The Funnel Technique: Better Deposition Questions, With or Without AI
- Open-ended who, what, where, when, why and how questions uncover what a witness knows; narrow did-you questions hide it.
- Close each topic by asking whether the witness has told you everything, then drill down to dates, documents and admissions.
- Let AI draft the outline and summarize the transcript, but do the live listening and follow-up questioning yourself.
AI can now draft your deposition outline and summarize the transcript afterwards. It can't sit across the table from the witness and hear the answer that doesn't fit. That part was always the job, and one of the oldest courses in the Solo Practice University archive teaches the questioning technique that makes it work.
In 2009, SPU ran a course called The Art of the Deposition. First on its syllabus was a questioning method the course credited to the National Institute for Trial Advocacy (NITA): the funnel.
Here's what the funnel is, why it works, and where AI fits around it.
The first-year mistake: questions that prove you already know
The instructor, Victoria Pynchon, opened the course with a confession. Within the first month of practice, the instructor was sent to take an "easy" deposition: the young man who caused the plaintiff's fall at a skating rink. What followed was a long string of yes-or-no guesses. Did you get there in the evening? No. Afternoon? No. Did you rent your skates? No. And on and on.
The diagnosis:
"I am making my life incredibly difficult because I want everyone to know that I know even though the point of the whole exercise is obviously to learn what I don't know."
That's the instinct the funnel corrects. New litigators, and plenty of experienced ones, ask narrow questions because narrow questions feel like control. They aren't. They hand the witness an easy "no" and leave you ignorant of everything around it.
The course's example of the classic version: you ask whether the witness and the witness's brother took Joan to the country club on the first of June. The answer is "No." What you never learn is that the witness went alone, on the second of June, met someone who made an introduction to someone else, who then drove the witness to meet Joan somewhere else entirely. The "no" was true. It was also useless.
"If you are asking "did you" questions… you'll either get no useful answers or, worse, answers that suggest responsive testimony but aren't."
The funnel: start wide, stay wide, then drill down
The fix is simple to state and hard to practice. Start at the wide top of the funnel with open-ended questions, and only narrow once you've exhausted what the witness knows.
"When you begin a deposition, it's best to know nothing and ask innocent questions about everything."
In practice, that meant the reporter's questions. The course told students that for roughly the first few hours of a deposition, "all your questions should begin with one of the following words: who, what, where, when, why and how," followed by "and then what happened?" and "what happened next?" The course syllabus put it more bluntly: when you take a deposition, you're being a news reporter.
The top of the funnel isn't just a way to open. It's also how you close each topic. Among the questions the course said every examiner should ask, two bracket the funnel: "Tell me everything you know about…" at the start, and "Have you now told me everything you know about…" at the end. That second question matters. It locks the witness into the universe of what they've told you, so a new story at trial looks like what it is.
Only then do you drill down: the specific dates, the documents, the admissions you need for a summary judgment motion. Narrow questions still have their place. They just come after the wide ones, not instead of them.
The funnel, top to bottom
Lean discovery: you don't need all of it
The funnel carried into a later lesson on designee depositions, the depositions of a person an organization produces to testify for it. The instruction for each topic on the notice was to "start and finish at the top of the funnel (open-ended who what where when why questions) for each category before you drill down deeper."
Two other points from that lesson still hold. First, discipline:
"STAY ON COURSE - if you think of a killer question in the first half hour of this deposition, write it down and come back to it"
Second, a provocation worth sitting with in an era when software can ingest every document in a case:
"after 25 years of practice, I'd say everyone could easily prove their case with 10-20% of the information they gather; the other 80-90% is simply a sign of insecurity."
The instructor also told students not to fight with opposing counsel: make the record and move on. (For more on that, see the SPU archive piece on negotiating with belligerent opposing counsel.)
One update, because the lesson is from 2009. Rule 30(b)(6) of the Federal Rules of Civil Procedure, which governs depositions of organizations in federal court, was amended effective December 1, 2020. The serving party and the organization now "must confer in good faith about the matters for examination," before or promptly after the notice or subpoena is served (Rule 30, Cornell LII; U.S. Courts). The course predates that requirement. State rules vary, so check your own jurisdiction's rules.
Where AI fits, and where it doesn't
The work around a deposition has changed a lot since 2009. The course's impeachment lesson pointed to real-time reporting as the technology of the day: "You can use the morning's testimony in the afternoon." SPU's archive tracked the next step too, when lawyers started putting transcripts and exhibits on mobile devices.
Now the tools draft and summarize. Thomson Reuters' CoCounsel, for example, has a "Prepare for a deposition" skill that drafts topics and questions from your description of the matter, and a Summarize skill the company pitches for long transcripts (CoCounsel help; Thomson Reuters).
Used that way, AI is a reasonable assistant at both ends of the deposition. It can get a solo from a blank page to a draft outline quickly, and it can turn a long transcript into something you can search and skim before trial.
But look at what the funnel actually asks of the examiner. The wide question only pays off if you hear the answer, notice the name you didn't expect, and follow it. One of the course's guest lessons, from a medical malpractice lawyer, said it plainly:
"Anyone can read from a list of prepared questions. It takes an experienced attorney to listen to the answers and know where you want to go…"
A generated outline is a list of prepared questions. Read it the way the course would: how many of them are "did you" questions? Rewrite those as who, what, where, when, why and how. Then put the outline down when the witness says something it didn't anticipate.
John O'Connor, founder of Associate and SPU's Professor of AI Fluency, names the risk directly:
"The lethal AI habit for lawyers is cognitive surrender of work product."
A deposition is about as pure an example of work product as a litigator has. The questions, the follow-ups, the decision to stop asking once a witness has been impeached: that is judgment, made live, under oath, on the record. O'Connor's advice is to "use AI to delegate everything around the practice, the business side," and to keep the thinking. AI tools also make mistakes, which SPU covered in its guest lecture on the ethics of using AI. A summary is a starting point. The transcript is the record.
AI at both ends, you in the middle
The litigators who can still read
In November 2026, O'Connor published ten predictions for the legal industry on LinkedIn. The third:
"Litigators who can still read and who don't use AI for work product become the enforcers."
Read that next to the funnel. The lawyer who has actually read the documents knows which open question to ask. The lawyer who listens hears the answer that suggests responsive testimony but isn't. The lawyer who wrote the outline knows when to abandon it. None of that is anti-technology. It's the same point SPU made years ago about emotional intelligence when AI arrives: the human skills become the scarce ones.
The instructor made a related point in a lesson filmed with a practice witness: the examiner has to be authentic, formal or informal, because that's what lets a witness drop their guard and talk. No tool does that for you.
If you're a newer litigator without a senior partner to watch, the funnel is learnable, and the old SPU advice on learning how to be a lawyer under fire still applies. Sit in on depositions. Read transcripts, including your own, and count the "did you" questions.
What else the course covered
The funnel was one unit of many. The Art of the Deposition went on to cover marking and authenticating documents, laying the foundation for the business records exception, impeachment with prior testimony, admonishing the witness, defending a deposition, expert witness outlines, and a series of lessons built on scenes from the film The Verdict. More from the SPU archive on tools and practice is on the Technology category page. Litigators get hired from their websites too, so make sure yours shows how you work, not just where you went to school. SPU offers a free website review for law firms.
Keep the thinking, delegate the rest
SPU membership gets you the community, monthly strategy sessions and the course library, with other solo and small-firm lawyers working out which parts of their practice to hand to AI and which to keep. See what SPU membership includes.
Get one practical lesson for your firm every week.
Free, from practicing lawyers. Every issue includes a lesson from the archive. Unsubscribe anytime.
The SPU Team writes and edits Solo Practice University's articles. Susan Cartier Liebel founded SPU in 2009 to teach lawyers the practical side of building a practice, the part law school leaves out, with courses taught by practicing lawyers. Today SPU is a community for solo and small-firm lawyers, with an AI Fluency program for firms putting AI to work. Nothing here is legal advice: check your own jurisdiction's rules.
About Solo Practice University →