies. He stated, "I don’t see where that work rises to the level of success by the scientific community." Similarly, Susan Adams, an FBI agent, noted that "No single indicator can be used to determine the likelihood of innocence or guilt." You should weave these quotes directly into your legal briefs to undermine the prosecution’s credibility.

To ground your challenge, you need precise, timestamped transcripts. Subjective interpretations crumble against objective data. A 911 call transcript analysis allows you to compare the caller’s statements against standardized dispatcher protocols. This reveals whether specific words or pauses were actually deviations or just normal stress responses.

• Identify "guilt indicators" like the word "please" or "thank you" in the transcript.
• Cross-reference these moments with the dispatcher’s standardized questions and instructions.
• Flag any inconsistencies between the 911 audio and the police report or bodycam footage.
• Use jail call transcription to align subsequent statements with the initial emergency call.

This systematic approach removes the guesswork. You are no longer relying on a witness’s vague recollection of tone. Instead, you are presenting a timeline where every "indicator" is measured against the standard protocol. The data often shows that the caller was simply following instructions or reacting to shock.

Evidence Copilot helps you build this narrative by surfacing candidate inconsistencies between the 911 call and other evidence automatically. When a report contradicts the audio, the system flags it for your review. You can then draft cross-examination questions tied directly to those specific, citable moments. This ensures your challenge is grounded in verifiable facts, not just legal theory.

The goal is to strip the "science" label from the prosecution’s case. By exposing the method’s lack of validation and its inherent bias, you force the court to look at the raw audio. You give the jury the context they need to see the call for what it is: a stressful emergency response, not a confession.

Cross-Reference Everything: Turning the Call Into Impeachment Material

A 911 call is not a standalone exhibit. It is one audio track in a case that includes a police report, bodycam footage, and jail calls, and the impeachment material lives in the gaps between them. Prosecutors have used Harpster-style "guilt indicators" in at least 26 states, and more than 100 documented cases show the method shaping arrests and convictions, even though it has failed every Daubert factor and the FBI's own 2020 study found the indicators inconsistent. Your job is to build the record that shows what the call actually says, second by second, next to everything else.

Start with a word-level transcript. You need timestamps precise enough to quote a phrase and drop a clip on it, not a summary paragraph. Speaker diarization matters too: you have to know which utterances belong to the caller and which belong to the dispatcher reading from a standardized script. The National 911 Program confirms call-takers ask a fixed set of questions about location, phone number, nature of the emergency, and details, then give instructions until they tell the caller to hang up. That protocol is your baseline. When the prosecution calls a caller's "please" or "thank you" a guilt indicator, you can show the dispatcher prompted the response.

Then cross-reference, in this order:

• Transcribe the 911 call with word-level timestamps and confirm speaker labels.
• Read the police report against the transcript and flag every description that doesn't match the caller's words.
• Pull the bodycam and dashcam audio for the same window and compare what officers said on scene to what the report records.
• Pull the jail calls for statements that contradict the 911 narrative or the report.
• List each inconsistency with both quotes and both timestamps, ready for a motion in limine or cross-exam.

The workflow breaks down when you review each file in isolation. A report that says the caller "calmly described the scene" means something different once you hear the dispatcher's instructions and the caller's stress responses in sequence. Defense teams using 911 call transcript analysis get timestamped, diarized transcripts they can search and quote, and pairing that with jail call transcription puts both audio sources on one synchronized timeline.

Evidence Copilot's auto-signals run this comparison on ingest: every new file is checked against earlier evidence, and candidate inconsistencies between report and bodycam or statement and call audio surface as unreviewed signals before anyone asks. Multi-camera sync links bodycam, dashcam, and interview-room footage to a single master clock, so synced playback and an interleaved transcript show you the exact second where the caller's account and the officer's report diverge. The FBI's 2022 study warned that applying this analysis "may exacerbate bias," which is precisely why your cross-reference has to be grounded in quoted sources and exact timestamps rather than impressions.

From Transcript to Cross-Exam: Building the Record That Protects Your Client

The transcript alone won't win the hearing. What wins is a record where every disputed call moment is pinned to a second of audio and every question you ask traces back to a quoted line. Building that record is a team job, and it follows a repeatable sequence.

Start by isolating dispatcher-instruction segments. The National 911 Program documents that call-takers ask a standardized set of questions (location, phone number, nature of the emergency, suspect description, injuries) and often give real-time instructions. Pull every span where the dispatcher directs the caller. When prosecutors later argue a call "sounded staged," you answer with protocol: the caller said "please" or "thank you" because the call-taker told them to stay on the line and follow steps. A 911 call transcript analysis with word-level timestamps lets you clip those spans and cite them by the second.

Then draft cross-exam questions tied to quoted call moments. This matters because the method prosecutors lean on has been documented in more than 100 cases across 26 states, frequently through lay testimony that never faces a Daubert hearing. Your questions should force the witness to separate observation from interpretation:

• "You testified the call sounded staged. Which specific words on the recording support that?"
• "At 1:14, the dispatcher instructs the caller to stay on the line. Did the caller comply?"
• "Is 'please' on your list of guilt indicators, or is it how people talk under stress?"
• "Are you familiar with the FBI's 2020 finding that these indicators were inconsistent?"

Build case briefs with citations to the exact second, not page numbers. Every claim in the brief should carry a timestamp so the judge can play the moment during argument. If your case also involves recorded jail calls, jail call transcription on the same timeline surfaces statement-versus-call conflicts you'd otherwise miss.

Educate the jury on protocol versus guilt indicators before the prosecution frames it. Under the Daubert and Frye analysis, the method fails all five reliability factors and general acceptance, and the FBI's own 2020 and 2022 studies undercut it. A jury that understands standardized dispatch procedure reads a polite caller as compliant, not guilty.

Close with the grounded-review standard: every AI output gets marked for human review, and originals stay immutable. Evidence Copilot's workspace flags each AI-generated excerpt as pending attorney review and keeps every original file untouched, so the record you build survives a challenge.