A client says you missed something. A different client says you put another client's interests first. Most firms respond to both the same way, with an apology and an offer to fix it, and there is evidence that one of those responses is aimed at the wrong problem.
Key Takeaway
Two studies found that trust was repaired more successfully when mistrusted parties (a) apologized for violations concerning matters of competence but denied culpability for violations concerning matters of integrity, and (b) had apologized for violations when there was subsequent evidence of guilt but had denied culpability for violations when there was subsequent evidence of innocence[1]. The second clause is the one that constrains the first, and it is almost always dropped.
What This Article Does Not Say
Stated first, because this finding is easy to misread in a damaging direction.
The research reports that denial repaired trust better than apology for integrity allegations, and separately that denial worked where subsequent evidence showed innocence[1].
Three things that does not mean, and we state them plainly.
It is not a finding that denying wrongdoing works. The second clause conditions the result on what the evidence later showed, and the study also found apology outperformed denial where evidence of guilt subsequently emerged.
It is not advice about how to respond to an allegation. How a regulated firm answers a complaint engages professional conduct rules, regulatory notification duties, insurer requirements and, potentially, legal privilege. None of that is addressed here, and all of it takes precedence over anything in a research summary.
And it is not legal advice. If your firm faces an allegation of any seriousness, the sequence is your professional body, your insurer and your lawyer, before anything else.
Our Grades For These Claims
Applying the scheme from the first article in this series.
The interaction between violation type and response is Grade B. It comes from two experimental studies in a leading journal, with random assignment and a mediation analysis, which is a good design. It is not Grade A because we located no replication and the studies used a single scenario type.
That the effect generalises to ongoing commercial relationships is Grade C, and the design section explains why that is a real gap rather than a formality.
The explanation for the asymmetry is ungraded, because the reasoning we give for it is our own and appears in no source we obtained.
A Note On Method
Everything here is verified to August 2026.
We obtained the published abstract in full from PubMed, the publisher's own record and four further reproductions, all identical[1][2][3][5][6].
We did not obtain the full text, and therefore report no effect sizes, no sample sizes, and no detail of the manipulations.
What we know about the experimental design comes from an announcement published by one author's university[4], which is not a neutral source and which we identify at each use.
The Bayesian illustration explaining the asymmetry is entirely ours, uses probabilities we invented, and appears in no paper.
This article reviews behavioural research. It is not legal, regulatory, professional conduct or crisis communications advice.
The Finding
The first clause of the result.
Kim, Ferrin, Cooper and Dirks published Removing the Shadow of Suspicion: The Effects of Apology Versus Denial for Repairing Competence- Versus Integrity-Based Trust Violations in the Journal of Applied Psychology, 89(1), 104–118, in February 2004[1].
The result: trust was repaired more successfully when mistrusted parties apologized for violations concerning matters of competence but denied culpability for violations concerning matters of integrity[1].
Two observations, ours.
This is an interaction, not a main effect. Neither response is better in general. What the study found is that the right response depends on which kind of violation is alleged, which means a firm with a single standard response to complaints is getting one of the two categories wrong.
And the two categories are routinely conflated in practice. A client who says "you got this wrong" and a client who says "you weren't straight with me" are usually handled by the same procedure, by the same person, in the same tone.
The Second Half
The clause that changes everything, and the one that gets dropped.
"(b) had apologized for violations when there was subsequent evidence of guilt but had denied culpability for violations when there was subsequent evidence of innocence."[1]
Three observations, ours.
The study manipulated what the evidence later showed, independently of the response given. So the design tests not only which response works, but which response works given how the facts turn out.
And the result is intuitive once stated: apologising ages well if you did it, and denying ages well if you did not. What matters is the match between the response and the eventual truth.
Which means the finding is not about rhetoric at all. It is about whether your response will still look right when the facts are known, and the research measured trust after that point rather than immediately.
What The Second Half Rules Out
Reading the two clauses together. This section is our own.
Taken alone, clause (a) reads as advice to deny integrity allegations. Taken with clause (b), it cannot.
Three consequences.
A denial by someone who did it is a denial that will meet subsequent evidence of guilt, and that is the combination the study found performed worse.
So the usable statement is conditional on innocence: where an integrity allegation is false, denying it repaired trust better than apologising for it did.
And the corresponding statement for competence is conditional on fault: where you did make the error, apologising repaired trust better than denying did.
Our own summary: the finding rewards being accurate about what happened, and the type of violation determines what accuracy sounds like. That is a considerably more useful and more defensible result than the version that circulates.
What Was Actually Tested
The setting, which is a real limit on how far this travels.
An announcement from one author's university describes the paradigm: participants were asked to rate the likelihood that they would hire the applicant and if so, what level of job responsibilities should be entrusted to the applicant[4]. The same announcement frames the research around public figures then facing allegations[4].
This is our only source on the design and it is an institutional press announcement, not the paper.
Three consequences, ours.
The judgment measured is a hiring decision about a stranger. Participants had no prior relationship with the person they were assessing, and nothing to weigh the allegation against.
An existing client relationship is a different situation entirely. A client of eight years has hundreds of prior observations, and the study's participants had none.
And the outcome is a stated intention rather than behaviour over time. Whether the same pattern holds when someone decides in six months whether to renew is not something these studies measured.
We think the finding is still worth knowing, because the underlying logic does not obviously depend on the setting. But the extension to ongoing commercial relationships is ours, not the study's.
The Mediation
One further result, which tells you what is moving.
The abstract reports: "the interactive effects of violation type and violation response on participants' trusting intentions were mediated by their trusting beliefs."[1]
Two observations, ours.
The chain runs through beliefs. The response changed what people believed about the person, and the changed belief changed what they intended to do. It is not a matter of the response being satisfying or the tone being right.
Which means the practical question is evidential rather than emotional: what does this response cause a reasonable observer to conclude about you? That framing is the one the rest of this article uses.
Why The Two Are Not Symmetric
Our own explanation. The study reports the finding; this reasoning appears in no source we obtained and we present it as an account you may test rather than as a result.
The asymmetry follows from how diagnostic a single observation is, and that differs sharply between the two traits.
Competent people fail regularly. Good advisers miss things, lose files and misjudge deadlines, because volume and complexity guarantee it. So one failure is weak evidence of incompetence.
Honest people almost never appear dishonest. A clear integrity incident is rare among honest people, so one such incident is strong evidence.
Two consequences.
The same single event updates a reasonable observer's belief far more in the integrity case than in the competence case, and it does so for a good reason rather than because observers are being unfair.
Which means denial is not a rhetorical preference in the integrity case. It is the only response that addresses what the observer has just concluded, because conceding the fact concedes the inference.
The Asymmetry In Numbers
Making that precise. Every figure below is invented by us to demonstrate a mechanism, comes from no study, and is illustrative only.
Suppose 20 percent of advisers are genuinely weak. A weak adviser misses a deadline in a given year 80 percent of the time; a strong one does so 25 percent of the time. After one missed deadline, a Bayesian observer moves from 20 percent to about 44 percent confidence that the adviser is weak.
Now suppose 5 percent of advisers would knowingly mislead a client. Such a person does so in a given year 60 percent of the time; an honest adviser appears to do so 2 percent of the time. After one such incident, the same observer moves from 5 percent to about 61 percent.
Two observations.
The competence belief roughly doubles. The integrity belief goes up more than twelvefold.
And that gap is not produced by prejudice or by observers being harsh. It falls straight out of the base rates, and any reasonable person processing the same information lands in the same place.
And The Same In Reverse
The consequence for recovery, which is the part that matters commercially. Ours, following the same logic.
Run the argument backwards, on good news rather than bad.
A strong adviser demonstrates competence constantly. Each demonstration is weak evidence, but there is an endless supply of them, and they accumulate.
A dishonest person behaves honestly nearly all the time. So honest conduct after an integrity allegation is almost no evidence at all, because it is exactly what the dishonest and the honest both produce.
Two consequences.
A competence problem can be worked off. Deliver well for eighteen months and the belief moves, because each delivery is informative in the right direction.
An integrity problem largely cannot be. Behaving honestly afterwards does not distinguish you from the alternative, which is why the allegation has to be defeated rather than outlived.
That, in our reading, is why the study found what it found, and it is the single most useful idea in this article.
Worked Off Versus Defeated
The practical distinction, stated once. Ours.
Two categories, two entirely different strategies.
A competence violation is repaired by time and performance. Acknowledge it, fix it, and then produce a long run of evidence pointing the other way. The apology matters because denying a demonstrable error adds an integrity problem to a competence one.
An integrity violation is not repaired by time and performance, because subsequent good behaviour carries almost no information. It requires the specific allegation to be shown to be untrue.
Two consequences.
The resource that fixes one is patience. The resource that fixes the other is evidence, and evidence is gathered at the time or not at all.
Which produces a documentation implication a firm can act on today, independent of whether it ever faces an allegation: the records that would defeat an integrity claim are the ones created before anyone makes one.
Classifying The Accusation
The first move, and the one most likely to be skipped. Ours.
Before responding to a complaint, the question is which category it is in. Two tests we would suggest.
Would a competent, honest person in your position have produced this outcome? If yes with bad luck or high volume, it is a competence question. If no honest person would have done it, it is an integrity question.
What would the client conclude about your character if it were true? Competence allegations conclude something about your capability. Integrity allegations conclude something about whose side you are on.
Two observations.
The client's framing, not yours, determines the category. You may regard a late filing as an administrative failure while the client regards it as evidence you deprioritised them for a larger account. Those are different accusations and the second is the integrity one.
And the categorisation should be done before drafting a response, because the default response fits only one of them.
The Cases That Sit In Between
Being honest about where this framework stops. Ours.
Three situations the two-category scheme does not cleanly handle.
Mixed allegations. A single complaint may contain both, and the study tested them separately. We do not know from this research what happens when a response has to address both at once.
Competence failures with integrity readings. A repeated error can stop being read as capability and start being read as indifference, which converts the category without any new fact.
And partial fault. The study's second finding concerns subsequent evidence of guilt or innocence, which are clean states. Most real disputes are neither, and a response that is accurate about a partial fault is not the same as either an apology or a denial.
We do not have answers to these from the research we obtained, and we would rather say so than extend the framework past its evidence.
In A Professional Firm
The application, and it is our own extension to a setting the study did not examine.
Competence allegations in an accounting or advisory practice: a missed filing, a computational error, a position that did not survive review, advice overtaken by a change nobody caught.
Integrity allegations: favouring one client over another, a conflict not disclosed, a fee not explained, a recommendation the client believes served the firm rather than them.
Three observations.
The second list is far more damaging and, on our reasoning above, far less recoverable, even though the first list is what firms have procedures for.
The second list is also where the professional conduct rules concentrate, which is not a coincidence. Independence, conflicts and disclosure obligations exist precisely because the profession's value depends on integrity beliefs that a single incident can destroy.
And a firm's complaints procedure is usually built for the first list, because competence complaints are more common. That is exactly the design error the study points at.
The Default Error
The specific mistake we think this research identifies. Ours.
The professional default under criticism is to apologise. It defuses, it is courteous, and it is what training in client care encourages.
Three consequences.
Applied to a competence complaint, that default is correct and matches the finding.
Applied to an integrity allegation that is false, it is on this evidence the worse of the two responses, and it also risks being read as concession of the fact.
And the default is applied without classifying, which means the error is systematic rather than occasional. A firm that apologises for everything will handle its most damaging category of complaint in the way the research suggests works least well.
We repeat the caution from the top: how to respond to any actual allegation is a matter for your professional body, your insurer and your lawyer, and this observation about a general default is not advice about a particular case.
What An Apology Is Not
A distinction worth drawing before anyone acts on the above. Ours.
Nothing in this research suggests being cold or unsympathetic about a client's experience.
Three separable things frequently bundled into the word apology.
Acknowledging the client's experience. That something went wrong for them, that it caused them difficulty, and that you take it seriously. This is not an admission of anything.
Accepting responsibility for the fact. That you did the thing alleged.
Accepting the inference. That the thing alleged means what the client thinks it means about you.
Two observations.
The study's contrast is between apology and denial of culpability[1], which is the second and third of those, not the first.
So on our reading, the first is available in every case. You can take someone's concern seriously and still say clearly that what they believe happened did not happen.
What To Do
Classify before responding. Competence and integrity allegations are different problems, and on this evidence the same response cannot serve both.
Use the client's framing, not yours. What you regard as an administrative slip may be read as evidence about whose side you are on, and the reading determines the category.
Read both halves of the finding. Denial worked where subsequent evidence showed innocence; apology worked where subsequent evidence showed guilt. The result rewards accuracy, not tactics.
Expect competence damage to be repairable and integrity damage not to be. Good performance afterwards is strong evidence in one case and almost no evidence in the other.
Create the records now. An integrity allegation has to be defeated with evidence, and the evidence that defeats it is generated before anyone alleges anything.
Do not let a repeated competence failure become an integrity question. Repetition changes what the client concludes without any new fact arriving.
Separate acknowledging an experience from accepting a fact. The study contrasts apology with denial of culpability, and taking someone's concern seriously is neither.
Check whether your complaints procedure was built for one category. Most are built for the common one, which is not the damaging one.
And for any real allegation, go to your professional body, insurer and lawyer first. That sequence takes precedence over everything above.
The Limits Of This Analysis
Several caveats matter. This article reviews behavioural research and is not legal, regulatory, professional conduct or crisis communications advice. How a regulated firm responds to an allegation engages conduct rules, notification duties, insurer requirements and potentially privilege, none of which is addressed here and all of which take precedence. Nothing in this article suggests denying something you did; the study's own second finding conditions the result on what subsequent evidence showed. Everything is verified to August 2026. We obtained the published abstract in full but not the full text, and report no effect sizes, no sample sizes and no detail of the manipulations. Our only source on the experimental design is an announcement published by one author's university, which is not a neutral source. The paradigm concerned hiring judgments about a stranger, not ongoing commercial relationships, and measured stated intentions rather than behaviour over time; the extension to client relationships is ours, not the study's. We located no replication and did not obtain any of the related work described in citing indexes. The Bayesian illustration is entirely ours, uses invented probabilities, comes from no study, and demonstrates a mechanism rather than estimating anything. The explanation for the asymmetry, the worked-off versus defeated distinction, the classification tests, the hard cases, the professional services application and the default-error section are all our own reasoning, not findings. Roughly twenty-two years of subsequent literature was not reviewed.
Frequently Asked Questions
What did the study find?
Is this saying to deny things I did?
Why should the two kinds differ?
Can trust be rebuilt by performing well afterwards?
How well does this transfer to a real client relationship?
What should I actually do if we get a complaint?
References
- Kim, P. H., Ferrin, D. L., Cooper, C. D., & Dirks, K. T. (2004). Removing the Shadow of Suspicion: The Effects of Apology Versus Denial for Repairing Competence- Versus Integrity-Based Trust Violations. Journal of Applied Psychology, 89(1), 104–118. DOI 10.1037/0021-9010.89.1.104, published abstract via PubMed, on two studies having been conducted to examine the implications of an apology versus a denial for repairing trust after an alleged violation; on results revealing that trust was repaired more successfully when mistrusted parties apologized for violations concerning matters of competence but denied culpability for violations concerning matters of integrity; on trust having been repaired more successfully when parties had apologized for violations when there was subsequent evidence of guilt but had denied culpability for violations when there was subsequent evidence of innocence; and on supplementary analyses revealing that the interactive effects of violation type and violation response on participants' trusting intentions were mediated by their trusting beliefs. Note: we obtained the published abstract in full; we did not obtain the full text, and report no effect sizes, sample sizes or manipulation detail. pubmed.ncbi.nlm.nih.gov
- American Psychological Association PsycNet record for Kim and colleagues (2004), reproducing the abstract identically and confirming the citation, journal, volume, issue, pages and DOI. Note: the publisher's own record, used as authoritative confirmation of the abstract text. psycnet.apa.org
- Semantic Scholar record for Kim and colleagues (2004), reproducing the abstract including the mediation finding, and carrying indexed descriptions of related trust-repair literature, including work demonstrating that once breached, cooperation can be reestablished and that actions as well as explanations and apologies can augment the process, and work indicating that when harm-doers apologized, victims refrained from severe aggression against them. Note: a bibliographic record with third-party indexed summaries; we obtained none of the related studies described. semanticscholar.org
- Washington University in St. Louis, institutional announcement on the 2004 paper, on the research addressing whether a person can regain trust under clouds of suspicion left in the wake of legal problems; on this depending upon the match between how they respond to allegations and the extent to which the alleged offense is perceived to involve their integrity or their competence; and on participants having been asked to rate the likelihood that they would hire the applicant and, if so, what level of job responsibilities should be entrusted to the applicant. Note: an institutional press announcement from one author's own university, not a neutral source and not the paper. This is our sole source on the experimental design. source.washu.edu
- Social Science Research Network record for Kim and colleagues (2004), listing the paper as Marshall School of Business Working Paper No. MKT 06-09, confirming the journal citation and reproducing the abstract, and recording the authors' institutional affiliations. Note: a working paper repository record, used as further independent confirmation of the citation and abstract. papers.ssrn.com
- Publisher record for Kim and colleagues (2004) hosting the first page of the paper, confirming the authors' departmental affiliations at the University of Southern California, the State University of New York at Buffalo, the University of Miami and Washington University in St. Louis; recording that the study was supported by a grant from the University of Southern California's James H. Zumberge Research and Innovation Fund; and carrying third-party text describing the finding as indicating that offering an apology effectively repairs trust while denying the situation can further damage trust in ability trust violations. Note: a publisher record; we obtained the first page only, and the characterisation quoted is third-party text rather than the paper's own. researchgate.net
This article reviews behavioural research and is not legal, regulatory, professional conduct or crisis communications advice. How a regulated firm responds to an allegation engages obligations that take precedence over anything here. Nothing in this article suggests denying something you did. The study was not obtained in full and no effect sizes or sample sizes are reported. The only source on its design is an institutional announcement from an author's university. The paradigm involved hiring judgments about strangers; the extension to client relationships is the authors' own. The Bayesian illustration uses invented probabilities and comes from no study.