The smokers picked the wrong ad

In 2012, Emily Falk and her colleagues put smokers in a brain scanner and showed them three TV campaigns for the US National Cancer Institute’s quitline, 1-800-QUIT-NOW.

They also asked the smokers a simple question: which campaign do you think will work best?

Then they compared both answers with what actually happened - how many more people called the quitline after each campaign aired. The smokers’ own predictions didn’t match the results. Activity in a patch of their medial prefrontal cortex did.

It’s the kind of finding neuromarketing was built on. The brain knows what the buyer doesn’t.

For your next campaign, it means less than the headline suggests, and more than the sceptics admit. It was three campaigns. When Falk and a large group of colleagues pooled 16 brain-imaging studies covering 572 people and 739 messages, brain activity explained on average 1.17% of the variance in how well messages worked at scale. Within individuals, it was 0.10%.

Real, but small. Neuroscience can forecast what a crowd might do a little better than a panel can - but it can’t read your buyer.

What the research actually says

This is the first essay in a new series on the mind behind the buyer: what neuroscience has found about attention, memory, value, risk and group decisions, and what a B2B marketer can do with it.

I studied cognitive science, and my final-year dissertation was on attention and decision-making. So I’ll hold this series to a simple rule. Everything I cite is peer-reviewed, and I’ll say plainly what hasn’t held up. One caveat applies to all of it: none of the studies I’ve found used B2B buyers, buying committees or long sales cycles. Every step from the lab to your pipeline is my inference, and I’ll mark it as one.

So I’ve broken this first essay into three parts: what brain data can predict, what it can’t do, and what it points to that you can test without a scanner.

1. What brain data can predict

The smokers weren’t a one-off. The same pattern has turned up in a handful of studies, and it’s an interesting one.

In 2017, Alexander Genevsky, Carolyn Yoon and Brian Knutson scanned 30 people as they looked at real projects on a crowdfunding website. Activity in the nucleus accumbens - a small region tied to anticipating reward - forecast which projects the wider market actually funded weeks later. It did better than the participants’ own choices, which didn’t forecast the market at all. A second study replicated it.

In 2015, a team led by Vinod Venkatraman tested 30-second TV ads with six families of methods - self-report, implicit measures, eye tracking, biometrics, EEG and fMRI - and linked them to how the ads actually moved sales. fMRI explained the most variance beyond the traditional self-report measures, with activity in the ventral striatum the strongest single predictor.

The common thread is an argument I find persuasive in the neuroscientist Robert Sapolsky’s book Determined: our choices are shaped by biology and circumstance before we’re aware of them, and we supply the reasons afterwards. You don’t have to go as far as he does - he concludes there’s no free will at all - to accept that we explain our choices after the fact.

In one lab study, Petter Johansson et al. showed people pairs of faces and asked them to pick the more attractive one. Then they secretly swapped the result, handing back the face the person had rejected, and asked why they’d chosen it. Often, people didn’t notice the swap - and went on to explain, in their own words, a choice they had never made. The researchers called it choice blindness.

I think that’s the real lesson for marketers. When you ask a panel “which ad will work?”, they answer a different question from the one the market answers.

2. What it can’t do

Now the limits, which are more important than the headlines.

The effect is small. That 1.17% from the pooled analysis works out to a small effect size - the authors put it at roughly d = 0.22 - and they note it was sensitive to analytical choices. It’s useful at the level of a population. It’s nearly useless at the level of a person.

Individual brain readings are also unreliable. When Maxwell Elliott and colleagues pooled 90 experiments with 1,008 participants, the average test-retest reliability of common task-fMRI measures was an ICC of .397 - poor. Their conclusion was that these measures aren’t currently suitable for studying differences between individuals. If the same brain gives a different reading next week, it can’t tell you much about one buyer.

And a region “lighting up” doesn’t tell you which thought caused it. Russell Poldrack named this problem of reverse inference back in 2006: most brain regions do many jobs, so activity in one doesn’t prove a particular feeling. That’s worth remembering the next time a deck says your ad “activated the brain’s reward centre”.

For me, that rules out the idea of neuroscience as a mind-reading tool for B2B. There’s no buy button. There’s a weak, real signal at the level of crowds.

3. What it points to

So what is the signal actually measuring? This is where it gets useful.

The pooled analysis found that the brain activity linked to message effectiveness sat in reward and social processing regions - areas tied to valuing things and to thinking about what other people think. Falk’s group had already shown the same pattern for sharing: whether people pass a news article on depends on how relevant it feels to them and to the people they’d share it with.

And relevance can be tested without a scanner. In six studies with 3,727 people rating 362 messages, Danielle Cosme and colleagues found that reframing messages to make them more relevant to the reader, and to their social circle, causally increased intentions to share them in a preregistered experiment.

That’s the most useful thing neuroscience has told marketers so far, and it doesn’t need a lab. In B2B, the reader is rarely the only decision-maker. The content that moves a buying group is the content that resonates and makes them feel it’s about their problem - and that they can imagine forwarding to a colleague with a one-line note.

What good looks like

A few golden rules I’d use for applying this science honestly:

• Treat panel predictions as weak evidence. “Which do you prefer?” tells you what people think will work, which is a different thing.
• Test in market. A real A/B test with real buyers beats any lab forecast, neural or not.
• Write for the reader and the forward. Ask who they’d send it to, and whether it would make sense to that person.
• Ask neuro vendors for validation. Out-of-sample, against real outcomes like sales or pipeline. I couldn’t find peer-reviewed evidence of that for vendor “attention metrics”.
• Never read one buyer’s mind from brain data. The measures aren’t reliable enough at the level of an individual.
• Watch for reverse inference. “It lit up the reward centre” isn’t evidence of anything on its own.

For me, yes and no

There’s a fair challenge here. If brain data explains about 1% of what makes a message work, why should a marketer care at all?

For choosing B2B creative, I don’t think they should. A scanner is an expensive way to get a small signal from a handful of consumers, and none of it has been tested on business buyers.

But the direction of the research matters more than the size of the effect. It keeps pointing at the same things: fast, emotional reactions predict a crowd better than considered opinions, and relevance to the reader and the people around them drives what spreads. Both of those can be designed for, and tested cheaply, in the market you actually sell into.

For me, neuroscience is a compass, not a map.

Where to start on Monday

1. Retire the preference panel. For your next creative decision, run an in-market test instead of asking a group which version they like.
2. Rewrite one piece for the forward. Pick a high-value asset and ask: who would the reader send this to, and what would they need to see?
3. Ask for the out-of-sample evidence. If a vendor sells you neuro or attention metrics, ask how well they’ve predicted real outcomes they weren’t trained on.

Three tests to run

1. A panel’s favourite might not be the market’s favourite. Before your next campaign, ask sales or a customer panel to rank three ads or subject lines, then run all three in market with an even split for a few weeks. Compare the panel’s ranking with click-through and meeting rate.
2. Copy about the reader’s problem could beat copy about your product. Split a landing page or paid social ad with the same offer: the control leads with product features, the variant leads with the buyer’s problem in their own words. Measure form completion and MQL-to-SQL conversion.
3. Content written for the buying group might reach more of the account. Split an account-based nurture: the control is written to the individual, the variant to what the problem means for their team, with a short summary they could pass on. Over a quarter, measure the number of contacts engaged per account and meeting rate.