The question funnel and cognitive load

This chapter holds the most surprising thing in the whole book: most of the gain comes not from reading better, but from designing the questioning so that the difference between truth and falsehood reveals itself — clearly enough for an untrained observer to see it.

The funnel method

When the conversation moves towards the target naturally, the subject finds it harder to dodge — because an apparently natural conversation has led gradually there. And once the target is reached you move away from it gradually and change topic, so that they do not notice what they have just given away.

Three reasons open questions matter

They support the subject's sense of autonomy — see the five rapport behaviours in chapter 24
They obtain additional and unexpected information — which a closed question, by definition, can never give
The subject talks more → more indicators for you to read

The real problem is not how to get them to talk

These two numbers reorient the whole chapter. Most people will talk. Your job is to create the conditions in which they talk at greater length and in more detail — because it is length and detail that actually discriminate.

Why talking more is what discriminates

When a truthful person is encouraged to say more, they can and they do — they draw on real memory to supply detail that makes the account sound more plausible.

A liar hesitates to supply more detail, for three reasons: they cannot draw on real memory, they lack the imagination to invent a detailed story that holds together, or they are afraid of giving themselves away. The more they say, the more places there are to get it wrong.

Five tested tools for making people say more
Tool	Note
Supportive interviewing	A supportive tone rather than an accusatory one — see chapter 24
Mirroring	One of the few techniques proven to make a truthful person say more — see chapter 20
A silent second interviewer, who only nods and smiles	Effective even when the subject looks at that person less than ten per cent of the time
Explaining the subject's role in the interview clearly	“You are the one who knows about this, I am not. So you need to give more detail than you think is necessary.”
Playing a model description	A recording describing an unrelated event but in great detail. It widens the difference between truthful and deceptive accounts by resetting expectations about how much detail is expected

Memory enhancement

Begin with a request for a free account
“Tell me, from the beginning, in your own words.” Then be silent.
Rely on open questions, and do NOT interrupt
Every interruption is a moment in which you replace their memory with your structure.
Guide them back into the event in their own mind
And to think about what they experienced through each sense in turn: sight, hearing, smell, touch, taste. This usually draws out detail that direct questions never reach.

The principle. One run of questions can draw out part of a memory, but never the whole of it. The rest comes when you ask again from another angle — and the difference between the two tellings is itself the data.

Strategic use of evidence

This is the technique that replaces the straight accusation, and it is far more effective because it forces the other person to commit before they know what you have.

What has to be recorded is not their explanation, but the inconsistencies within one statement and between statements as the level of evidence rises. Each step is a chance for them to correct themselves — and the way they correct is data.

Cognitive load

The theory is compact: lying is multitasking, telling the truth is not.

The whole cognitive load strategy rests on this asymmetry: add one more job to the list on the left and the system overloads; add one to the list on the right and nothing happens.

The largest gap is on the middle row: deception detection, 67% against 46% — meaning that on that row the traditional approach is worse than chance.

Seven techniques for raising load

#	Technique	Mechanism	Data
1	Telling it in REVERSE order, from the last event back to the first	Memory is encoded in forward sequence; telling it backwards forces reconstruction instead of replay	75% correct detection, against 18% when it is not requested
2	Asking for a DRAWING of the scene	Demands detailed spatial information that the spoken account does not contain	78% correct classification. Truthful people draw in more detail and are far more likely to draw the people too, with individual features; liars avoid drawing people because that is one more source of verification
3	UNANTICIPATED questions	A liar prepares for the questions they predict. Ask about process, about planning or about space and they have nothing ready	The procedure: ask the predictable questions first to set a standard, then ask the unpredictable ones → the liar's level of detail drops sharply
4	Asking them to hold eye contact	People tend to look away while recalling; holding eye contact is distracting and adds load	More indicators appear under this condition
5	Asking for VERIFIABLE detail	A liar wants to give plenty of detail in order to make a good impression, but is wary of giving traceable detail — so the detail they do give cannot be checked	Possibly the one tactic with no effective countermeasure: only a truthful person can supply genuinely verifiable detail. And it is useful even when you never actually go and verify anything
6	Drawing and answering at the same time	Two tasks in parallel	With a truthful person this draws out more information; with a liar it makes holding the line harder
7	Interviewing in pairs	Two people who both know about an event, questioned separately	Truthful people use a “tell everything” strategy; deceptive pairs draw the scene less alike than truthful pairs

Three exceptions to the cognitive load model — you have to know them

A prepared lie is far easier than a spontaneous one. The load is low, so the model weakens.
A small lie embedded in a large truth is very hard to detect, because most of the account is true and costs nothing to produce.
Sometimes telling the truth costs more than lying — when the relevant memory has not been accessed for a long time, or when the truth requires a long and complicated explanation. In these cases the honest person will display exactly the indicators of cognitive load.

Cognitive load indicators in the body

Analysis of interviews shows that a prepared lie usually comes with a triad of its own, and that triad is the reverse of the anxiety triad:

This is the same distinction that appeared in chapter 17, seen from the other side. Someone sitting very still, speaking slowly with pauses, moving their hands little — that is not a calm person. That may be a person working very hard inside their head.

Putting the whole chapter together

A well-designed questioning session runs as follows, and it uses almost everything in Book Four:

Round one — a shallow pass, building the heat map
Ten to fifteen topics, one to two minutes each. No digging. Use the heat-map technique to put them in order.
Round two — funnel into the hottest topic
Open out, narrow down, hit the target, then widen again. Do not stop at the target — that is the moment they learn what you care about.
Round three — raise the load where it is needed
Use only one or two load-raising techniques, and only on a topic that has already produced several indicators. Using all seven on every topic is the surest way to destroy rapport.
Round four — strategic use of evidence
Only when you have evidence, and only after they have committed to a version.
And then — independent verification
Chapter 26. Without this step, the four rounds above produce nothing but a neatly arranged hypothesis.

Chapter exercises

One. The next time someone tells a long story, ask them to tell it again from the end back to the beginning. Try it on something harmless first, to see how hard it is even with a wholly genuine memory.

Two. In one meeting, practise the funnel: start with “Tell me about…”, and ask no closed question in the first three minutes.

Three. Prepare three “unanticipated” questions for a topic you will be discussing this week — about process, about planning, or about the layout of a space. Ask them after two questions they could have predicted.

Chapter summary

The most surprising thing in the whole book: most of the gain comes not from reading better, but from designing the questioning so that the difference between truth and falsehood reveals itself — clearly enough for an untrained observer to see it.
The funnel: open out → open probes → narrow down → clarifying questions → the target → widen again, change topic. The lower half of the funnel matters as much as the upper: do not stop at the target, because that is the moment they learn what you care about.
The problem is not “how do I get them to talk”. In 600 interviews, 80% cooperated and answered readily; in 1,067 interviews, only about 5% stayed completely silent. The problem is how to get them to talk a LOT.
Why length discriminates: a truthful person who is encouraged can say more, because they are drawing on real memory. A liar hesitates — the more they say, the more places there are to get it wrong.
Begin with a request for a free account, then be silent. And do NOT interrupt: every interruption is a moment in which you replace their memory with your structure. Guiding them back into the event in their own mind, sense by sense, usually draws out detail that direct questions never reach.
Strategic use of evidence replaces the straight accusation: disclose nothing → ask for a free account → ask specific questions to force them to commit to one exact version → produce evidence of vague source and low specificity → only then produce evidence of precise source and high specificity. What has to be recorded is not their explanation, but the inconsistencies as the level of evidence rises.
Lying is multitasking, telling the truth is not — seven jobs at once against one. The whole load-raising strategy rests on that asymmetry. The cognitive approach reaches 71% overall against 56%; on the deception-detection line alone it is 67% against 46% — meaning that on that line the traditional approach is worse than chance.
The two techniques with the strongest figures: telling it in reverse order gives 75% correct detection against 18% when it is not requested, and asking for a drawing of the scene gives 78% correct classification. Liars avoid drawing people, because that is one more source of verification. And asking for verifiable detail may be the one tactic with no effective countermeasure — useful even when you never actually go and verify anything.
Three exceptions to the cognitive load model that you have to know: a prepared lie carries a low load; a small lie embedded in a large truth is very hard to detect; and sometimes telling the truth costs more than lying — when the memory has not been accessed for a long time, or when the truth requires a long explanation. In those cases the honest person will display exactly the indicators of load.
The cognitive load triad is the exact reverse of the anxiety triad: more pauses in speech, less blinking, less hand movement. Use the whole triad to discriminate, never a single indicator. Someone sitting very still, speaking slowly with pauses, moving their hands little is not a calm person.
The sequence of a well-designed questioning session: round one, a shallow pass over ten to fifteen topics to build the heat map → round two, funnel into the hottest topic → round three, raise the load, only one or two techniques and only on a topic that has already produced several indicators → round four, strategic use of evidence, only after they have committed. Using all seven techniques on every topic is the surest way to destroy rapport.

Previous chapter24 · First impressions and rapport Next chapter26 · Read, steer, verify