What Forgiving Someone Actually Changes

Somebody has probably already told you to forgive. It is the most common advice given to a person who has been hurt by their partner, and it usually arrives as though the benefit were obvious and the only real question were whether you are big enough to do it.

There is genuine evidence behind forgiveness, including one of the largest randomised trials ever run on the subject. The evidence is narrower than the advice, and the gap between the two is worth understanding before you decide anything.

What the study did

In 2024 a team reported a randomised trial of a self-directed forgiveness workbook run across six sites in five countries: Colombia, Hong Kong, Indonesia, South Africa and Ukraine. It is preregistered, and after exclusions it analyses 4,598 people, which makes it very large for this field.

Participants were randomly assigned either to receive the workbook immediately or to receive it after a two-week delay, and everyone was measured at two weeks. The workbook takes two to three hours and walks a person through a structured sequence for forgiving a specific person who hurt them. The person who did the hurting is never contacted and takes no part.

Behind this trial sits a pooled analysis from 2014 covering 54 published and unpublished reports of forgiveness interventions, which found much the same thing across 2,323 people: more forgiveness than no treatment, by a standardised difference of 0.56, and more than active alternatives by 0.45. Two independent bodies of evidence, pointing the same way.

What it found

The workbook worked on what it was aimed at.

People who got it immediately reported substantially less unforgiveness toward the person who hurt them than the people still waiting, by a standardised difference of 0.53, with a confidence interval running from 0.47 to 0.58. With a sample this size that interval is tight, and the finding is not in much doubt.

Depression symptoms and anxiety symptoms moved as well, by 0.22 and 0.21. Real, consistent across the sites, and roughly a third to a half the size of the effect on forgiveness itself.

Now the part that changes how all of it should be read.

What was on the outcome sheet, and what was not

Every outcome in this trial is one person reporting on their own internal state. Unforgiveness, forgiveness, forbearance, depression, anxiety, flourishing.

No relationship outcome was measured. Not relationship satisfaction, not relationship quality, not reconciliation, not whether the other person changed their behaviour, not anything reported by a partner. The 2014 pooled analysis has the same blind spot: its outcomes were forgiveness, depression, anxiety and hope, and nothing else.

There is also a detail about who these people were forgiving. In the pilot data for the 2024 trial, the person who had caused the hurt was a romantic partner or spouse in about one case in six. Friends were more common. This is a body of research about being hurt by somebody, not specifically about being hurt by the person you live with.

The separation is deliberate rather than an oversight. Researchers in this area draw a hard line between forgiveness, which happens inside one person, and reconciliation, which is a decision two people make about how they will live. You can do the first without the second. Some people should.

An honest reading of the limits

The comparison group got nothing. People in the delayed condition sat on a waiting list. There was no attention-matched or placebo condition, so the effect cannot be separated from the effect of being given something, being paid attention to, and knowing what the study wanted.

Nobody was blinded and everything is self-report. Look at the shape of the results: the largest effect by some distance is on the exact construct the workbook spends two hours teaching people to think about and then asks them to rate. The wellbeing effects are a fraction of it. That pattern is consistent with a real benefit. It is also consistent with people reporting what they were just taught.

The people who built the intervention ran the trial. The developer of the workbook is an author on the trial evaluating it. This is normal in psychotherapy research and it reliably inflates effects, and a reader should know it before weighing the number. The same problem runs through the 2014 pooled analysis, where a large share of the trials tested models developed by its own authors.

Two weeks. That is the whole follow-up window. The authors say plainly that a waiting-list design cannot evaluate whether anything lasts. Nothing here supports a claim about where somebody is in a year.

The samples are volunteers, not patients. Community members recruited through universities, congregations and community leaders across five countries, median age 26 and about three-quarters women. Distress levels are not clinical.

The workbook was translated but not culturally adapted, and the size of the effect differed significantly across sites. A single pooled number describes no particular place well.

And none of it says whether you should. A trial shows that a procedure moves a measure. It cannot tell you that forgiving is owed, deserved or right in your situation. Nobody is under an evidentiary obligation to forgive anybody.

What this means for the fight you keep having

If the hurt is old and you are the one still carrying it, this evidence supports something quite specific: structured work on forgiveness tends to reduce what you are carrying, and tends to lift your mood somewhat along with it. That benefit appears to be available whether or not the other person participates, because in these trials the other person never does.

What the evidence does not support is the version of this advice that gets used as leverage. Forgiving has not been shown here to repair a relationship, to make anybody safe, or to be a step you owe. If the person telling you to forgive is the person who hurt you, that is a negotiating position rather than a finding.

The last note is about sequence, and it cuts both ways. Because forgiving does not require the other person to have changed, it stays available to someone whose partner never will. For the same reason, it is not evidence that they have.

Where a session with Elena fits

Elena is an AI coach. She is not delivering a forgiveness intervention, and this article would be dishonest if it implied otherwise — what was tested above is a structured protocol with a preregistered trial behind it, and a conversation is not that.

What a session can do sits earlier and smaller. Most people arrive at this subject with three questions fused together: what happened, what they still feel about it, and what they are going to do about the relationship. Elena is useful for pulling those apart and working out which one you are actually stuck on, before anybody decides anything about forgiving.

One limit worth stating plainly. This research studies offences that are over. If what happened is still happening — an affair that is continuing, or any pattern of control, intimidation or fear — the findings above do not apply to it, and neither does a session. That situation needs a human professional who can see the whole picture, and if it is yours, that is where to start rather than here.