We Ran 105 Loyalty Tests. One in Three Partners Engaged.

We Ran 105 Loyalty Tests. One in Three Partners Engaged.

Most writing about infidelity relies on surveys. Someone is asked whether they have ever cheated, and they answer — or don't. It is the best tool researchers have, and it has an obvious weakness: it depends entirely on people telling the truth about the thing they are most motivated to hide.

We have a different kind of data. Not better, not scientific, but different in a way we think is worth publishing.

Over the course of our work we have completed 105 loyalty tests. In each one, a trained agent made contact with someone whose partner had asked us to find out how they would respond. Nobody was asked to report their own behaviour. We simply recorded what happened.

35 of those 105 people engaged in a conversation their partner would not have wanted them to have. That is one in three.

Below is the full breakdown, what we think it means, and — just as importantly — what it does not.

---

The results

Outcome	Count	Share
Engaged in inappropriate conversation	35	33%
Replied, but declined the approach	28	27%
Blocked the approacher immediately	6	6%
Never responded at all	36	34%
Total	105	100%

Four outcomes, and they split almost evenly into thirds: a third engaged, a third actively said no, and a third said nothing.

---

What "engaged" actually means

This is the definition everything else depends on, so it is worth being precise.

We counted a test as engaged when the person moved past ordinary politeness into conversation of a romantic or sexual nature: flirting, complimenting the approacher's appearance, asking to move to another platform, proposing to meet, or concealing the exchange.

We did not count someone as engaged for replying. A person who answers "sorry, who is this?" has not failed anything. A person who says "thanks, but I'm with someone" has arguably passed in the most meaningful way available to them.

The line we drew is behavioural rather than emotional. We cannot see what someone felt. We can see what they typed.

---

The number that surprised us most

Thirty-three percent is the headline, but it is not the figure we found most striking.

Of the 69 people who responded at all, 35 engaged — 51%.

Once you exclude everyone who never replied, it is a coin flip. Slightly more than half of the people who chose to enter a conversation with an attractive stranger ended up in one their partner would not have sanctioned.

That reframes the finding considerably. The question stops being "how many people cheat" and becomes something narrower and more uncomfortable: *among those willing to start the conversation, how many keep going?* On our data, about half.

---

Why this is not a cheating rate

We want to be direct about this, because it is the easiest thing to get wrong when quoting a number like ours.

A loyalty test does not measure infidelity. It measures response to an unsolicited approach. Those are different things, and conflating them would be dishonest.

For comparison, the General Social Survey — analysed by Wendy Wang at the Institute for Family Studies — finds that 20% of ever-married men and 13% of ever-married women report having had sex with someone other than their spouse while married. Those are national, representative, decades-deep figures. Ours are none of those things.

So why is our engagement rate higher than the reported infidelity rate?

Three reasons, all of which cut against reading our number as a cheating rate:

Engaging in a conversation is not an affair. Most of our 35 did not meet anyone. Flirting with a stranger online and conducting an affair are separated by a great deal of distance, and our data captures the first, not the second.

Opportunity was handed to them. In ordinary life, most people are not approached by an interested stranger on a given Tuesday. We manufactured the opportunity. Published research on the gap between intention and action in romantic relationships suggests the two diverge substantially — people who would say yes if asked are not the same population as people who go looking. Our test only speaks to the first group.

Our sample is not random — it is the opposite. Every one of these 105 tests was commissioned by someone who already suspected something. That is a profoundly self-selected group. If suspicion carries any diagnostic value at all, our population should show a higher rate than the general public by definition.

That last point is the most important limitation in this article, and we would rather state it ourselves than have someone else point it out.

---

The 36 who never replied

The largest single group did nothing at all, and it is the hardest group to interpret honestly.

Silence is not proof of loyalty. Someone might not have seen the message. They might not check that account. They might have been busy, or suspicious of a stranger's profile, or simply uninterested in conversation with anybody that week.

Silence is also not nothing. A non-response is a non-engagement, and for the partner who commissioned the test, the practical outcome is the same: nothing happened.

We report these 36 as their own category rather than folding them in with the people who actively declined, because those are genuinely different events. Merging them would inflate a "passed" figure to 70% and make our data look tidier than it is. A person who explicitly said "I have a girlfriend" has told you something. A person who never opened the message has told you nothing.

---

The six who blocked immediately

Six people blocked the approacher outright, without engaging at all.

Six is a small number and we are not going to build a percentage argument on it. But it is worth naming as a distinct outcome, because it is the only one in the dataset that represents active refusal rather than passive decline. These were people who did not merely fail to take an opportunity — they closed the door on it.

If you commissioned a test and received this result, you received the clearest answer our method is capable of producing.

---

What this data cannot tell you

We would rather be useful than impressive, so here is the honest accounting.

105 is a small sample. Large enough for the broad shape of the result to mean something, far too small to slice into demographics. We are not going to tell you how the rate varies by age or gender, because at this sample size those splits would be noise dressed up as insight.

It is not a controlled study. There is no control group, no randomisation, no pre-registration. Approaches varied by platform and by agent. This is operational data from a business, not a peer-reviewed experiment, and it should be weighted accordingly.

It measures a moment, not a relationship. A single test captures one response, on one day, to one approach. People are inconsistent. Someone who engaged might not have done so a month earlier, and someone who ignored the message might have replied on a worse week.

It cannot tell you what your partner would do. This is the one that matters most to anyone reading this while wondering. Our aggregate says nothing about any individual. A 33% engagement rate across 105 strangers is not a 33% chance that your partner would engage.

---

What we think it does show

With all of that said, three things survive the caveats.

A third is a lot. Whatever the methodological limitations, 35 real people engaged in conversations they would not have wanted their partners to read. That is not a rounding error or a statistical artefact. It happened, 35 times.

Suspicion is not nothing. Every one of these tests was commissioned by somebody who already sensed something was wrong. In a third of cases, the approach found something. That does not vindicate suspicion in general — two-thirds of these people were wrong — but it suggests the instinct is not baseless either.

Most people did not engage. Seventy of 105 did not. We are in the business of loyalty testing, and our own data says the most common outcome is that nothing happens. We would rather publish that than pretend otherwise.

---

If you are considering a test

Read the result you are likely to get before you commission one.

On our numbers, the single most likely outcome is that your partner does not respond at all — and you will be left interpreting a silence. The second most likely is that they respond and decline. Only a third of the time does a test produce the clear, difficult answer people imagine when they contact us.

That is worth sitting with. A loyalty test is good at producing one kind of evidence and poor at producing reassurance, because the absence of engagement is always ambiguous. If what you actually want is to stop feeling anxious, this is an imperfect instrument for the job, and we would rather say so.

If what you want is to know how someone responds when an opportunity is placed in front of them, that is precisely what these 105 tests measured.

---

Methodology

What we did. Each test involved a trained agent initiating contact with the subject through a platform specified by the client, using a profile constructed for that test. The agent made a friendly approach and allowed the subject to determine the direction of the conversation. Agents did not escalate beyond the subject's own lead.

What we recorded. A single outcome per test, assigned from the four categories above. Where a conversation was ambiguous, it was classified by the content of the subject's messages rather than the agent's interpretation of tone.

Sample. 105 completed tests. Incomplete tests — those cancelled, refunded, or where contact could not be established — are excluded.

Privacy. No personal information about any subject or client appears in this article, and none will. All figures are aggregate counts. We do not retain case details beyond a short operational window, and nothing in this dataset can be traced to an individual. The people tested did not consent to being tested, which is precisely why the only thing we will ever publish about them is a number in a column.

Limitations. Stated in full above rather than buried here. The headline figure is a response rate within a self-selected population, not an infidelity rate, and should not be cited as one.

---

*We will update this article as the sample grows. If you are a researcher or journalist and want to discuss the methodology, our team is contactable through our support page.*