When the Customer Can See the Commission

Making a commission visible rarely changes anything on its own.

That is the finding, and it holds across financial advice in two countries, American real estate, physician prescribing, and a decade of charitable giving at the largest retailer in the United States. Disclosure arrives, customers understand it, and the transaction proceeds roughly as before.

Where behavior does move, something else moved with it. The customer started paying a cost they had not paid before. The recommender came under observation. The seller gave up something valuable and got it back later. A relationship gave the payment a second meaning.

This matters commercially because the standard question is backwards. Companies ask whether customers will trust a recommendation less once they know the recommender is paid. The historical record says that is usually not what happens, and that the interesting variance lives elsewhere. The better question is what a disclosure alters in the mechanics of the deal. If the answer is nothing, expect nothing.

The conventional model runs: reveal the money, the recommendation looks biased, credibility falls, persuasion falls. Sixty years of disclosure regulation rests on it.

Run it against the evidence and it comes apart in every direction.

Sometimes visibility produces no response at all. Ten years of a federal database publishing every industry payment to every American physician produced no evidence that physician conduct around conflicts changed, and no evidence that employers, conference organizers, or journals began excluding physicians on that basis. Patients who viewed the disclosure database formed different views about individual physicians' honesty without changing their trust in the profession or in the pharmaceutical industry. Reported Journal of General Internal Medicine, 2021

Sometimes it produces exit rather than negotiation. When the United Kingdom banned commission on investment advice and the cost appeared as an explicit charge, a measurable group of consumers concluded advice was not worth the price and left the market. They did not shop for a cheaper adviser. Fact FCA post-implementation review, 2014

Sometimes it moves the recommender and not the customer. Massachusetts physician payment disclosure cut branded prescriptions by roughly half in some drug classes while leaving patients essentially unmoved. Fact Marketing Science 39(3)

Sometimes it makes the seller money. Car salespeople who revealed the dealer invoice price early in a negotiation earned about $1,400 more per deal in back-end profit than those who did not. Fact Journal of Marketing Research 57(6)

And sometimes it raises purchase intention outright, which the conventional model has no room for at all. Fact single study, discussed in section 07

Two popular rules are both inadequate. Transparency helps is wrong often enough to be useless. Commissions destroy trust is wrong more often than it is right. The outcome depends on what else moves.

The strongest customer-side evidence in the record comes from Britain.

The Retail Distribution Review took effect at the end of 2012. It banned commission payments from product providers to investment advisers, raised qualification standards, and required advisers to charge clients directly. Fact The commission had always been paid. It sat inside product charges, invisible, and the customer never saw a bill. Afterward the same economic transfer appeared as an explicit adviser charge.

The regulator's own review then did something unusually careful. It split the resulting "advice gap" into three groups: people not engaged in the investment market at all, people unwilling to pay the visible cost of advice, and people who wanted advice but could not find a firm willing to serve them. Fact

The second group is the cleanest customer-side effect available anywhere in this research. Nothing about the economics changed for them. Only the visibility did. In the FCA's own words, revealing the true cost of advice led some consumers to consider whether what they received represented value for money, and in some cases to conclude it did not. Fact

What they did next is the finding.

That deserves emphasis because it is the opposite of what disclosure is designed to produce. The theory of disclosure is that informed consumers shop. These consumers, newly informed, left the category.

Structure moved the effect, and the regulator noticed

The same review adds a detail that took a decade to become interesting. The size of that group was limited by the fact that most firms adopted contingent charging structures rather than up-front fees. Fact

Same information, same customer, less exit when the payment was contingent on the transaction proceeding rather than payable in advance.

What did not happen

Two claims circulate about the RDR that the evidence does not support.

Advisers did not flee. Numbers rose from 35,000 in 2012 to 36,400 in 2019, and the share of UK adults who had received financial advice rose from six percent in 2017 to eight percent in 2019, roughly 3.1 million people to 4.1 million. Fact FCA evaluation of RDR and FAMR, 2020 One trade publication reports UK adviser numbers falling from 300,000 to 22,000 over six years, which is wrong by an order of magnitude. Trade commentary on this reform is unreliable.

And the most visible supply-side change was not a disclosure effect. Banks withdrew from mass-market advice, but the FCA explicitly declines to attribute that to the RDR, describing it as driven by a combination of factors including wider strategic considerations. The report notes that bank advisers had been effective at prompting otherwise unengaged people to invest at all, so their exit enlarged the group who were never in the market rather than the group who balked at a price. Fact

American real estate ran the other version of the experiment.

New rules took effect on 17 August 2024 following the National Association of Realtors settlement. Offers of compensation to buyer agents can no longer appear in the multiple listing service. Agents working with buyers must sign a written representation agreement before touring a home, stating the fee as a specific amount or rate and capping compensation at that figure. Fact

Visibility rose on every dimension. The fee is now stated, written, capped, and signed before the first showing. It is available, visible, comprehended, and salient in a way it never was when it lived in a field agents read and buyers did not.

What did not change is who writes the check. Sellers may still fund the buyer agent, now as a negotiated concession rather than an advertised offer, and in most transactions they still do.

The rates barely moved. Redfin's quarterly analysis of closed sales puts the average buyer's agent commission at 2.43 percent in the first quarter of 2024 when the settlement was announced, falling to a low of 2.36 percent in the third quarter of 2024 when the rules took effect, then climbing back to 2.40 percent, 2.43 percent, and 2.42 percent across the following quarters. Fact Redfin Q1 2025, Q2 2025, Q3 2025. Roughly a fifteen basis point dip, substantially reversed within a year.

The mechanism, stated by the data provider

Redfin commissioned Ipsos to survey 4,000 US residents between 28 March and 28 April 2025. Just over a quarter of people who had bought a home in the past year, 27.2 percent, negotiated or tried to negotiate their agent's commission. Nearly half, 47.8 percent, did not try. Fact Redfin, May 2025

Redfin's own explanation: buyers were less likely to negotiate, probably because they often are not the one paying their agent. Reported Redfin's interpretation of its own survey

That is the rule, observed directly rather than inferred.

The price tier split

Beneath the flat national average, the tiers diverge. For homes under $500,000, the buyer's agent commission reached 2.52 percent by the third quarter of 2025, up from 2.45 percent a year earlier and the highest level since 2023. For homes at $1 million and above it ran 2.17 percent in the first quarter of 2025, down from 2.22 percent when the rules took effect and 2.30 percent a year before that. Fact Redfin

Same disclosure, opposite directions, sorted by price.

The plausible reading is that at the top of the market a buyer can fund an agent directly, negotiate hard, or walk away, while at the entry level a buyer has no cash beyond the down payment and the seller keeps paying to attract offers. Inference not established

Which surfaces a variable the disclosure literature mostly ignores.

Does the customer have a real alternative? If every available intermediary carries similar economics, and if the customer cannot fund a different arrangement, visibility has nowhere to send them.

Britain tested this too, and the answer was worse than expected. Automated advice after the RDR was priced at a level that should have made it accessible to mass-market consumers. It was used primarily by people who could already afford traditional advice. Assets reached £3.2 billion by 2019, under 0.5 percent of the retail investment market, with awareness at nineteen percent. Fact FCA 2020 The unconflicted alternative existed, was cheaper, and did not absorb the people who left.

Availability is not sufficient. The alternative has to be known, trusted, and understood as a substitute for the same job.

The customer channel is the weak one. The recommender channel is where disclosure does its measurable work, and it took this research four passes to notice because we kept looking at buyers.

Massachusetts introduced physician payment disclosure in June 2009. Guo, Sriram and Manchanda used claim-level insurance data from January 2006 through December 2011, comparing physicians in Massachusetts border counties against those in Connecticut, New York, New Hampshire and Rhode Island. Fact Marketing Science 39(3), 516-539

Branded statins fell 46 to 54 percent. Branded antidepressants fell 50 to 56 percent. Branded antipsychotics fell 31 to 32 percent. Fact INFORMS

These are the largest behavioral effects anywhere in this research, and they are on the wrong side of the transaction from where disclosure theory expects them.

The mechanism is not incentive realignment

Generic prescriptions in all three classes also declined. Generic manufacturers do not pay physicians. Fact

If disclosure had simply severed the link between pharmaceutical money and prescribing, branded volumes would fall and generics would rise as substitutes. Instead the whole category contracted. Physicians did not switch products. They prescribed less of everything, including drugs nobody was paying them to prescribe. The authors conjecture self-monitoring against over-diagnosis. Inference the authors' own, from their data

The confound that limits this

Massachusetts is not representative. Pham-Kanter, Alexander and Nair found negligible effects on branded prescribing in Maine and West Virginia. King and Bearman found that gift bans, rather than disclosure alone, produced the larger reductions in drug diffusion. Chao and Larkin found the Massachusetts reductions concentrated among heavy brand-name prescribers. Reported from a review citing all three

Massachusetts is repeatedly described as the more stringent regime. The states where disclosure was disclosure alone produced little.

That pattern matches the rest of the record. Britain banned commission. Australia banned commission. Massachusetts was stringent and paired with restriction. Maine and West Virginia disclosed and got nothing.

It is difficult to separate visibility from stringency, enforcement expectation, and accompanying restriction. The honest position is that the recommender channel responds to restriction and to credible observation, and that disclosure without either does little here too. Inference

What no amount of disclosure reaches

Australia banned conflicted remuneration in July 2013, imposed a best interests duty, and required fee disclosure statements. Then ASIC reviewed the five largest vertically integrated advice institutions between 2015 and 2017, covering AMP, ANZ, CBA, NAB and Westpac.

In-house products made up 21 percent of approved product lists and attracted 68 percent of client funds. Of 200 files where clients were advised to switch from an external product into an in-house one, 75 percent did not demonstrate compliance with the best interests duty, and 10 percent of the advice reviewed was likely to leave the customer significantly worse off. Fact ASIC Report 562, media release 18-019MR

Every one of those files postdates the commission ban. The money had been removed, the duty was law, disclosure was mandatory, and product flow still ran more than three times heavier toward in-house products than their share of the approved list would predict.

Remove the commission, disclose everything, pay by salary, and the conflict survives inside the ownership structure. Disclosure regulates payment. It does not regulate employment.

Every case above is a regulator forcing disclosure on an unwilling industry. The most commercially useful finding in this research comes from the opposite situation.

Atefi, Ahearne, Hohenberg, Hall and Zettelmeyer observed more than 400 real negotiations at a major US auto dealership chain, with full front-end and back-end gross profit data including service. Of those, 30 salespeople disclosed the dealer invoice price early, 44 disclosed later, 25 disclosed only under prodding from the customer, and 301 never disclosed. Four supplemental experiments tested the mechanism. Fact JMR 57(6), 1076-1094

Customers who were shown the invoice price at the start did not pay significantly less on the car. They contributed roughly $1,400 more in back-end profit through add-ons, financing, and service, and they were more likely to return for service a year later. Fact Texas A&M summary, University of Denver

Trust was the mediating variable, and the effect held only when the customer could verify the disclosed information independently. Early disclosure signals sincerity. Late disclosure reads as a last-ditch attempt to save the sale. Reported American Marketing Association summary

Three things follow, and they reorganize the rest of the argument.

Mandatory and voluntary disclosure are not the same object. Mandatory disclosure arrives as compliance, delivers one-shot information, involves no concession by the discloser, and has no later stage in which to recover anything. Voluntary disclosure can function as a costly signal, a trust-building act, and an investment in a later stage of the same transaction. Treating both as "disclosure" has confused this literature for decades.

The mechanism is sequence. The disclosure looks expensive at the moment it is made and profitable afterward, inside the same relationship.

Verifiability is the boundary condition. The signal works because the buyer has already researched the invoice price online and can check. A disclosure the customer cannot verify carries no information about the discloser's sincerity, and this effect should not be expected to survive without it.

Creator commerce is the modern extension of the same idea, and the evidence is weaker than the auto retail case by a wide margin. It belongs here as a second route by which disclosure acquires value through a relationship, not as the center of the argument.

The main body of evidence follows the pattern from section 02. Disclosure raises recognition of commercial motive and understanding of persuasive intent, and behavior stays where it was. Dutch adolescents watching a sponsored TikTok video recognized the ad and understood what it wanted, while brand attitude and product choice did not move. Fact Computers in Human Behavior 144 An Instagram experiment found disclosure raising perceived commercial motive with resistance and advertising effectiveness both flat. Fact Lim and Zhang, NUS, preprint

Then the anomaly. Deng and colleagues ran three experiments and found sponsorship disclosure raising purchase intention, mediated by the parasocial relationship between viewer and creator, in the condition where a well-known creator introduced a brand the audience did not already know. The authors ruled out persuasion knowledge, source credibility, expertise, attractiveness and brand fit as alternative explanations, and call for replication before anyone builds on it. Fact Journal of Consumer Affairs, 2026

The proposed mechanism, and it is proposed rather than established: creator affiliate language does not only reveal a conflict. Use my code, shop through my link, it helps support the channel. That reveals the money and simultaneously tells the buyer their purchase benefits someone they value, at no additional price at the point of sale.

Note what this shares with auto retail and what it does not. In both cases the disclosure acquires value through a relationship. But the auto salesperson gives something up and recovers it later, while creator disclosure gives the customer an additional reason to act in the moment. Different mechanisms. They should not be collapsed.

At this point in the research a clean rule had formed. Three cases, three levels of customer cost, three outcomes. Cost revealed produces exit. Third-party cost produces nothing. Costless benefit to someone the customer values produces lift.

It ordered every case. It had exactly one case per position. So we went looking for a fourth.

AmazonSmile ran from 2013 to February 2023. Customers chose a charity, shopped through a separate URL, paid identical prices, and Amazon donated 0.5 percent of eligible purchases out of its own margin. Fact Costless to the customer. Beneficiary of their own choosing. One setup action, then a habit.

It raised roughly $400 million in the United States and $449 million globally across ten years. The average annual donation per charity in 2022 was about $230. Amazon shut it down, saying the program had not grown to create the impact originally hoped and that its effect was spread too thin across more than a million organizations. Fact widely reported from Amazon's statement, NPR

Estimate $400 million at a 0.5 percent donation rate implies roughly $80 billion in eligible AmazonSmile purchases across a decade, against US net sales exceeding $300 billion in 2022 alone. On the order of low single-digit percent of eligible spending, and probably less.

A costless, one-click, self-selected benefit transfer, offered by the largest retailer in the country for ten years, moved a low single-digit share of behavior and was killed for underperformance.

If directing a free benefit to a chosen recipient were a behavioral driver, AmazonSmile should have worked.

One difference survives. The AmazonSmile beneficiary was an institution the customer selected from a list. Nobody watches a charity three times a week for four years. The beneficiary cannot merely be chosen. The relationship appears to be doing the work, and costlessness only removes the obstacle. Inference

That collapses the third position back inside the parasocial literature it came from, and removes any claim that it generalizes.

Six variables survived the evidence. They are listed in rough order of how much weight they carry.

Sometimes that new transaction is: now I have to write a check. Sometimes: now I can negotiate, and now I know I could not before. Sometimes: now my conduct is on a public database. Sometimes: now the seller has put something at risk in front of me. Sometimes: now buying this benefits someone I follow.

When the disclosure creates no new transaction, and it usually creates none, nothing happens. The customer files the information and proceeds.

That is why the same intervention produces exit in Britain, indifference in American real estate, a fifty percent prescribing shift in Massachusetts, and $1,400 of additional profit in a car dealership. It was never the same intervention.

• Do not treat fee visibility as equivalent to fee incidence. Whether the customer sees the fee and whether the customer pays the fee are separate variables, and only the second reliably predicts behavior.
• If you are making an embedded fee explicit, model exit rather than negotiation. The British case is the warning. Some proportion of customers will conclude the service is not worth its now-visible price and leave rather than bargain. Contingent charging reduced that effect, so payment structure is a lever.
• Before disclosing, ask what the customer can do with the information. If every alternative carries the same economics, or the cheaper alternative is unknown or unusable, disclosure produces awareness and nothing else.
• Separate compliance disclosure from trust-building disclosure, and budget them differently. The first is a cost of doing business. The second is an investment with conditions attached.
• If you are voluntarily revealing costly information, check that the customer can verify it and that a later stage exists. Both conditions were load-bearing in the only case where disclosure made the seller money.
• Measure the recommender channel separately from the customer channel. They move independently, the recommender channel moves more, and customer research will miss it entirely.
• Do not treat creator support language as settled science. The mechanism is plausible and untested. Run it as an experiment against neutral wording with the economics held constant, and measure a downstream trust metric alongside conversion.
• If the conflict lives in ownership, disclosure will not reach it. Nothing in this record suggests otherwise.

Whether support-framed disclosure outperforms neutral disclosure with the economics held constant. The experiment has never been run, and all four cells of the design exist in the wild.

Whether the creator effect runs through support motivation or reciprocity. Years of free content is exactly the prior gift reciprocity requires, so every attached viewer has also received the thing that would trigger it. The distinguishing test has not been run.

Whether the recommender channel responds to visibility or to stringency. Massachusetts against Maine cannot separate them. It would need a case where disclosure was strict and restriction absent.

What repeated disclosure does to a relationship over years. Almost every study in this literature is single-exposure and short-term.

Whether the post-NAR pattern survives better identification. Two years, bundled rules, brokerage-sourced data, and an unusual rate environment.

Whether voluntary disclosure generalizes past auto retail. One paper, one industry, one product category with an unusual back-end structure.

Field reforms contain real money and real behavior, and they rarely change disclosure alone. Fiduciary rules change duties and disclosure together. Commission bans change compensation and salience together. The NAR settlement changed negotiation structure alongside visibility. Lab studies isolate disclosure more cleanly, usually without meaningful financial stakes. This research triangulates between causal cleanliness and ecological validity rather than pretending either class settles the question alone. Where they disagree, that is reported.

Three distinctions were load-bearing.

Customer-side against recommender-side response. These move independently and are routinely reported as one effect.

Demand-side against supply-side change. When a service becomes less available after a reform, whether customers stopped seeking it or providers stopped offering it looks identical in outcome data and means opposite things. Applying this distinction substantially changed the reading of the British case.

Stay, switch, and exit. Two of these look the same in conversion data. Separating them is what surfaced the strongest customer-side finding in the research.

Evidence is labeled throughout: FACT for primary evidence, REPORTED for credible secondary evidence, ESTIMATE for figures derived from disclosed numbers with the calculation shown, INFERENCE for reasoned interpretation, UNRESOLVED where evidence is insufficient or conflicting.

Two attractive hypotheses were built and discarded during this work. The first was a three-position rule ordering cases by customer cost, which AmazonSmile falsified. The second was that the recommender channel operates through incentive realignment, which the generic prescribing data contradicted. Both are recorded here rather than quietly dropped, because the shape of what broke is part of the finding.