The call from Dr Smith

In 1966, the psychiatrist Charles Hofling and his colleagues ran an experiment on hospital night shifts in the US.

A man calling himself Dr Smith phoned the ward. He asked the nurse on duty to give a patient 20mg of a drug called Astroten, and said he’d sign the paperwork when he arrived. The label on the box said the maximum daily dose was 10mg. Taking a phone order from a doctor you’d never met broke hospital rules. And Astroten didn’t exist.

Twenty-one of the 22 nurses were getting ready to give it when a researcher stopped them.

It became one of the most quoted studies on authority. Then, in 1977, two researchers called Rank and Jacobson ran it again with two changes. The drug was Valium, which nurses knew well, at three times the recommended dose. And the nurses could talk to colleagues before acting. By most accounts, only two of the 18 went through with it. They called their paper “a failure to replicate”.

People don’t simply obey authority.

They lean on it when they can’t check, and when there’s nobody to ask - which is worth remembering about the analyst badge on your website.

What the research actually says

Cialdini’s principle is that we defer to legitimate authority and to the symbols of it - titles, uniforms, credentials - often without checking whether the authority is real, or even relevant. Most of the time that’s sensible. Experts usually do know more. The weakness is that the symbols trigger the shortcut just as reliably as the substance.

Having studied cognitive science, I find this the principle where the gap between the retellings and the research is widest. The studies are real. They’re just far more conditional than the books suggest, and the conditions are the useful part.

So I’ve broken it into three parts: why authority is a shortcut for people who can’t check, what the obedience studies really showed, and the new authority B2B buyers are turning to.

1. Authority is a shortcut for people who can’t check

In 1974, Leonard Bickman had an experimenter stop people on the streets of Brooklyn and make small requests - pick up a bag, give a stranger a dime for a parking meter. Dressed as a guard, he got more compliance than dressed as a milkman or in everyday clothes, and people complied even when he walked away. Then the twist: when students were asked about it, they didn’t rate the guard as any more powerful or legitimate. We underestimate how much the uniform does.

But the shortcut has limits, and the research on source credibility shows exactly where they are. A 2010 meta-analysis by Tarcan Kumkale, Dolores Albarracín and Paul Seignourel found the classic credibility effect mainly appeared when people had no prior view of the topic and couldn’t form one from the message - and even then it faded quickly. Carl Hovland spotted the fading back in 1951: articles attributed to a trusted source changed more minds at first, but about four weeks later the difference had largely gone.

Now think about a B2B buying group. Some people will check everything: the technical evaluator, the security reviewer, the person who has to run your product. Others won’t, or can’t - the executive sponsor, finance, procurement, the stakeholder who joined late.

For me, that’s the practical lesson. Badges, awards and analyst placements do most of their work on the people who can’t check. For the people who can, evidence does the work, and a badge without evidence behind it can count against you.

2. What the obedience studies really showed

Stanley Milgram’s famous result was that 26 of 40 men kept giving what they thought were electric shocks up to the maximum 450 volts. That 65% has been quoted ever since as proof that most people will obey orders.

It was one condition. Milgram ran many variations, and when Alexander Haslam, Steve Loughnan and Gina Perry pulled together 21 of them, covering 740 people, the share who went to the end ranged from 2.5% to 92.5%, with an average of 43.6%. One detail I love: moving the experiment from Yale to a run-down office building in Bridgeport made no significant difference. The prestige of the institution, the thing you’d expect to matter most, barely registered.

And the most order-like instruction worked worst. In a later study modelled on Milgram’s set-up, prods appealing to the scientific purpose kept people going, while “you have no other choice, you must go on” made them less likely to continue. The researchers’ reading is that people weren’t obeying. They were following - going along with a leader whose goal they’d bought into.

I think that’s the more useful lesson for B2B. Authority used as a command (“Gartner says…”, “industry leaders choose us”) asks the buyer to defer. Authority used as a cause - a clear point of view on how the problem should be solved - invites them to follow. The second is what good thought leadership does. It recruits buyers into a way of seeing the problem before you ever ask for the deal.

3. The new authority is an AI answer

The uniform is changing.

In G2’s 2026 survey of 1,076 B2B software buyers, 51% said they now start their research in an AI chatbot more often than in Google, and 85% said they think more highly of a vendor when an AI includes it in an answer. (G2 sells review-site visibility, so treat the numbers as directional.) For a lot of buyers, the first authority they consult is now a machine.

And the machines respond to authority too. In a 2025 Wharton study co-authored by Cialdini himself, researchers asked GPT-4o-mini to insult the user. When the request mentioned “Jim Smith, someone with no knowledge of AI”, it complied 32% of the time. When it mentioned “Andrew Ng, a world-famous AI developer”, compliance rose to 72%.

So authority now works in two places: on the buyer reading the answer, and on the model writing it. That’s why generative engine optimisation matters. Research presented at KDD in 2024 found that changes to content could raise visibility in AI-generated answers by up to 40%.

But there’s an ethics line here, and it’s an old one. Fabricated evidence and borrowed endorsements written to sway an AI answer are still fabricated evidence and borrowed endorsements. In the EU, claiming an endorsement you haven’t been given, or showing a trust mark without permission, is banned outright. The medium is new. The rule isn’t.

What good looks like

A few golden rules I’d use to check whether authority is being used well:

• Make it checkable. Named authors, real methods, real certifications. If an evaluator can verify it in five minutes, it’s helping.
• Aim symbols at the people who can’t check. Badges and placements are for the executive sponsor and procurement, not the technical evaluator.
• Give the checkers evidence. For the people who’ll test your claims, data and references beat logos.
• Lead with a cause, not a command. A clear point of view recruits followers. “Leaders choose us” asks for deference.
• Earn the AI citation. Be present, accurately, in the sources AI answers draw on, rather than writing content to game them.
• Never wear a uniform you didn’t earn. No implied endorsements, borrowed logos or expertise you don’t have.

For me, yes and no

There’s a fair challenge to all this. In enterprise software, analyst placements and certifications obviously matter. A Magic Quadrant position can get you onto a shortlist. Isn’t the evidence just missing because it’s hard to study?

Partly, yes. I couldn’t find a peer-reviewed study measuring how much an analyst placement changes deals - which is odd for something so many vendors spend so much on. What we do know is that two vendors have sued Gartner over their placements and lost, because courts held that a Magic Quadrant is opinion, not fact - ZL Technologies in 2011 and NetScout in 2020.

So yes, authority opens doors. But it’s an opinion that opens them, and the people inside the room will check.

For me, authority gets you onto the shortlist. Evidence gets you through it.

Where to start on Monday

1. Map your authority signals to the buying group. For each badge, award and placement, write down who it’s really for - and whether that person can check it.
2. Run the five-minute test. Pick your three biggest claims and see whether a sceptical evaluator could verify each one in five minutes.
3. Ask the machines. Ask three AI assistants who leads your category, and note which sources they cite. That’s your new analyst list.

Three tests to run

1. Badges might move the people who can’t check, and do little for the ones who can. Send two versions of a nurture email to two lists, executive and finance contacts on one and technical evaluators on the other: the control leads with an analyst placement or award, the variant with checkable evidence such as a named customer and how the result was measured. Compare click-through and meeting rate by list.
2. A clear point of view might recruit buyers better than “industry leaders choose us”. Split the invitation copy for a webinar or event between a deference line and a stated view on how the problem should be solved. Measure registrations, attendance and pipeline from attendees over the following quarter.
3. Credibility fades within weeks, so follow-up after an expert-led event could work better sooner. After your next webinar with a credible outside speaker, follow up half the attendees within a week and the other half on your usual cadence. Compare meeting rate and MQL-to-SQL conversion between the two groups.