Why the Transcript Is Not the Interrogation

A transcript reads like a clean, orderly Q&A. The recording almost never does. What gets typed up is a selective coproduction, shaped by whoever did the typing and what the case needed to look like.

Martha Komter's research on Dutch police interrogations is blunt about this. Police transcripts, she writes, "orient to their relevance for building a case," and the choices made in transcription reflect the transcriber's purpose and anticipated audience. In one theft case she studied, the written record jumps from a denial to a confession with nothing in between: between "I bit her because she was pulling at me" and "I have not told the whole truth" sits half an hour of interaction that is not written down. No record of what the interrogator actually did to get there. The monologue style of the report, Komter notes, "transforms the interaction into a seemingly volunteered narrative by the suspect."

That gap matters because of what Saul Kassin's research shows about final statements. Judges, juries, and lay people, he found, cannot tell the difference between a true confession and a false one just by watching the confession. In a prison study outside Boston, observers couldn't distinguish taped confessions to real crimes from fabricated ones. And roughly 30 percent of wrongful convictions overturned by DNA testing involved a false confession, according to the Innocence Project.

So when you review a police interrogation transcript, treat the document as a lead, not the evidence. Ask:

• Does a recording exist, and does the transcript match it minute for minute?
• What's missing between the denial and the admission?
• Did the statement use tentative language like "I guess I did it" before firming up?
• How long was the interrogation, and was the suspect a juvenile or otherwise vulnerable?

The average false confessor was interrogated for up to 16 hours before admitting to a crime they didn't commit. If your transcript doesn't show that time, the recording does. That's why 911 call transcript analysis and jail call transcription belong in the same review as the interrogation footage: grounded, timestamped sources you can cite by the second, with every AI output marked for your review.

The False Confession Problem and What the Numbers Show

Nearly 30 percent of wrongful convictions overturned by DNA testing involved a false confession, according to the Innocence Project's data cited by psychologist Saul Kassin. That single figure should change how you read every police interrogation transcript that crosses your desk.

The mechanism behind those numbers is documented. Kassin's research shows that people who falsely confessed were interrogated for up to 16 hours before admitting to crimes they didn't commit, per the Innocence Project. Prolonged interrogation degrades reliability, and the populations most exposed to that risk are children, people with intellectual disabilities, and suspects navigating a language barrier.

Kassin sorts false confessions into three types, and each leaves fingerprints in the transcript:

• Voluntary: offered without police pressure, often by someone seeking attention or protecting another person.
• Compliant: the suspect confesses to escape a stressful situation, then recants once the pressure lifts. Kassin notes that recantations rarely carry the weight of the original confession.
• Internalized: the suspect comes to believe they committed the crime, confessing in tentative, fragmentary language like "I guess I did it" or "I guess I must have done it."

That last category produces the most dangerous transcripts. The script barely changes across cases, only names, dates, and places shift. And once a confession exists, the system tends to trust it: Kassin's prison study found that observers, including judges and juries, could not distinguish true confessions from fabricated ones.

This is why treating a transcript as a neutral record fails your client. Transcript review has to extend to the recording, the omitted stretches of interaction, and the tactics that preceded the statement. For defense teams managing that volume of audio and video, a workspace like Evidence Copilot keeps every quote tied to its source moment, whether you're running a 911 call transcript analysis or indexing jail call transcription alongside the interrogation room footage.

Recording Statutes and How Non-Recording Helps the Defense

A missing recording is rarely a neutral absence. When the interrogation room goes dark, the defense gains a statutory foothold that can reshape the entire suppression argument.

California's recording mandate, codified at Cal. Penal Code § 859.5, took effect January 1, 2017 after more than a decade of vetoed reform efforts. The statute requires electronic recording of custodial interrogations of any adult or minor suspected of murder. It does not, however, create a categorical duty. The legislature carved out eight exceptions, and each one gives the prosecution a potential rebuttal you have to anticipate:

• Recording is not feasible.
• The suspect refuses to speak unless the interrogation is not recorded.
• The interrogation occurred in another jurisdiction under that jurisdiction's law.
• Officers did not reasonably believe the suspect committed murder at the time.
• Recording would jeopardize officer or suspect safety or disclose a confidential informant.
• A device malfunction caused the failure.
• The interaction was routine booking or processing.
• The suspect is in custody for specific charges but the interrogation is unrelated to those offenses.

When none of those exceptions apply, the statute gives you two concrete remedies. First, the failure to record becomes admissible to support a claim that the defendant's statement was involuntary or unreliable. Second, courts must give a cautionary jury instruction directing jurors to view statements from unrecorded interrogations with caution. That mandatory instruction is often the single most powerful piece of language in the case, because it tells the jury, in the judge's own voice, that something is wrong with this evidence.

The preservation requirement matters too. Law enforcement must maintain the original recording, or an exact copy, until conviction is final and every appeal (direct and habeas) is exhausted. A missing or altered file late in the process can support a spoliation argument and, in some jurisdictions, evidentiary sanctions.

This is where the NACDL seventeen-benefit framework becomes a working checklist rather than an academic inventory. Benefits like "determining whether suspect statements were made freely and voluntarily," "permitting innocent suspects to establish improper interrogation tactics," and "determining the amount of time involved in the interview" map directly onto the three litigation grounds NACDL training identifies: Miranda violations, involuntariness, and lack of reliability. When the recording is missing, each of those benefits becomes an argument the defense can make about what the absent recording would have shown.

The practical workflow looks like this: pull the police report and any written transcript, check the case file for a recording, and if none exists, run the eight exceptions against the facts in discovery. Pair that with a careful review of related evidence, including any 911 call transcript analysis or jail call transcription that may contradict the timeline the interrogators claim. Where the statutory mandate was triggered and no exception fits, you have both an evidentiary hook and a jury instruction waiting to be requested.

For defense teams handling heavy discovery dockets, organizing this analysis alongside the rest of the evidence, including bodycam, interrogation footage, and jail calls, in a single workspace with timestamped transcripts and full-text search makes the gap between "we have no recording" and "we can prove why that matters" much shorter to bridge.

A Transcript Review Workflow for Defense Teams

A police interrogation transcript is not the interrogation. It is one written account of one, and the research is blunt about what that means in court. Peer-reviewed work on police transcription describes these documents as coproductions between interrogator and suspect, oriented to case-building rather than objective representation. In one studied theft case, half an hour of interaction between denial and confession never made it into the written record at all. Your review workflow has to start from that premise.

If your jurisdiction produced a recording, the recording is the exhibit. The transcript is a secondary artifact you can challenge line by line. Where no recording exists, the absence is itself an issue you can preserve under statutes like California's Cal. Penal Code § 859.5, which makes unrecorded custodial statements admissible to support involuntariness or unreliability claims and triggers a cautionary jury instruction (NACDL). Build the workflow so the recording drives every annotation, not the other way around.

False-confession researcher Saul Kassin has documented the language pattern that shows up again and again in internalized confessions: tentative, fragmentary phrasing like "I guess I did it" or "I guess I must have done it" before a fuller admission (APA Speaking of Psychology). Pull those moments out of the transcript and timestamp them against the recording. They are not proof of anything by themselves, but they are a signal worth flagging for cross-exam and for any expert you may call under the framework NACDL's training materials lay out (NACDL).

A transcript only makes sense against the rest of the discovery. The minimum cross-check set:

• The police report narrative, compared line by line to the transcript for paraphrases that elide questions, answers, or pauses.
• Bodycam or interview-room footage synced to the same master clock, so you can see what the officer was doing while the transcript shows silence.
• 911 call transcript analysis from the same case, since the call often establishes the timeline the interrogation is later measured against.
• Jail call transcription, which frequently captures the defendant describing the interrogation in their own words shortly afterward.

When something is missing from the transcript, that gap is data. The Dutch research documented a case where the suspect moved from denial to confession with no record of the intervening interrogation (Frontiers in Communication). Your checklist should treat every omitted beat the same way: mark it, measure the gap against the recording's running time, and decide whether the omission is impeachable.

The Innocence Project has helped free more than 250 wrongfully convicted people, and roughly 30% of DNA-overturned wrongful convictions involved a false confession (Innocence Project). Those cases did not fail because defenders lacked a transcript. They failed because nobody read the transcript against the rest of the record. A grounded evidence-review workspace like Evidence Copilot lets you put the interrogation footage, the police report, the 911 audio, and the jail calls on a single timeline with timestamped transcripts and OCR'd documents, so every omission is visible the moment you open the case. Every AI answer carries a citation back to the exact second of video or page of PDF, and every output is marked for human review.

For a walkthrough of how this looks on a real case file, book a 20-minute walkthrough.

From Transcript to Motion: Building the Suppression Case

The transcript you reviewed yesterday only becomes leverage when it's pinned to a legal theory. NACDL's interrogation-case training organizes the litigation around four grounds: Miranda violations, involuntariness, lack of reliability, and contamination of the statement by outside details. Each maps onto specific things you already found in the document.

Start with the gaps. Peer-reviewed research on police transcripts found that in one theft case, half an hour of interaction between the suspect's denial and her confession was never written down, and the record gave no information about what the interrogator actually did to persuade her. If your client's transcript shows a similar jump from denial to admission, that omission is your cross-examination spine: "Officer, what happened between those two lines?" An officer who can't answer from the four corners of the document has just conceded the record is incomplete.

Then screen for the reliability markers. Saul Kassin's research documents that false confessions often begin with tentative, fragmentary language like "I guess I did it", and that details matching what interrogators fed the suspect are a contamination red flag. His work also shows judges and juries cannot reliably distinguish true from false confessions by watching the statement alone, which is why the Innocence Project advocates pretrial reliability hearings before confessions are admitted. If your transcript shows fed details, tentative language, or an interrogation lasting hours (falsely confessing suspects were interrogated up to 16 hours on average, per the Innocence Project), a reliability hearing is worth briefing.

• Miranda: quote the exact transcript line where rights were waived, and pair it with the timestamp from the recording.
• Involuntariness: build a duration-and-pressure timeline from the recording, since the written record rarely shows the effort behind the confession.
• Reliability: flag tentative language, fed details, and vulnerable-population factors like youth or language barriers.
• Non-recording: where your state's recording mandate applies and police didn't record, raise the statutory consequence affirmatively.

California gives you a model for the last point: under Cal. Penal Code § 859.5, an unexcused failure to record is admissible to support involuntariness or unreliability claims, and courts must instruct juries to view unrecorded statements with caution. Check the eight statutory exceptions before you file, because each is a potential prosecution rebuttal.

The workflow that ties this together is quote-to-theory: every cross-examination question tied to a cited transcript moment, every motion paragraph anchored to a timestamped excerpt. Tools like Evidence Copilot support that structure by generating cross-exam drafts where each suggested question carries a quoted source, and by drafting a suppression motion outline with both sides quoted and cited, always marked for attorney review. The judgment stays yours; the citations just have to hold up when the officer takes the stand.