Can a Machine Cross Like Pozner? How DUI Defense Lawyers Can Partner with GenAI to Master the Art of Impeachment
Leveraging Generative AI to Generate Devastating Chapters of Cross - As If Larry Pozner Himself Were Sitting Second Chair
The Pozner Blueprint - Impeachment by Failure to Follow Training
Larry Pozner’s method of cross-examination is grounded in structured, repeatable frameworks that shape the narrative in favor of the defense. At its core are chapter bundles, tightly organized sequences of leading questions designed to control the witness and advance the defense story, one fact at a time, with one fact per question.
The purpose of this method is not merely to elicit information but to narrate the defense’s case through the opposing witness.
In his June 2025 NACDL Champion article, Pozner identifies impeachment by violation of training as the most potent of these cross-examination chapter structures:
“Impeachment by violation of training is a powerful technique because it simultaneously attacks the professionalism of a witness while undermining a portion of the testimony of that witness.”
The underlying premise is simple but powerful. When a witness is tasked with any investigative role, be it collecting evidence, performing tests, or interviewing suspects, that task is governed by specialized training.
Once a witness acknowledges that they were trained for the task, any failure to adhere to that training calls their professionalism into question and diminishes the credibility of their testimony. This approach reframes deviations from protocol as choices, not mistakes.
Pozner teaches that this impeachment method follows a four-part structure:
Establish the witness’s role in the case.
Commit the witness to specialized training for that role.
Establish the specific procedure the witness was taught.
Prove the violation of that training.
Because the method is both formulaic and adaptable, it can be applied across a wide range of witnesses and fact patterns, particularly in cases where the witness’s investigative actions are critical to the prosecution’s theory.
From police officers to lab technicians, any witness who gathered evidence or performed testing can be impeached for failing to follow their own training.
This structured method aligns seamlessly with the way generative AI learns, allowing models to replicate and extend Pozner-style cross-examinations across diverse fact patterns with remarkable precision.
By codifying the four-part framework of impeachment by violation of training, GenAI systems can help defense attorneys quickly identify deviations from protocol and generate targeted cross chapters designed to highlight those failures in court.
Why This Form of Cross-Examination Is Perfect for DUI Defense
Nearly all police officers investigating DUIs have their training based on the National Highway Traffic Safety Administration's Standardized Field Sobriety Test program.
Those who operate and maintain breath test instruments have similarly standardized training. If blood was tested the test was performed based on the lab's SOP.
This training, when followed, has the potential to produce reliable outcomes. But when it's not, the failure becomes a liability. Once the witness has committed to their training, any deviation becomes a credibility failure.
A trooper fails to hold the stimulus slightly above eye level.
A technician skimps the observation period before a breath test.
A lab analyst deviates from SOPs when preparing blood samples for testing.
This form of cross is also scalable. From the initial roadside stop to toxicology reporting, each phase offers procedural applications where training can be established, specific steps confirmed, and violations exposed.
For the defense lawyer who understands the structure, and now has GenAI to replicate it, each witness becomes an opportunity to shift the create reasonable doubt.
Why the Pozner Method Is Made for Large Language Models
When Larry Pozner distilled cross-examination into replicable chapter bundles with a series of discrete, immutable rules, he was, in effect, building an algorithm.
His four-step structure functions like a legal recipe: follow the sequence, control the facts, and shape the narrative. This predictable, pattern-based approach mirrors the logic large language models are designed to learn, replicate, and execute.
The four-part logic of impeachment by violation of training acts as a framework for GenAI to populate with case-specific facts and language, producing cross-examination chapters that are rigorously structured.
This consistency allows GenAI systems to identify the underlying pattern, replicate its architecture, and adapt its components to the facts of a given case. In doing so, the GenAI acts as both a creative writer as well as a disciplined assistant.
Using the information provided or reviewed, GenAI applies the formula, then fills in each element with precision, while always maintaining the integrity of Pozner’s method. From there, the application can be scaled to varied fact patterns and witness types.
As Pozner writes:
"Defense lawyers can accomplish several common impeachment techniques through formulaic chapter bundles."
Once the formula is understood, it becomes easier to identify, within discovery and investigation, the facts that populate each step. Those facts are then converted into distinct chapters of impeachment. Formulaic structures, by their nature, lend themselves to codification.
Using Persona Prompts to Channel Pozner
One of the most effective techniques for leveraging GenAI in this context is persona prompting, a strategy that instructs the GenAI to assume the voice, mindset, or strategic reasoning of a particular individual.
As discussed in my article, AI-Powered Advocacy: Transforming Criminal Defense Through Prompt Engineering, NACDL/The Champion (Jan./Feb. 2025), persona prompts help generate more targeted, context-aware results because they focus the model’s outputs through a legal and rhetorical lens that is already calibrated for impact. When defense attorneys assign the GenAI a specific persona, the resulting content is both responsive and strategic.
By framing the GenAI's task in the role of Larry Pozner and instructing it to apply the four-part chapter structure, you can generate targeted cross-examinations that reveal procedural lapses, expose training failures, and undermine prosecutorial credibility.
Use role prompting to instruct the AI how to think:
Prompt:
“You are Larry Pozner preparing a cross-examination of a police officer in a DUI case. Focus on the officer’s failure to offer a woman wearing high heels the option to remove them before the walk-and-turn test. Using the Pozner method, construct a four-part impeachment chapter bundle: (1) confirm the officer’s role in conducting field sobriety tests; (2) establish that the officer was trained under the NHTSA manual; (3) elicit that this training required offering removal of heels over two inches before administering balance tests; and (4) show the officer observed the heels but failed to follow this training. Use leading, matter-of-fact questions to expose the violation and discredit both the officer’s professionalism and the reliability of the test.”
You can also use personal prompting to guide the GenAI to internalize your litigation style:
Prompt:
“You are assisting me, a seasoned DUI defense attorney, in preparing cross-examination for trial. Draft a four-part Pozner-style impeachment chapter bundle targeting a lab technician who failed to properly pipette the internal standard while preparing blood samples for gas chromatography. Begin by confirming the technician’s role in handling and analyzing DUI blood samples. Establish that this role required formal training in sample preparation and strict adherence to validated lab protocols. Elicit that the technician was specifically trained to pipette precise volumes of the internal standard, a step critical to ensuring accurate and reliable chromatographic analysis. Finally, prove that the technician deviated from this training by mishandling the pipetting process. Structure your questions in my voice: clear, relentless, leading, and neutral in tone, never sarcastic or editorializing. The goal is to undermine the validity of the test result and the technician’s reliability through a focused, disciplined impeachment by violation of training.
Or try targeted task prompting, which breaks complex outputs into modular parts:
Prompt 1:
“Extract from this police report all potential training-based violations.”Prompt 2:
“For each, write a four-part cross chapter in the style of Larry Pozner.”Prompt 3:
“Simulate officer responses—first cooperative, then evasive.”
DUI-Specific Use Cases - From Discovery to Cross
Let’s walk through how you can build a Pozner-style impeachment strategy using GenAI at each trial stage.
1. Discovery Review
Create a RAG but uploading the police report, the most recent SFST training manual, and video transcript. Be sure all identifying information has been removed to protect client confidentiality. Prompt GenAI with:
“Act as a DUI defense attorney reviewing the attached documents for cross-examination preparation. Analyze the documents to identify any factual elements that suggest the officer failed to follow standardized training protocols, particularly those outlined in the NHTSA manual or relevant departmental procedures. Focus on deviations from required practices in traffic stops, field sobriety testing, suspect interviews, chemical test rights, or evidence collection or handling. Highlight each potential training violation clearly and concisely, and explain its significance to undermining the officer’s credibility or the reliability of their findings.”
GenAI will return a list like:
Failure to use standard instructions on Walk and Turn
Failure to inquire about medical issues
Incomplete documentation of breath test calibration
2. Chapter Building
Prompt:
"Act as a DUI defense attorney. Write a four-part cross-examination chapter bundle using Larry Pozner’s method. The subject is a police officer who failed to properly instruct a DUI suspect during administration of Standardized Field Sobriety Tests (SFSTs). Structure your response as four chapters."
3. Drafting Leading Questions
Prompt:
“Act as a DUI defense attorney preparing cross-examination. Write a four-part chapter bundle using Larry Pozner’s method, targeting a police officer who failed to properly instruct a suspect on Standardized Field Sobriety Tests (SFSTs). Structure the bundle into four chapters: (1) establish the officer’s role in administering SFSTs; (2) confirm formal training in NHTSA-standard procedures; (3) prove specific training on the standardized verbal instructions for each SFST; (4) establish the officer’s deviation from those instructions in this case.
4. Simulation and Rehearsal: Dynamic Prompting for Adversarial Practice
This is where generative AI becomes not just a strategist, but a sparring partner. Simulation prompting allows you to rehearse cross-examinations by casting the GenAI in the role of a witness, often an uncooperative one.
Unlike static output, this method trains your courtroom reflexes in real time. It forces you to maintain control, redirect evasive answers, and remain anchored to your chapter structure under pressure.
By instructing the GenAI to simulate a hostile or hedging witness, you can practice delivering questions with precision, navigating evasions, and calibrating your tone. It's courtroom prep, without the courtroom.
Example Prompt:
“Assume the role of Officer Smith, a police officer testifying in a DUI case. Respond to my questions with evasive or nonresponsive answers. Your objective is to avoid direct admissions. I will cross-examine you as part of impeachment for violating SFST protocol. Do not break character.”
“Act as Officer Smith, a trained DUI enforcement officer. Answer my cross-examination questions evasively, avoiding clear yes/no responses unless forced. I am rehearsing impeachment by violation of SFST training. Your role is to simulate a resistant, hostile witness. Stay in character at all times.”
To deepen the realism, you can enhance these prompts by layering in known information about the officer: prior testimony patterns, documented training history, known areas of deflection, or their rank and role in similar cases.
For example, if the officer has a history of insisting his memory is superior to documentation, or routinely dodges questions on calibration logs, instruct the GenAI to reflect those habits.
The more context you provide, the more convincingly the GenAI will simulate the actual witness. This transforms rehearsal from generic practice into a tailored stress test, training you for the precise dynamics you’re likely to encounter in court.
Augmentation, Not Automation
Concerns about artificial intelligence replacing lawyers often overshadow its most powerful capability: augmenting human expertise through structured, high-efficiency support.
The integration of generative AI into cross-examination preparation does not displace professional judgment, it reinforces it. By automating the tactical execution of repeatable frameworks like Pozner’s chapter bundles, GenAI enables defense attorneys to allocate more cognitive resources to case theory, narrative coherence, and courtroom adaptability.
In this model, the lawyer remains the architect of advocacy. AI functions as a precision instrument, an associate that drafts, tests, refines, and simulates under your direction. It can process discovery at scale, identify recurring procedural errors, and generate tightly tailored impeachment strategies with remarkable consistency.
Rather than diluting the rigor of trial practice, generative AI enhances it. Pozner gave us the blueprint for how to control the narrative through disciplined questioning.
Generative AI provides the processing power to implement that blueprint under compressed timelines and evolving fact patterns.
“Defense lawyers can accomplish several common impeachment techniques through formulaic chapter bundles. Once we understand the formula, we can easily identify… the facts that populate the steps of that impeachment.”
—Pozner, Impeachment by Violation of Training, supra
Prompts, Tools, and Workflows
Here’s a sample toolkit:
Prompt: Cross Chapter Generation
“You are Larry Pozner. Create a cross-examination chapter about a trooper who didn’t offer my client the chance to remove 3-inch heels before SFSTs. Use four chapters: role, training, specific procedure, violation.”
Prompt: Cross Simulation
“Roleplay Officer Martinez avoiding questions about Intoxilyzer breath tube handling. You should deny, hedge, and deflect.”
Prompt: Inconsistency Finder
“Compare the officer’s narrative in this report with the trial transcript. Highlight inconsistencies in SFST administration and observation period.”
Workflow:
Upload discovery (redacted to protect client confidences)
Identify training-linked tasks
Draft chapter bundles
Simulate adversarial exchanges
Rehearse control techniques
Concluding Reflections: The Strategic Convergence of Pozner’s Method and Generative AI
The strategic alignment of Pozner’s structured cross-examination method with generative AI marks a potential advancement in trial advocacy.
Processes that once demanded exhaustive preparation and rote practice can now be systematically modeled, tested, and refined using GenAI’s pattern-recognition capabilities. Each deviation from protocol, each lapse in adherence to training, becomes an engineered point of impeachment, an opportunity for targeted narrative control.
Importantly, this partnership does not supplant legal judgment. As I noted in Rethinking Generative AI in Legal Practice: Toward a Trustworthy Paradigm, NACDL/The Champion (July 2025), GenAI excels when deployed within well-defined frameworks and under the careful direction of the attorney.
GenAI is a drafting partner, a decision-maker, and its outputs should always be carefully reviewed, refined, and adapted to the case strategy. Generative models serve best as accelerators of rigorous analysis, not as substitutes for legal reasoning.
Also discussed in Trustworthy AI Isn’t Accurate, it’s Auditable, the effective and ethical GenAI use requires adherence to the lawyer’s duty of competence. Defense attorneys must ensure that any GenAI-generated cross-examination material is thoroughly vetted, and that all questions are relevant and assist the Defense in telling the jury their alternative story. The questions proposed by GenAI must also be aligned with both evidentiary standards and the client’s best interests.
Always consider GenAI output to be simply a good working draft, a starting point for strategic thought, not an off-the-shelf solution.
When used properly, for criminal defense attorneys confronting the credibility of trained witnesses, the GenAI-augmented application of the Pozner method offers a compelling advantage.
It allows for deeper preparation, more responsive cross-examinations, and sharper identification of procedural weaknesses. By integrating GenAI thoughtfully and critically, defense counsel can elevate both the precision and the power of their trial advocacy.
Cross-examine with intention. Cross-examine with structure. Cross-examine with generative intelligence.
As trial practice continues to evolve, the integration of GenAI with techniques like the Pozner method will not only improve cross-examination outcomes but help democratize elite advocacy tools for solo and small firm lawyers.
Citation:
Larry Pozner, Impeachment by Violation of Training, 49 Champion 12 (June 2025).
About the Author - Learn About the Author’s DUI Defense Practice in Michigan: Patrick T. Barone Esq. | Michigan DUI Lawyers Barone Defense Firm.