Why advocacy by human beings may become more valuable as AI recommendations become ubiquitous
In Brain File 01 we argued that as intelligence becomes abundant, distinctly human capabilities may become more valuable. Brain File 02 explored emotion: the force that makes an experience matter. Brain File 03 explored trust: the confidence that makes us willing to act.
But there is a further human advantage. We do not usually decide what to believe based only on our own thoughts and feelings. We look to other people to help us come to that decision. A friend puts their reputation behind a restaurant. A colleague recommends a new hire. A respected leader publicly backs an unpopular idea. The information matters. But so does the person willing to stand behind it.
This fourth Brain File explores the difference between recommendation and endorsement. A recommendation helps us choose. An endorsement can make us brave enough to act. And in a world overflowing with machine-generated suggestions, knowing who stands behind something matters more than ever.
AI can predict what you might like, humans put their name behind it
AI recommendation engines can identify patterns across millions of choices, personalise suggestions and surface products, content and people we might never discover ourselves. This is a genuine advantage. In many settings, algorithmic recommendations can outperform unaided human judgement. But imagine two messages. One says: “People like you frequently choose this.” The other says: “I chose this. I believe in it. And I am happy for you to know that.”
Both can influence behaviour. Only the second has a person behind it. It carries identity, relationship and reputational risk.
AI can rank the options. It cannot stake its name on one.
It can predict what you will like. It cannot vouch for it.
It can write a glowing testimonial. It cannot be embarrassed when it is wrong.
The behavioural science behind endorsement
Declassified brain science
Robert Cialdini’s principle of social proof explains why other people’s choices become especially influential when we are uncertain. If others appear to have chosen something, it can feel safer and more legitimate. Yet social proof is not the same as personal endorsement: a popularity signal tells us what others do; an endorsement tells us who is prepared to stand behind a choice.
Classic work by Hovland and Weiss (1951) demonstrated that the perceived credibility of a communicator can shape persuasion. The same message may land differently depending on who delivers it. Source matters, not merely content.
Why does putting your name to something carry such weight? Signalling theory offers an answer. Economist Michael Spence (1973) showed that a signal is believed when it is costly to send: a qualification means something because it takes real effort to earn. Biologist Amotz Zahavi (1975) found the same principle in nature. His handicap principle explains the peacock’s tail: it is heavy, conspicuous and makes the bird easier for predators to catch, so only a peacock that will be terrifying in a fight and very difficult to pin down can afford one. The cost is precisely what makes the signal honest. Endorsement works in the same way. When someone attaches their name to a choice, they put their reputation at risk if they are wrong. That cost is what makes it credible.
Consumers say the same thing. In Nielsen’s Global Trust in Advertising survey (2015), recommendations from ‘people we know’ were rated the most credible form of advertising, trusted by 83% of respondents worldwide.
And research by De Veirman, Cauberghe and Hudders (2017) complicates the simplistic equation of influence with follower numbers. Larger audiences increased perceived popularity and likeability in their studies but did not consistently translate into opinion leadership; for some distinctive products, larger influencer audiences could even weaken brand attitudes.
The lesson?
Reach is not credibility
Popularity is not conviction
Recommendation is not endorsement.
And none of this means people always prefer human advice: in six experiments, Logg, Minson and Moore (2019) found that participants often weighted algorithmic advice more heavily than equivalent human advice in estimation and forecasting tasks. The human advantage is not universal predictive superiority. It is the social meaning that comes from someone personally standing behind a choice.
Department B evidence file
The Reputation Test
Think of a restaurant recommendation. A platform may calculate that you have an 87% chance of liking it. Useful but a close friend who knows you says, “Trust me. Take someone special there.” Suddenly the choice is embedded in a relationship. If the evening disappoints, that friend may hear about it. Their endorsement has a social consequence.
Now translate that into business. An employee who recommends their employer to a friend; a customer who publicly defends a brand; or a CEO who personally sponsors a difficult decision, are doing more than circulating information. They are associating their own standing with the outcome. Consider Martin Lewis. His recommendation is so trusted that scammers have repeatedly faked it, putting his name and face on fraudulent adverts and, more recently, AI-generated deepfake videos. In 2019 he settled a lawsuit against Facebook over scam adverts using his image, with Facebook agreeing to donate £3 million to Citizens Advice and launch a UK scam ads reporting tool. When an endorsement is worth counterfeiting, you know it is worth something.
Why AI Cannot Fully Replicate This
Not because the technology is immature but because endorsement is not information, it is a commitment. An AI system can generate a convincing testimonial, select a spokesperson, or simulate a recommendation in someone’s voice. It can even become a source that users rely upon. So it would be wrong to claim that people never trust or follow AI advice. The structural difference is responsibility. A model does not itself possess a human social identity, personal reputation among peers or lived obligation to the people it influences. Those stakes sit with the humans and institutions that build, deploy and stand behind it.
AI has recommended a million restaurants, but:
It has never had to face a friend after a bad one
It has written countless testimonials
It has never had its own name on one
It has read about reputation.
It has never had one to lose.
An AI-generated endorsement can sound sincere. But sincerity is not a writing style. It is a claim about the intentions and commitments of an accountable person. AI can reproduce the language of conviction. Humans can be answerable for conviction.
The organisational risk nobody is talking about
As organisations automate recommendation, targeting and content production, they may become spectacularly efficient at telling people what to consider while neglecting the people who make those recommendations believable. Employee voices become scripted. Customer advocacy becomes incentivised. Influencer partnerships become exercises in audience size. Influencers are scooped up by brands once they achieve a sufficiently high follower count and instantly lose the power of the individual endorsement - paid by brands they can relinquish validity. Leadership communication becomes perfectly polished and strangely ownerless.
There is also a more serious danger: synthetic reviews, undisclosed sponsorship and AI-generated testimonials can blur the boundary between genuine advocacy and manufactured approval. Once audiences suspect that endorsement is for sale or machine-made, the value of real human recommendation may be harder to recognise. In the UK, this is no longer simply a reputational risk. Since April 2025, the Digital Markets, Competition and Consumers Act has banned fake reviews and concealed incentivised reviews, and the CMA can fine businesses up to 10% of global turnover. Manufactured approval is now a board-level compliance issue as well as a brand one.
The danger is not AI making recommendations, it is organisations mistaking scalable recommendation for earned advocacy and automating away the independent human voices that give a brand credibility.
What this means for leaders
The competitive question is not simply “How can we optimise our recommendation engine?”. It is “Who would voluntarily put their name behind us, and why?”
For CMOs, this means treating authentic advocacy as an asset distinct from reach and impressions. For CEOs, it means understanding that visible personal commitment can matter more than another immaculate corporate statement. For people leaders, it means creating workplaces employees would genuinely recommend, rather than asking them to share pre-written praise.
"The most powerful endorsement rarely comes from the loudest celebrity or the largest influencer. It comes from someone whose judgement matters deeply to a particular audience."
Crawford Hollingworth
M&S Food offers a timely example. In May 2026, Kathryn Turner, its Director of Product Development, launched her own TikTok and Instagram accounts. She was not a celebrity or a professional influencer. She was someone who has worked at M&S since 1995 and knows the food intimately. Within months, her behind-the-scenes videos had attracted more than 50 million views, around 300,000 followers on TikTok and 230,000 on Instagram – more than the official M&S Food account. Audiences were not responding to reach, they were responding to real expertise, and to a person prepared to put her own name behind the food.
There is a further twist. AI recommendations are increasingly built on human endorsement. AI assistants and search tools draw on reviews, forums and real people’s posts to decide what to suggest. So genuine advocacy now works twice: once when a person vouches for you directly, and again when the machines learn from what people have said. Earned endorsement is not the opposite of AI recommendation. Increasingly, it is the raw material.
Department B evidence file
The endorsement audit
Review your last major campaign, product launch or leadership announcement. Identify the people who publicly advocated for it.
Did they have genuine experience?
Were they free to disagree?
Did their audience understand any commercial relationship?
Did their endorsement require them to put something of their own standing on the line?
If your strongest proof is a collection of anonymous five-star ratings, you may have recommendation signals. You may not yet have earned endorsement.
C-SUITE ACTION
Ask your executive team:
Then run a simple test: ask 10 customers and 10 employees whether they would recommend you to someone whose opinion they value. This is, in essence, the Net Promoter question. Fred Reichheld introduced the Net Promoter Score in Harvard Business Review in 2003, arguing that a single question – how likely are you to recommend us to a friend or colleague? – was a better predictor of growth than lengthy satisfaction surveys. It has since become one of the most widely used customer metrics in business. But do not stop at the score. Ask what they would actually say, and what might make them hesitate. That hesitation may be your most useful insight.
In Brain File 02 we explored how emotion gives communication significance. In Brain File 03 we explored how trust makes belief possible. Brain File 04 adds the social dimension: belief can travel when someone we value is willing to attach their name to it.
"AI will make recommendations ever more precise but competitive advantage may come from inspiring something no recommendation engine can itself be personally accountable for: a real person saying: I stand behind this."
Crawford Hollingworth
AI can tell the world what to choose. Only humans can put their name to it.
Next Brain File
Human advantage brain file 05 - AI Scales Data. Humans Scale Belief.
References
Hovland, C. I. & Weiss, W. (1951). The Influence of Source Credibility on Communication Effectiveness. Public Opinion Quarterly, 15(4), 635–650.
De Veirman, M., Cauberghe, V. & Hudders, L. (2017). Marketing through Instagram influencers: the impact of number of followers and product divergence on brand attitude. International Journal of Advertising, 36(5), 798–828.
Logg, J. M., Minson, J. A. & Moore, D. A. (2019). Algorithm appreciation: People prefer algorithmic to human judgment. Organizational Behavior and Human Decision Processes, 151, 90–103.
Cialdini, R. B. (2009). Influence: Science and Practice (5th ed.). Pearson.
Spence, M. (1973). Job Market Signaling. The Quarterly Journal of Economics, 87(3), 355–374.
Zahavi, A. (1975). Mate selection – a selection for a handicap. Journal of Theoretical Biology, 53(1), 205–214.
Nielsen (2015). Global Trust in Advertising. The Nielsen Company.
Reichheld, F. F. (2003). The One Number You Need to Grow. Harvard Business Review, 81(12), 46–54.
Digital Markets, Competition and Consumers Act 2024, Schedule 20 (banned practices relating to consumer reviews), in force 6 April 2025.
