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

911 call analysis is showing up in your case files because prosecutors have spent nearly two decades building it into homicide investigations, and by 2026 it's a technique your clients can be convicted on. The most widely used version, the Harpster/Adams method, claims to detect guilt from a caller's word choices, and it has spread to prosecutors and investigators in at least 26 states, with its creator consulting on more than 2,500 homicide investigations as of March 2024, according to a University of Cincinnati Law Review article that calls for courts to end its use entirely.

The method works like this: an analyst scores a caller's statements against a proprietary tool called the COPS Scale, which sorts dozens of "guilt indicators" into six categories. The indicators are oddly specific. Commenting on the victim's eyes suggests guilt. So does asking whether to touch the body. Saying "bleeding" suggests innocence, but "blood" suggests guilt. In real investigations, analysts have treated a caller's first word, "hi," and the phrase "I need help" as signs of deception, per the same law review analysis.

Here's the problem for your cross-exam prep: the entire method rests on a research base of roughly 100 to 200 calls. Harpster's 2006 thesis used 100 calls from cases that ended in conviction or plea, and his 2017 book expanded the sample to just 200 calls. No one has replicated the findings, Harpster and Adams have never released their data, and the sample excluded non-English speakers while ignoring anxiety, education, culture, and mental health, as defense attorney critiques of the method document.

Why does junk science keep reaching juries? Because prosecutors rarely put a qualified "911 call analyst" on the stand. Instead, they use lay-testimony workarounds:

• Harpster himself has published tips coaching prosecutors to introduce the method without seeking expert qualification, including having officers "guide juries through both the audio and transcript of the calls."
• The creator of the method has never testified as an expert in a criminal trial, so the reliability fight never happens on his watch.
• Officers trained through Harpster's 8-hour commercial course present COPS Scale conclusions as lay opinion under FRE 701, sidestepping the Daubert and FRE 702 standards that would apply to expert testimony.

Courts have started pushing back. Harpster was rejected as an expert at a 2009 Alabama trial, a Nevada judge rejected a detective trained in the method, and Michigan overturned Riley Spitler's murder conviction because 911 call analysis "should not have been admitted." A University of Michigan policy brief goes further, concluding the technique reinforces investigative confirmation bias and has contributed to wrongful convictions.

Because the method turns on exact wording, your first move is verifying the analyst's quotes against the actual audio. Word-level, timestamped review of the call, the kind of grounded 911 call transcript analysis we built Evidence Copilot for, exposes misquotes and context-stripping before the analyst's testimony hardens into fact. The same discipline applies to jail call transcription when the prosecution pairs a client's recorded calls with the 911 audio to build a narrative.

The Junk Science Record: Building Your Daubert and FRE 701 Attacks

The record against 911 call analysis is unusually clean for a defense motion. The method's creator, John Harpster, built the COPS Scale© on a 2006 thesis of just 100 calls and a 2017 book expanding to 200 calls, and neither he nor co-author Adams has ever made the underlying data public, "contrary to standard practice for this sort of research" according to a 2025 University of Cincinnati Law Review article. No one has replicated the original findings.

Map that onto Daubert's factors and the motion nearly writes itself:

• No known error rate: nothing in the published record quantifies accuracy, so factor three fails outright.
• No peer-reviewed validation: a couple of tiny studies of roughly 100 calls each, with opaque call selection and analysts who knew case outcomes, is not testable, transparent science.
• No standards: the "indicators" shift with the author (16 factors in 2006, 18 in 2009, 38 in the 2017 book), and the scale itself is proprietary.
• No widespread acceptance: Harpster has never taken the stand as an expert in a criminal trial. The method's own creator won't submit to Daubert.

Cite the adverse rulings: Harpster was rejected as an expert at a 2009 Alabama trial, a Nevada judge rejected a detective trained in the method, and Michigan's Riley Spitler murder conviction was overturned because the analysis should never have been admitted. A practitioner critique also flags the conflict of interest: Harpster sells an 8-hour training course to the departments whose cases then "validate" his scale.

Expect the lay-testimony workaround. Per the UC Law Review, Harpster has supplied prosecutors ten tips for getting the method in front of juries without expert qualification, essentially coaching them to "guide juries through both the audio and transcript" while a trained detective offers "observations" under FRE 701. That's your motion in limine: any testimony that an opening word like "hi" was "too polite," or that "I need help" signals guilt, applies specialized COPS Scale knowledge dressed as lay opinion. Object before the witness walks the jury through the call; once the indicators land, the prejudice is done.

One cross-exam point worth keeping loaded: peer-reviewed research shows trained professionals detect deception at only about 54%, barely above chance. An analyst claiming to read guilt from word choice is claiming to beat a baseline the science doesn't support.

Because the method turns on exact wording ("bleeding" versus "blood," first words, cadence), verify the analyst's quotes against a word-level, timestamped transcript of the actual audio. Grounded transcript review of 911 call audio, with every quote citable to the exact second, is core workflow for this kind of challenge, and it's how we built 911 call transcript analysis into Evidence Copilot for defense teams handling call evidence alongside jail call transcription.

The Hearsay Trap: Your Client's Own 911 Call Isn't Automatically In

Every defense attorney wants to put their client's 911 call in front of the jury. The instinct is understandable: the client sounded panicked, coherent, and consistent with innocence. But a 2026 Virginia appellate decision confirms what experienced practitioners already suspected. When the defense offers the call, it has to independently clear a hearsay exception, and the bar is higher than most lawyers assume.

In Jones v. Commonwealth (Va. App. Sept. 22, 2026), the court affirmed exclusion of the defendant's own call. Two facts killed it: a gap of at least 10 minutes between the shooting and the call, and three intervening advice-seeking conversations. The excited utterance exception requires a startling event, a statement made under the stress of excitement from that event, and a statement relating to the event, and the test is not mechanical. It turns on whether the declarant was incapable of thoughtful reflection. Ten minutes plus three conversations gave the court plenty of room to find reflection.

The asymmetry is the part that stings. Calls offered by prosecutors typically ride in through the excited utterance exception with little resistance, per practitioner analysis of 911 hearsay doctrine. Your client's call gets a hostile, minute-by-minute scrutiny of every intervening event. Plan accordingly:

• Document the event-to-call timeline precisely: minutes elapsed, who the client spoke with, and what those conversations covered.
• Brief every candidate exception fully (present sense impression, state of mind, residual exceptions), not just excited utterance.
• Make an offer of proof on the call's content so the appellate record shows what exclusion cost you.
• Check the transcript against the audio at the word level before you rely on any quoted phrasing.

That last point on briefing deserves emphasis. In Jones, an isolated citation without developed argument got the issue treated as waived. A thin record is how admissible exculpatory evidence dies quietly.

The practical workflow here is timeline reconstruction. Pull the call audio, the bodycam, and any witness statements, and build a minute-by-minute sequence from the startling event to the moment your client dialed. Word-level, timestamped transcripts are what make that sequence defensible; when the spontaneity fight turns on seconds and intervening conversations, a vague summary loses. Tools like Evidence Copilot generate diarized, timestamped transcripts of 911 audio (see 911 call transcript analysis) and related recordings like jail call transcription, with every answer cited to the exact moment in the source file, so your hearsay brief rests on the record rather than recollection.

Assume the call is out until you've proven otherwise. That assumption changes how you prep the motion, and it changes what you promise the client.

Verifying the Transcript Against the Audio: The Word-Level Workflow

The whole method lives or dies on exact wording. If the analyst says your client's caller said "blood" when the recording actually says "bleeding," that flips a claimed guilt indicator into an innocence indicator, and the difference is one syllable buried in a panicked, overlapping, dispatcher-interrupted call. You cannot catch that from a summary or a prosecution-prepared transcript. You need your own word-level, timestamped transcript checked against the audio.

That's not paranoia. Prosecutors are coached to "guide juries through both the audio and the transcript" without ever qualifying as experts, which is how Harpster-style indicators get in front of juries as lay opinion, per a University of Cincinnati Law Review analysis. When the entire exercise rides on first words, cadence, and phrases like "hi" or "I need help" being treated as guilt signals, a transcript that's even slightly wrong quietly manufactures indicators that don't exist in the recording.

So build the verification into your workflow. Get the call audio in discovery, produce a transcript with word-level timestamps and speaker diarization, and check every quoted indicator line by line. A grounded workspace like Evidence Copilot handles this by generating diarized, word-timestamped transcripts of the 911 audio (and related jail call audio, via jail call transcription), with every AI answer citing the exact second it came from and every output flagged for human review. The attorney still listens and decides; the tool just makes every quote checkable against its source moment.

Here's the checklist I'd run on any prosecution 911 analysis:

• Pull each "indicator" the analyst claims, find it in your transcript, and note its timestamp down to the word.
• Listen to that span of the raw audio yourself. Confirm the exact word, the speaker, and whether the dispatcher interrupted or talked over it.
• Restore the context. An indicator quoted in isolation often sits inside a sentence that means something else entirely, and the analyst may have stripped surrounding lines.
• Check the first words and opening seconds specifically, since callers' first word "hi" has been treated as a guilt indicator in real investigations.
• Document any mismatch in writing, with both the audio timestamp and the analyst's quote side by side, so it's exhibit-ready.

The mismatches matter beyond impeachment. If the analyst's quotes don't match the audio, that goes to reliability under Daubert factors and to FRE 701 in a motion in limine. And even where the quotes are accurate, the underlying premise is shaky: peer-reviewed research puts human deception-detection accuracy, including trained professionals, at roughly 54 percent, barely better than a coin flip. That baseline is a fair, grounded cross-examination question for any witness who claims to read guilt from a caller's word choice.

Do the word-level pass early, before you brief motions. It's tedious, it's the least glamorous part of 911 call analysis, and it's where the case gets won.