Tools

WhatsApp Chat Analyzer: Private, No Upload

A man and a woman hold hands.
Chat X‑Ray is a free tool that analyzes an exported WhatsApp chat in your browser: who writes how much, who starts conversations, how fast each of you replies, when you write and how often you say “we.” Nothing is uploaded, and no AI reads your messages.

Example

55messages
49:51Percent of messages, Alex and Sam
16 of 18Active days

This is what your result looks like. The numbers come from a made-up chat between Alex and Sam; your own result appears further down once you drop your export.

  1. Export the chat
  2. Drop the file
  3. Read the numbers

Only for chats you write in yourself. Analyzing another person’s chat without their knowledge is a breach of trust and can be against the law. Help with control or abuse

Your chat stays on your device. Neither we nor an AI reads it. The messages of the people you chat with belong to them too: share the numbers, not their words. Check it yourself: load this page, turn on airplane mode, then drop the file. The analysis still runs. Meant for people 18 and older.

Example loaded: 55 messages from 2 people. Drop your own file above.

With this choice, Chat X‑Ray also shows values per person.

Example, from Alex’s side

55 messages

From September 1, 2026 to September 18, 2026, with messages on 16 of 18 days.

Fewer than 100 messages: with this little text, these numbers swing a lot.

Who contributes what
MetricYouSam
Messages27 (49%)28 (51%)
Words112121
Words per message4.34.5
Longest message (words)88
Conversation starts12 (57%)9 (43%)
Questions129
Emoji78
We-words23
Thanks53
Affection34
Apologies01
Late night21
Reply time (median)5 minfrom 17 replies5 minfrom 19 replies

The last message in the file is from Sam.

7 daysLongest streak
16 of 18Active days
7 AM and 6 PMBusiest hours (tied)
TuesdayBusiest day
5.5%Late night (midnight to 4:59 AM)
40.7 hrLongest gapSeptember 5 to 7, 2026
8 wordsLongest messageits length only, never its text
0.45We-words per I-word (our measure)
September 18, 2026Last day with messages

When you write

All times come from the phone that exported the chat. For someone in another time zone, “3 AM” may be their morning.

Show as a table

When each person writes

You
Sam

Each row is scaled to that person’s own busiest hour: it shows when someone writes, not how much.

Messages per day

The days with the most messages:

  • September 1, 2026: 8 messages
  • September 4, 2026: 6 messages
  • September 7, 2026: 5 messages
  • September 5, 2026: 4 messages
  • September 16, 2026: 4 messages
Every day (16 days with messages)

Most frequent words thanks 6 · tomorrow 4 · good 4 · night 4 · tonight 4 · morning 3 · today 3 · bring 3 · sleep 2 · tell 2

Your most frequent words thanks 4 · tomorrow 3 · good 3 · night 3 · sleep 2

Most frequent words from Sam tonight 3 · morning 2 · tell 2 · trail 2 · earlier 2

Most frequent emoji ❤️ 3 · ☀️ 2 · 😘 2 · 😂 2 · 🙌 1

Your most frequent emoji ☀️ 1 · 🙌 1 · 😂 1 · 🌙 1 · 😌 1

Most frequent emoji from Sam ❤️ 2 · 😘 1 · 🍕 1 · 😂 1 · 👍 1

Only on this screen, never on the share card.

We-talk

We-words: You 2 (1.8 per 100 words), Sam 3 (2.5 per 100 words). Across the whole chat there are 0.45 we-words for every I-word (5 we, 11 I); this ratio is our own measure.

Thanks, affection, apologies

Messages with thanks, affection or an apology: You 5, 3 and 0; Sam 3, 4 and 1.

Conversation starts and questions

Conversation starts: You 12, Sam 9. Questions: You 12, Sam 9.

What studies of couples say

All three studies looked at romantic or married couples. For family, friends or work, they establish nothing.

A meta-analysis of 30 studies of romantic couples found that we-talk is usually linked to how well the relationship is doing, a partner’s we-talk more strongly than one’s own. An association, not a cause; it didn’t study friends or family. More on the research

In married couples, more negative and less positive expression in conversations went with lower satisfaction and separation (Gottman and Levenson). They measured conversations, not chats, and Chat X‑Ray doesn’t compare your numbers with any rule of thumb. More on the research

Driver and Gottman observed newlywed couples and saw early support that small everyday moments matter; strictly observational, the direction is open. That a conversation start in a chat is such a moment is our reading. More on the research

One small step for this week

At least 40 percent of the conversation starts and questions come from you, each counted on its own. If this connection matters to you: this week, ask Sam how something on their mind right now is going.

If something about this chat weighs on you, none of these numbers matter. Here is where to find help.

What was read
  • Lines in the file: 56
  • Messages: 55, of which media and attachments (photos, videos, voice messages, documents, locations, polls): 1, deleted: 1
  • Notices from WhatsApp (encryption, group events, calls), not counted: 1
  • Lines before the first message, skipped: 0

Detected: Android export, dates with the month first, 12-hour clock.

Does your chat start later than you expected? The first message in this file is from September 1, 2026. If your chat began earlier, the beginning is missing from the file. Why that can happen

Not in the file: photos, voice messages and other media (an export without media leaves only a placeholder). View once messages: WhatsApp’s export help says they aren’t exported. Disappearing messages: the help for exporting on WhatsApp Web says they aren’t exported, and whatever has already disappeared is no longer in the chat anyway. Where messages are missing, the gap looks like a silence.

Numbers about a chat, not a verdict on a relationship. Help lines in the US, the UK, Ireland and Germany are listed below.

Share

The card shows only shared numbers: how the messages split, how many there are and on how many days, the longest streak and how many we-words there are per I-word. It calls you “You” and the other person “Person B.” No names, no messages.

Saves a calendar file (.ics) for 30 days from now on your device. The reminder is in your calendar only once you open the file. The entry is only called “Take another look”: it names neither this tool nor the chat and holds no numbers, no names and no link. Anyone who can see your calendar still sees it.

If you feel unsafe

If you feel controlled or unsafe, a local helpline or support service can help. None of the numbers on this page matters more than your safety. In immediate danger, call your local emergency number.

  • US: National Domestic Violence Hotline, free, confidential, 24/7: 1.800.799.SAFE (7233), or text START to 88788 (message and data rates may apply)
  • US: 988 Suicide & Crisis Lifeline, free, call or text: 988
  • UK: National Domestic Abuse Helpline, freephone, 24 hours: 0808 2000 247
  • UK and Ireland: Samaritans, free from any phone: 116 123
  • Ireland: Women’s Aid National Freephone Helpline, 24 hours: 1800 341 900
  • Germany: Hilfetelefon Gewalt gegen Frauen, for women affected by violence, around the clock and free, with interpreters in 18 languages, from a phone in Germany: 116 016
  • Germany, in German: TelefonSeelsorge, free of charge within Germany. From abroad, it says calls are free from a mobile phone with a German SIM card in another EU country, and that it can be reached by internet telephony with a number registered in Germany; it lists +49 800 111 0 111 and +49 800 111 0 222 and says not every way of dialing works from every country.

For anyone under 18 who lands here anyway:

  • US: love is respect, the National Teen Dating Abuse Helpline, for young people: 1.866.331.9474, or text LOVEIS to 22522
  • UK: Childline, free and confidential, for anyone under 19: 0800 1111
  • Ireland: Childline, free, 24 hours: 1800 66 66 66

If someone can see your phone, your browser history or your calendar, look for help from a device only you use, or call one of the lines above directly. love is respect warns that browser history can be monitored without your knowledge and can never be wiped completely.

How to export a WhatsApp chat: iPhone, Android, Mac and Web

To get a chat as a file, you export it. Here is how WhatsApp’s Help Center describes it, with the menu names exactly as WhatsApp writes them (retrieved October 21 to 25, 2026).

On iPhone

  1. Open the individual or group chat.
  2. Tap the contact’s or group’s name at the top, then “Export chat.”
  3. If the chat contains media, WhatsApp asks you to choose “With all media,” “With selected media…” or “Without media.” For Chat X‑Ray, “Without media” is all you need.
  4. Choose where the file goes. WhatsApp gives Message, “Save to Files” or Notes as examples. In “Save to Files,” it’s best to choose “On My iPhone.” Apple describes that as the place for files saved locally on your device, while files in “iCloud Drive” sync across your devices. Then you can drop the file here.

“Your chat history will be exported as a ZIP file.” Chat X‑Ray opens the ZIP itself; you don’t need to unzip anything. Any text editor can show the .txt inside it, too.

On Android

  1. Open the individual or group chat.
  2. Tap the three-dot menu (in the Help Center: “More options”), then “More” and “Export chat.”
  3. Tap “Without media.” With “Include media,” WhatsApp also attaches the most recent media, which Chat X‑Ray doesn’t need.
  4. Choose where the file goes; WhatsApp gives Quick Share, Gmail or Chat as examples. WhatsApp hands the file to the app you pick here, and that app decides where it ends up. The simplest is a files app and a folder you can find again. If you email it to yourself, the chat also sits with your email provider afterward.

On a computer: the Mac app and WhatsApp Web

The WhatsApp app for Windows doesn’t export chats: the Windows tab of WhatsApp’s export article says the feature isn’t available on that platform. For the Mac app, the Mac tab says: “Click your contact’s name or group name > Export chat.” Then choose “Without media.” “Your chat history will be exported as a ZIP file.” We haven’t tested Chat X‑Ray with an export file from the Mac app, because we don’t have a real one; if it can’t read such a file, use the export from your phone.

WhatsApp Web can export chats: open the chat, click the three-dot menu, then “Export chat.” Then choose “All messages” or “Custom date range” and click “Export.” This also creates a ZIP file. “Media isn’t included, unlike exporting chats on your mobile device.” “View once and disappearing messages won’t be exported.” We haven’t tested Chat X‑Ray with an export file from WhatsApp Web, because we don’t have a real one. If it can’t read such a file, use the export from your phone.

Very long chats

WhatsApp’s current Help Center article on exporting states no cap on the number of messages. An older version of WhatsApp’s help, from 2020, said: “When exporting with media, you can send up to 10,000 latest messages. Without media, you can send 40,000 messages.” Press reports repeat that figure, for example Android Police in 2024, while How-To Geek wrote in 2024 that you can get up to 100,000 recent messages. We can’t check any of these numbers. If the first message in your result comes later than your chat began, the beginning is missing from the file. On WhatsApp Web you can then export earlier periods one at a time (“Custom date range”), and Chat X‑Ray reads each file on its own, though we haven’t tested it with a real export file from WhatsApp Web.

Export or backup?

A backup is something else. WhatsApp uses it to restore chats inside the app: “WhatsApp chats and media can only be restored from the last successful backup that occurred on your Android.” An export is a text file to read, and it can’t be imported back. For Chat X‑Ray, you need the export.

If “Export chat” is missing

One possible reason is “Restricted chat.” When it’s on in a chat, nobody in it can export the chat; WhatsApp says people in it “Can’t export the chat.” In individual chats either person can turn it on and off again, in groups the admins. WhatsApp describes its purpose: “You can limit how other users can share your messages and media outside of WhatsApp.” and “This update helps prevent the use of unofficial clients or AI agents in chats.” It is replacing the earlier setting “Advanced chat privacy”; chats that still have that setting on don’t switch over by themselves. If the other person turned on Restricted chat, that’s a clear answer: they don’t want the chat to leave the device. Respect it.

What the file contains, and what it doesn’t

In the exports we know, each message starts on a new line with date, time, name and text; longer messages continue on the lines below. Photos, voice messages and other media show up only as placeholders in an export without media. “View once messages will not be exported.” And: “Your chat history can’t be re-imported because it’s a text file and not a backup file.” Chat X‑Ray knows the placeholders WhatsApp writes in English and German, including shared locations and live locations. It recognizes polls in English iPhone exports, the only kind we found a real export line for. Add to that placeholders and deletion notices in other languages, each with evidence. For iPhone in Spanish, Brazilian and European Portuguese, Italian, French, Dutch and Turkish, that is the text WhatsApp itself stores for them in its iPhone app, as a published research data set holds it; for Spanish, European Portuguese, Italian and French also real iPhone export lines. For Android, it is real export lines in French, Italian and Spanish, and for Portuguese the word list of another analysis program. If an export looks like another language, it tells you. In an iPhone export without any English or German notice, a line Chat X‑Ray doesn’t know counts as a message of an unknown kind from the person whose name it carries, so the shares stay right; that can also be a call, and “What was read” says how many there are. On Android, unknown media may count as text messages, and the word lists fill up with that language’s filler words.

WhatsApp’s own statistics in the app

Many guides talk about “WhatsApp statistics.” WhatsApp’s help pages describe two usage screens for that, not an analysis of a chat: under Settings > “Storage and data” > “Network usage,” WhatsApp shows how much data you’ve used, and in a chat under “Manage storage,” how much space its media take up: “This will show you how much space the chat's media is taking up, with files sorted from largest to smallest.” A German dpa guide (t-online, 2024, in German) adds that Network usage also counts how many messages were sent and received, and that resetting the statistics sets every counter back to zero. The guide describes one total there; it doesn’t describe it split by chat or by person. Who writes how often in a particular chat, who starts and how fast replies come is only in the export.

WhatsApp chat analysis: what your result shows

At the top: how many messages the file holds and which dates it covers. That matters, because an export shows what is in this chat, not your whole history. Below that, for two people side by side, who contributes what: messages, words, words per message, the longest message, conversation starts, questions, emoji and we-words. Values that say more about a single person, such as thanks, affection, apologies, late-night messages, reply times and most frequent words, appear per person only once you’ve picked who you are. For two people, a small block sets the last 30 days against the time before: who writes how much, who starts, and, once you’ve picked who you are, how fast replies come. It shows whether something is changing right now, and it is what you compare when you drop the chat again after the reminder. In groups, Chat X‑Ray shows up to 8 of the most active people and names roles; words per message and each person’s longest message open below them. Groups get no late-night values, no hours and no reply times per person: the analysis is for your own reflection, not for judging colleagues or anyone else. With your own file you can pick months, the last 12 for example, and count how often the two of you wrote a word or phrase. The tiles sum up the chat, the grid shows which days and hours people write, the monthly bars show how that changes over time, and the day chart shows when the most was going on.

None of these numbers is a verdict. Writing more doesn’t necessarily mean loving more, and a slower reply may just mean a full day. Chat X‑Ray doesn’t try to tell whether someone is cheating or lying; counting messages says nothing reliable about that. The numbers show you what is worth talking about. What Chat X‑Ray doesn’t do: no PDF of your chat and no reading by an AI.

How Chat X‑Ray works

Your browser reads the file line by line, in the background for large files so the page stays usable; you can cancel at any time. A new message starts with a date, a time and a name; lines without that start belong to the message before. Chat X‑Ray reads exports from iPhone and Android with a 12-hour or 24-hour clock. Whether a date puts the day or the month first is decided once for the whole file: it takes the reading with the fewest impossible dates and the smallest jumps from day to day. If that isn’t clear-cut, it tells you and offers the other reading. It doesn’t read a file as a chat when every sender is named like a log entry. Under “What was read” you can see what it detected and what it skipped, including whether any characters in the file couldn’t be read.

Every number follows a fixed rule. The rules are our own definitions, not findings of a study, and they are all here.

How Chat X‑Ray counts: the definitions

Messages
Every line that starts with a date, a time and a name, plus the lines below it. Notices from WhatsApp (encryption, group events, calls) don’t count. In an iPhone export without a single English or German notice, a line Chat X‑Ray doesn’t know counts as a message of an unknown kind from the person whose name it carries, if that person writes otherwise too; that can be media, but also a call. The percentage is the share of all messages, in whole percent, rounded so that they add up to 100.
Media and attachments, deleted
Placeholders like “<Media omitted>,” “image omitted” or “document omitted,” shared locations, polls from English iPhone exports, and messages Android exports as “null” count as messages, but without words, questions or emoji. So do notices like “This message was deleted.” Add to that the placeholders and deletion notices WhatsApp writes on iPhone in Spanish, Portuguese, Italian, French, Dutch and Turkish, and some on Android.
Words
Anything between two spaces that contains at least one letter or digit. Chinese and Japanese characters, which are written without spaces, count as one word each and stay out of the word lists.
Words per message
All of a person’s words in text messages divided by the number of their text messages; media and deleted messages don’t count. Rounded to one decimal place.
Longest message (words)
The most words in a single text message, per person and as a tile for the whole chat, counted like the words. Chat X‑Ray shows only the length, never the text.
Conversation starts
The first message in the file and every message after at least 6 hours of silence, even when it is a late answer: a reply that comes 6 hours or more after the message before it counts as a new conversation start, and after more than 12 hours it no longer counts as a reply either (see reply time). The percentage works as for messages.
Questions
Messages with a question mark; a question mark inside a link doesn’t count.
Emoji
Each emoji counts once, even when it’s made of several characters, such as a flag, an emoji with a skin tone, or a family. Characters like ©, ™, arrows or ♥ count only in their emoji form, the way phone keyboards insert them. The basis is the Unicode Consortium’s emoji list.
We-words
Each person’s we-words (“wir”, “uns”, “unser”, “unsere”, “unserem”, “unseren”, “unserer”, “unseres”, “unsre”, “unsrem”, “unsren”, “unsrer”, “unsres”, “we”, “us”, “our”, “ours”, “ourselves”, “we’re”, “we’ve”, “we’ll”, “we’d”, “let’s”), plus their share per 100 words, rounded to one decimal place.
Thanks
Messages with one of these words: “danke*”, “dank”, “bedanke*”, “merci”, “thx”, “thnx”, “thanx”, “thank*”. An asterisk means the word starts this way. Each message counts at most once, for affection and apologies too.
Affection
Messages with “liebe dich”, “lieb dich”, “hab dich lieb”, “hab dich so lieb”, “hab dich auch lieb”, “hab dich sehr lieb”, “hdl”, “hdgdl”, “hdal”, “love you”, “love u”, “luv you”, “luv u”, “ily”, “ilu” or one of these emoji: ♥️ ❣️ ❤️ 💋 💌 💏 💑 💓 💕 💖 💗 💘 💙 💚 💛 💜 💝 💞 💟 🖤 😍 😘 😻 🤍 🤎 🥰 🧡 🩵 🩶 🩷 🫶 (the heart suit ♥ only in its emoji form).
Apologies
Messages with “sorry*”, “sorri”, “sry”, “entschuldig*”, “verzeih*”, “tut mir leid”, “tut mir so leid”, “tut mir echt leid”, “tut mir wirklich leid”, “my bad”, “apolog*”, “forgive me”.
Late night
Messages from midnight to 4:59 AM, by the time in the file, which is the time of the phone that exported the chat; for someone in another time zone, those hours may be their morning. The tile shows their share of all messages, rounded to one decimal place. Groups get no late-night value per person.
Reply time (median)
A reply is a message from a different person at most 12 hours after the message before it; the gap is the reply time. The median is the middle value: half the replies came faster, half slower. Shown in whole minutes, from one hour on in hours with one decimal place. Below each median: how many replies it rests on; with fewer than 5 replies Chat X‑Ray shows no median, only “too few replies.” Only for two people: groups get no reply time per person, because in a work group it would be a measure of availability.
Last 30 days / before
For two people: the last 30 calendar days up to and including the day of the last message in the file, next to everything before. For each person, messages and conversation starts with their share and, once you’ve picked who you are, the median reply time with the number of replies. Whether a message is a conversation start or a reply is decided on the whole chat, as above. The block appears as soon as there are messages before those 30 days.
Values per person
Thanks, affection, apologies, late night, reply time, the hours, each person’s most frequent words and emoji, how often each person wrote a phrase you searched for (in groups only how often you wrote it), and who wrote last appear only once you’ve picked who you are above. Until then, Chat X‑Ray shows these values only for everyone together. In groups there are no late-night values, no hours, no reply times and no phrase counts per person, with chosen months too.
Active days
Days with at least one message, out of all days from the first message to the last.
Longest streak
The most days in a row with at least one message each.
Busiest hour, Busiest day
The hour and the weekday with the most messages; on a tie, up to three are named, plus how many more are tied (the week starts on Monday).
Longest gap
The largest gap between two messages in a row, with both its days; on a tie, the earlier one. Under two days in hours, otherwise in days.
Last day with messages
The latest day with a message in the file.
We-words per I-word (our measure)
All we-words divided by all I-words, rounded to two decimal places. This page’s own measure, not the studies’. I-words: “ich”, “mir”, “mich”, “mein”, “meine”, “meinem”, “meinen”, “meiner”, “meines”, “i”, “me”, “my”, “mine”, “myself”, “i’m”, “i’ve”, “i’ll”, “i’d”.
When you write
Messages by weekday and hour. The stronger a square, the more messages, in four steps up to the busiest square; a square without messages is only outlined. The table below it has the totals per weekday and per hour.
When each person writes
Each person’s messages by hour, every kind of message; each row is scaled to that person’s own busiest hour. Only for two people, and only once you’ve picked who you are.
Messages per month
Every calendar month from the first message to the last, including months without messages; shown from two months on. When single months don’t fit the width, Chat X‑Ray groups them into quarters or years; the table still lists every month.
Messages per day
Every day from the first message to the last as a bar; days without messages stay empty. When the days don’t fit side by side, each bar sums several consecutive days, and a line below says how many. Below it, the 5 days with the most messages, earlier first on a tie, and every day with messages on request.
Most frequent words
Words of 3 or more characters, counted and shown in lower case, without filler words like “and” or “und,” without we- and I-words and without plain numbers; the 10 most frequent in the whole chat, and each person’s 5 most frequent. On a tie, what came first comes first.
Most frequent emoji
The 5 most frequent emoji, counted as above; once you’ve picked who you are, each person’s 5 most frequent too.
Roles in this group
In chats with three or more people, among the up to 8 with the most messages: Starts the most conversations, Asks the most questions, Sends the most emoji. On a tie: whoever has more messages, then whoever wrote first. No late-night role and no fastest-replier role: the analysis is for your own reflection, not for judging anyone.
One small step for this week
Only for two people, and only once you’ve picked who you are. A share counts only when it rests on at least 20 of what it shares out (20 messages for the message share, 20 conversation starts for the share of starts), and it is the share as the table shows it, in whole percent. First: if at least 65 percent come from the other person, in the whole chat or in the last 30 days, or they wrote at least 5 messages in those 30 days and you wrote none (this holds in a short chat too), Chat X‑Ray never suggests more contact: you don’t have to keep up, and if it feels like pressure or control, the note links to help. Then: if at least 65 percent come from you, or you wrote at least 5 messages in the last 30 days and they wrote none, it names that and doesn’t suggest more contact either. The sentence names only the shares that meet one of these rules. When photos, voice messages and other media are what lift a share to 65 percent, the sentence says so and gives the share of text messages, rounded as in the table. If none of these rules applies and the whole chat has fewer than 20 messages or fewer than 20 conversation starts, it only says the chat is still too short for a suggestion. Otherwise: if your share of conversation starts is below 40 percent, the step is about starting; otherwise, if the two of you wrote at least 10 questions and your share of them is below 40 percent, about asking; otherwise, the same for thanks (at least 10 messages with thanks), about thanking; otherwise it’s a check-in. If the last message is more than 30 days old (by your device’s date; not applied to the example above), there is no suggestion. With chosen months, all of this covers those months. A suggestion, not a standard.
Months
Only for your own file that spans at least two months: from one month to another, or “Last 12 months” (counting back from the chat’s last month), “This year” (your device’s calendar year) or “Everything.” Chat X‑Ray then reads the chat again with only the messages of those months: the first of them is a conversation start, the last 30 days end with their last message, and this week’s step covers those months too. The people stay those of the whole file: a group stays a group, with every group rule, even if only two of them write in those months, and anyone who writes nothing there shows zero. When the file begins is still said for the whole file under “What was read.” The card shows the numbers of those months without naming them.
How often did you write …?
How often a word, a phrase or an emoji appears in the text messages: in total, per month (with the first and the last month) and, once you’ve picked who you are, per person; in groups only how often you wrote it. Case and accents don’t matter (“Café” finds “cafe”), but two dots over a letter do (“schön” doesn’t find “schon,” “naïve” doesn’t find “naive”); only whole words in that order (“coffee” doesn’t find “coffeehouse”), and links don’t count. An emoji counts in its emoji form, as for the emoji above. Two in a row count twice, overlapping ones don’t. With chosen months, only those count. None of it is stored, and none of it goes on the card.

The word lists count English and German, exactly as listed above. They miss irony, typos, other languages and context: “thanks for nothing” counts as thanks. The times come from the file, as your phone wrote them. Chat X‑Ray doesn’t correct for daylight saving time: a reply time or a silence across the change can be off by an hour.

What the research says, and what it doesn’t

We-talk. Karan, Rosenthal and Robbins pooled 30 studies of romantic couples in a meta-analysis. They looked at each person’s we-talk separately, one’s own and the partner’s. Overall, we-talk was positively associated with how well the relationship and the people in it were doing, and a partner’s we-talk usually more strongly than one’s own: “Partner use of we-talk was generally more strongly related to relationship functioning than own use.” The authors also write that we-talk sometimes relates negatively. It remains an association, not a cause. That’s why Chat X‑Ray shows each person’s we-words; the ratio of we-words to I-words across the whole chat is our own measure, not the studies’.

Positive and negative. Gottman and Levenson studied 73 married couples in 1983 and 1987. Couples they classed as nonregulated, using their balance theory of marriage, showed more negative and less positive emotional expression in conversations, were less satisfied, were in poorer health at the second time point, and more often considered or went through separation. They measured conversation behavior, body signals such as finger pulse, health and satisfaction, not words in chats; this too is an association, not a prediction for any one couple. The well-known rule of five positive for every negative is often attributed to Gottman’s work on conflict conversations from 1993 and 1994; the 1993 abstract doesn’t state it, and we couldn’t check the 1994 book. Chat X‑Ray doesn’t compare your numbers with it.

Small everyday moments. Driver and Gottman observed 49 newlywed couples, in a 10-minute dinnertime interaction in what the study calls a “seminatural setting” and in a 15-minute conflict discussion. They looked at how playfulness and enthusiasm in everyday life relate to humor and affection during conflict, and saw early support for the importance of small daily moments for couples. The study was strictly observational and therefore correlational: “These findings provide preliminary support for the importance of daily moments in couple relationships, but this research was strictly observational and therefore correlational, so further research is necessary to determine direction of causation.” Our reading, not shown by the study: a conversation start or a question in a chat is one of those small moments.

Other relationships, groups, single writers. All three studies looked at romantic or married couples. For a chat with your mother, a friend or a colleague they establish nothing, and for group chats or a single writer Chat X‑Ray shows numbers only, no research.

Why no AI? Many people look for a reading by ChatGPT or another AI. We decided against it, because a chat belongs to two people: a cloud AI would also send the other person’s messages to someone else’s server, without anyone asking them. What you get here are numbers your browser computes, and fixed sentences that give every research finding its limit.

Example: the made-up chat of Alex and Sam

The example above is a made-up chat over 18 days with 55 messages. Alex writes 27, Sam 28. Alex started 12 of 21 conversations; both reply after a median of 5 minutes. There are 0.45 we-words for every I-word, and the longest streak is 7 days in a row.

What does that mean? On its own, nothing final. Alex starting more often might come down to habit. It gets interesting when the two of you recognize something in it, or when a pattern changes over months. That’s what the reminder is for: export again in a few weeks, and the “Last 30 days / before” block shows the time since today.

Privacy: what happens to your file

Your chat stays on your device. Neither we nor an AI reads it. The messages of the people you chat with belong to them too: share the numbers, not their words.

Your browser reads the file on your device. Chat X‑Ray sends nothing from your file: no upload, no server, no AI, and it stores none of it, not the months you pick and not a phrase you search for either. Close the tab and the file is gone from this page. You can check this with airplane mode. Your browser also makes the share card and the calendar file on your device; the calendar entry is only called “Take another look” and holds nothing from your chat, not even a link to this tool. The card shows only shared numbers: you as “You” and the other person as “Person B.” Whether and with whom you share it is up to you; posting it publicly makes a private chat a little bit public, which is why there are no names on it.

We still want to know whether the tool gets used, but without content: if you agreed to statistics on this website, we count four events: a file read, a card made, a card shared, a reminder saved. We give each one only the tool’s name and the page’s language; as with every page view it measures, Google Analytics adds the page address, the page title, the time, device and browser details, the random ID from the analytics cookie, and the approximate location Google derives from the IP address. No number, name or word from your chat reaches us or Google.

The other person’s messages are their data. It’s best to look at the result together, or to ask first. Chat X‑Ray is meant for people 18 and older. More in our privacy policy.

Frequently asked questions

How do I analyze a WhatsApp chat?

With Chat X‑Ray above: export the chat in WhatsApp without media and drop the file here. Your browser counts messages, conversation starts, reply times, questions and we-talk. No account, no upload, no AI.

How do I export a WhatsApp chat?

On iPhone: open the chat, tap the contact’s or group’s name, “Export chat,” then “Without media.” On Android: open the chat, tap the three-dot menu, “More,” “Export chat,” then “Without media.” In the Mac app: click the contact’s or group’s name, “Export chat,” then “Without media.” On WhatsApp Web: the menu, “Export chat,” “All messages” or “Custom date range,” then “Export.” WhatsApp’s help says the Windows app doesn’t offer it. You don’t need to unzip the file.

Is my chat uploaded, and does it work offline?

No upload: your browser reads the file on your device, and Chat X‑Ray sends nothing from your file. Once the page has loaded it needs no connection, for the months, the phrase count and the card too. Check it yourself: turn on airplane mode, drop the file, and the analysis still runs.

Why is “Export chat” missing?

One possible reason: “Restricted chat” is on in that chat. WhatsApp says nobody in the chat can export it then. In individual chats either person can turn it on and off again, in groups the admins. If the other person did, that’s a clear answer, and it deserves respect.

Can ChatGPT analyze a WhatsApp chat?

Yes, if you upload the file, but then every message the other person wrote ends up with an AI provider too, without anyone asking them. That’s why Chat X‑Ray has no AI: the numbers come from your browser, and nobody but you sees the chat.

Does the other person see that I exported our chat?

WhatsApp’s export instructions mention no notification to the others in the chat; whether that holds for every version, we can’t check. It’s fair to tell them anyway, because their messages belong to them too.

Who should text first in a relationship?

There’s no rule, and this tool doesn’t make one up. It shows each person’s share of conversation starts. A lopsided share over months is a good reason to talk, not proof that someone cares less.

Does Chat X‑Ray work for group chats?

Yes. In chats with three or more people it shows up to 8 of the most active people and names roles like “starts the most conversations.” It shows no late-night values per person in groups: the numbers are for your own reflection, not for judging colleagues. The share card then shows only numbers about the whole group, no names.

Sources

All sources retrieved October 21 to 25, 2026 and archived in the project. For the studies we read the abstracts; the full texts are not freely available, and this page says nothing beyond the abstracts. · Karan A, Rosenthal R, Robbins ML: Meta-analytic evidence that we-talk predicts relationship and personal functioning in romantic couples. Journal of Social and Personal Relationships 36 (2019), pages 2624 to 2651, doi:10.1177/0265407518795336 · Gottman JM, Levenson RW: Marital processes predictive of later dissolution: Behavior, physiology, and health. Journal of Personality and Social Psychology 63 (1992), pages 221 to 233, PubMed 1403613 · Gottman JM: The roles of conflict engagement, escalation, and avoidance in marital interaction: A longitudinal view of five types of couples. Journal of Consulting and Clinical Psychology 61 (1993), pages 6 to 15, PubMed 8450108 · Driver JL, Gottman JM: Daily marital interactions and positive affect during marital conflict among newlywed couples. Family Process 43 (2004), pages 301 to 314, PubMed 15386956 · WhatsApp Help Center: “How to export your chat history” (Android, iPhone, Web, Mac, Windows), “About restricted chat”, “How to back up your chat history”, “How to free up storage on WhatsApp”, “About WhatsApp charges”, and the older export article from 2020 in the Internet Archive · Apple Support: “Find files on your iPhone or iPad” · Press on the network usage screen: t-online (2024, in German) · Press on the export limit: Android Police (2024), How-To Geek (2024), netzwelt (2021, in German), t-online (2022, in German) · Help lines: National Domestic Violence Hotline, 988 Lifeline, National Domestic Abuse Helpline, Samaritans, Women’s Aid (Ireland), love is respect, Childline (UK), Childline (Ireland), German federal ministry on the Hilfetelefon (in German), Hilfetelefon on phone counseling (in German), TelefonSeelsorge (in German) · Export lines from other people’s real exports, as evidence for placeholders in other languages, locations and polls (facts only, no code copied): whatsparser, WhatsApp-Analyzer, WhatsApp-Chat-Converter, whatsapp-converter, chat-export, whatsapp-text-analysis2, kapta-mobile, whatsapp-chat-parser-website, WhatsR, rwhatsapp · WhatsApp’s own iPhone texts, from the EvidenceData research data set (the texts only): Spanish, Brazilian Portuguese, European Portuguese, Italian, French, Dutch, Turkish · Real iPhone export lines from the test data of whats_json (facts only, no code copied): Spanish, Portuguese, Italian, French · Word lists of other analysis programs for Spanish and Portuguese placeholders (the words only, no code copied): WhatsApp-Analyzer, whatsapp-group-wrapped, whatsapp-wrapped, qualichat · Unicode Consortium: emoji-data.txt, version 18.0, for recognizing emoji. Recorded verification runs are listed in our verification record (in German).

Published October 25, 2026

Not a relationship test and not a diagnosis: Chat X‑Ray counts what an export file contains and cannot see what happens between the lines. The studies cited describe associations, not causes. The content is general information and does not replace individual medical, legal, tax or financial advice. Not affiliated with WhatsApp or Meta.

More from the garden

You can find every calculator and tool under Tools. For the bigger picture of your life areas, relationships included: the wheel of life. To turn this week’s step into a habit: the habit calculator. And before a difficult conversation: the breathing exercise timer.

Wild Garden Editorial Team

Content and tools on Wild Garden are published by the Wild Garden Editorial Team. How we research, check and correct errors is set out in our editorial standards.

Read next