We Read 482 Opening Messages. Two Thirds Were Under Twenty Characters.

Advice about opening messages is almost entirely assertion. Be playful. Ask a question. Reference something in their profile. Never say hey. It is all plausible and almost none of it is checked against anything.

We have the messages, so we checked.

What we counted

Every conversation on Arranged where a real member sent the first message, which is 482 of them once we had cleaned the data. For each one we asked a single question: did the other person ever send anything back.

That is a deliberately low bar. Not a date, not a good conversation. Just whether the message landed.

Overall, 44 of 482 got an answer. Nine per cent.

One thing about that number before anything else. Our first pass said eleven per cent, and then we noticed a single account had sent the identical line to 239 people. Removing that one sender moved the headline by two points and changed several of the findings below. Every number here excludes it, and it is a decent illustration of how easily a handful of heavy senders distort a dataset this size.

Length: the one clear finding

This was the strongest and least ambiguous pattern.

Openers under twenty characters — hi, hey, hello, and slight variations on them — were answered 6% of the time.

Openers between twenty and fifty characters, which is roughly one short sentence, were answered 15% of the time.

Longer than that it flattens out. Fifty to a hundred characters came in at 10%, and over a hundred at 14%, on smaller numbers where the difference is not meaningful.

So the return is not on length. It is on clearing a very low bar at all. One sentence roughly doubles your odds against one word, and writing three paragraphs buys you nothing further.

The part worth sitting with: 308 of the 482 openers were under twenty characters. Two thirds of everyone sending a first message is sending a greeting and nothing else. The bar is low and most people are walking under it.

Photos mattered more than words

We expected this to be a footnote. It was the largest effect we found.

Senders with three or more photos on their profile were answered 11% of the time. Senders with one or two were answered 7%.

Senders with no photo at all were answered zero times out of twenty-one attempts.

Twenty-one is a small number and we would not claim zero is the true rate. But nobody in our data has ever replied to a message from an empty profile, and the gap between one photo and three is larger than the gap between a one-word opener and a written sentence.

Which is a slightly deflating conclusion for anyone workshopping their opening line. The message is being read next to your profile, and the profile is doing more of the work.

Two findings that contradict the usual advice

Copy-and-paste was not punished

Openers that had been sent to more than one person were answered 9% of the time. Openers written once, for one person, were answered 10%.

That is not a real difference on these numbers, and it is not what we expected to find. The standard advice is that recycled openers are obvious and get ignored.

We are genuinely unsure what to make of it. The most likely explanation is that a reused sentence and a bespoke one look identical to somebody reading a first message from a stranger — the recipient has no way of knowing it was reused. If that is right, personalisation may matter later in a conversation and not at the very start.

We would not go so far as to recommend copying and pasting. We would say we could not detect a penalty for it.

Asking a question did not help

In our first pass, openers containing a question mark did worse than openers without one. That result was heavily affected by the high-volume sender described above, and once removed the sample became too small to say anything honest about.

So: we cannot confirm the standard advice to always ask a question, and we cannot refute it either. We are noting it because leaving out the inconclusive results is how data gets used to say whatever the author already believed.

What we would actually tell someone

Put three photos up before you worry about your opening line. It is the biggest lever in our data and it takes two minutes.

Write one sentence rather than one word. Not a paragraph — a sentence. That single step is worth more than every piece of phrasing advice combined.

Then stop optimising the opener. The difference between a good sentence and a great one is not visible in our numbers, and the difference between having photos and not having them is enormous.

If you are answering rather than sending, our data on what happens when messages cross a border found the opposite of what most apps assume about distance.

What this does not tell you

We only see replies that happen on Arranged. If somebody moves a conversation to another app immediately, which is common and occasionally a warning sign, it counts here as no reply.

Nine per cent also sounds bleak, and reply rates on dating apps generally are. It is not a number to take personally at the individual level.

And this is one app, several hundred messages, over a period of weeks. We will publish an update when the sample is larger, including if it reverses.

Frequently asked questions

What is a good reply rate on a dating app?

Ours is 9% overall, rising to about 15% for people who write a full sentence and have a few photos up. Published figures elsewhere vary hugely depending on who is counting and what they count, so we would rather compare our own numbers to each other than to somebody's marketing page.

Does the first message really matter that much?

Less than most advice implies. Clearing the one-word bar matters. Beyond that, our data says your photos are doing more work than your phrasing.

Is it worth writing something personalised?

We could not detect a benefit at the opening-message stage, which surprised us. It is very likely to matter once a conversation is running — we just cannot show that from first messages.

Why did you exclude one sender?

A single account had sent the same line to 239 people, which is more than a third of the dataset from one person. Leaving it in meant reporting that account's results as though they were everybody's.

How many messages was this?

482 opening messages from 124 different senders. Only two senders accounted for more than twenty openers each, so after that exclusion it is reasonably spread rather than a handful of people.

Will these numbers change?

Almost certainly. We are a young app and the sample is small. We also changed how message notifications work shortly after pulling these figures, which will affect reply rates on its own — so the next version of this post may well disagree with this one.