Want texts like the winners in this study?
Paste your conversation and get options tuned to the actual moment.
We Analyzed 383,886 AI-Generated Texts. Here's What People Actually Send.
TL;DR
We analyzed 383,886 AI-generated texts across 39,553 real conversations. People send short (median 63 characters), mostly skip questions (91% of winners have none), and reject 83% of everything we generate. The biggest surprise: more flirting happens on Instagram than on every dating app combined.
Key Takeaways
- •The median text people choose to send is 63 characters — one sentence
- •76% of sent texts are under 80 characters
- •Chosen texts ask questions 1.5× more often than rejected ones — but 91% of winners ask none
- •Chosen and rejected texts are nearly identical in length: content decides, not word count
- •Instagram hosts more of these conversations (10,759) than every dating app combined
- •People reject 83% of AI suggestions — the value is options, not one "perfect" answer
The dataset: 383,886 texts across 39,553 real conversations
Since January 2026, 15,509 people have pasted real conversations into SmoothRizz and generated 383,886 reply options. For each one we know a simple, brutal fact: did the person copy it to send, or swipe it into the void?
Over the last 90 days, users chose 7,274 texts and rejected 36,514 — an 83% rejection rate. That sounds harsh until you realize it's the point: the right reply depends on the person, so we generate options and let you pick. This study is about what the picks have in common.
We've published smaller studies before — 4,381 dating requests (what people ask for) and which tones get replies. This is the biggest one yet, and it answers a different question: out of everything the AI writes, what do people actually send?
Finding 1: The 63-character rule
The median text people choose to send is 63 characters. That's this long:
"bold words from someone who replies in one-word sentences"
76% of everything people send is under 80 characters. Almost nobody sends paragraphs. This matches what messaging research has found for years — an academic analysis of student text messages found the typical message runs about 14 words (Lyddy et al., Journal of Computer-Mediated Communication), and business-texting data shows messages under 100 characters get dramatically higher response rates (TextUs).
If your draft needs a scroll bar, it's not a text — it's a letter. Cut it.
Finding 2: Questions are overrated
Every texting guide says "end with a question." Our data says: sort of, but mostly no.
Texts people chose to send contained a question 1.5× more often than rejected ones (9.2% vs 6.2%). Questions help. But flip that number around: 91% of winning texts don't ask anything.
The pattern in the winners is a confident statement that invites a response without demanding one — a tease, an observation, a callback to something they said. "What's your favorite restaurant?" is an interview. "you strike me as a person with a strong taco opinion" is a conversation.
Finding 3: Length doesn't decide winners — content does
Here's the honest finding most studies would bury: chosen texts and rejected texts are almost the same length. Median 63 characters for winners, 64 for losers.
Everything the AI generates already follows the short-text rule — so length isn't what separates the text someone copies from the four they reject. What separates them is whether the message builds on what the other person just said (what improv calls "yes, and") instead of pivoting to a new topic. You can't fix a text by trimming it. You fix it by making it about them.
That's also why single-suggestion AI tools underperform: the right reply depends on the moment. Our flirty text generator reads the conversation and gives you several angles on it, because the data says you'll reject most of them — and that's the feature, not the bug.
Finding 4: The flirting happens on Instagram, not dating apps
When users tell us where their conversation is happening, the leaderboard isn't close:
| Platform | Conversations (90 days) |
|---|---|
| Instagram DMs | 10,759 |
| iMessage | 3,034 |
| 2,837 | |
| Facebook Messenger | 1,048 |
| Hinge | 1,003 |
| Discord | 808 |
| Bumble | 604 |
Instagram alone hosts more of these conversations than every dating app combined. The dating apps are where you match; the DMs are where it either becomes a date or dies. That gap — the unstructured, no-prompts, no-bio DM conversation — is exactly where people run out of script and come looking for help.
Before/after: what the winners look like
Composite examples of the pattern the data keeps rewarding: short, no interview questions, built on what they said.
They teased you about your gym selfies
✗"haha yeah I go a lot I guess. do you work out? what gym do you go to? is it nice?"
✓"someone had to raise the average in this chat"
46 characters. No question. Builds on their tease instead of changing the subject.
Instagram story reply, they posted concert footage
✗"That looks so fun!! Who did you see? How was it? I love concerts too"
✓"front row energy from the person who said they hate crowds"
60 characters — right at the median. References something real about them.
Dry "lol" after your best material
✗"lol so anyway how was your day today? do anything fun?"
✓"putting that in my quote wall. you’re buying the first round for it"
Statement, not interview. Gives them something to push back on.
Want the same treatment on your own conversation? That's literally what the SmoothRizz app does — paste the chat, get options, send the one that sounds like you.
Frequently Asked Questions
How long should a flirty text be?
Short. Across 7,274 texts people chose to send, the median length was 63 characters — roughly one sentence. 76% were under 80 characters. Long paragraphs read as try-hard in dating chats.
Should every text end with a question?
No. Texts people chose to send contained a question 1.5× more often than rejected ones (9.2% vs 6.2%), but 91% of winning texts had no question at all. A confident statement that gives the other person something to react to usually beats an interview-style question.
Where do people actually need texting help — dating apps or DMs?
DMs. In our 90-day sample, Instagram accounted for 10,759 conversations — more than every dating app combined. iMessage (3,034) and WhatsApp (2,837) also beat Hinge (1,003) and Bumble (604). Most flirting happens after you leave the dating app.
Does message length predict whether a text gets picked?
Barely. Chosen texts had a median of 63 characters; rejected texts, 64. At the same length, what the text says — whether it builds on the other person’s message — is what separates winners from losers.
How was this data collected?
SmoothRizz generated 383,886 reply options across 39,553 real conversation sessions from 15,509 users between January and July 2026. We compared the texts users chose to copy and send against the ones they rejected. All data is aggregated and anonymous.
Keep reading
Get weekly rizz tips
Real templates, data-backed insights, and new features — straight to your inbox.
No spam. Unsubscribe anytime.
Your next text, backed by 383,886 data points
Paste the conversation, pick your spice level, and send something that sounds like you on a good day.