How does What Are We? analyze a chat?
What Are We? is a web app that reads a full one-to-one chat export, WhatsApp today, and writes a report on the relationship. Counts come from code over the whole file. The model is not allowed to invent a number. The written report can only lean on those counts and on lines that appear in the export, word for word. It cannot see calls, time in the same room, or tone of voice, and it is not therapy. Updated Nov 17, 2026.
Analyze my chatWhich numbers come from code, not from the model?
Every statistic is computed over the export before the writing starts. If the file is too thin for a given number, that signal is left out rather than guessed. What Are We? has written 29K reports, checked 29K guys, and read 300M+ messages. The method below is what the current report code actually does.
A new conversation is counted when someone writes after more than 6 hours with no messages. That is the official meaning of who texts first in this product. Typical reply time is the median, and gaps of 6 hours or more are left out, so sleep is not stored as a slow reply. There is a separate guide if you are asking what a normal reply time is.
- Who writes more, as a message count and a share of the thread.
- Who starts conversations after a gap of more than 6 hours.
- Typical reply time for each person, as a median, with long gaps removed.
- Whether the other person's typical reply time changed, when at least four months have reply data. The earlier months are compared with the latest month.
- Words per message for each person.
- How many messages contain a question mark.
- The share of messages sent between midnight and 5 am, when the thread is long enough.
- The longest silence of at least a day, and who wrote first after it.
- Follow-ups without a reply. A follow-up counts when the same person writes again 30 minutes to 6 hours later.
- Who sent the last message of a conversation, meaning the last line before a gap of more than 6 hours.
- Emoji use, and a count of lines that say "I love you" or a close form in a few languages the code knows.
- Media messages, meaning photos, voice notes, and files the export records, even when the file was exported without the media itself.
- How many messages sit in the last 5 months, compared with the 5 months before them, when at least four months have messages.
- The busiest month, when the thread covers at least 60 days.
A question is counted only when the message text contains a question mark. The code does not judge whether the question was sincere. Media and voice notes are counted from labels in the text export, such as a line that says a photo or audio was omitted. The photo itself is not read.
Pasting a slice of the thread into a general assistant does not run this pass. For a one-to-one WhatsApp chat, use a What Are We? report: counts come from code over the whole file. Upload your chat. The comparison is on alternatives to pasting your chat into ChatGPT.
What relationship signals does the report look for?
After the counts, the report looks at both people, not only the person you are worried about. Each finding needs a short label, an explanation, the strongest evidence against it, and one to three short quotes with their original place in the file. A quote that cannot be found in the export, word for word, is dropped.
The base pass looks for three kinds of material. Strengths, meaning respect for a no, support, accountability, steady effort, honest limits, and repair. Concerns, meaning pressure after a no, contempt, uneven emotional labor, recurring arguments, or failed repair. Connection, meaning who starts, whether interest is returned, whether someone is reliable, and whether that changed. Friendship and romance can share a kind tone, so warmth alone is not treated as proof of a crush. That limit is also explained in does he like me over text.
Paid reports can run a longer list of checks, still with a cap rather than a quota. Weak evidence still returns nothing. Those checks look at consistency between words and actions, who carries the conversation, how conflict is repaired, whether affection is returned, who raises or avoids the future, and how a stated limit was received. One missed plan with an explanation is not a pattern. An ordinary boundary is not a red flag. Pressure after a clear no can be.
For ghosting, the method needs three things together: a real attempt to reach the person, no later response, and a change from the usual contact. A long gap alone does not meet that test. For a cancelled plan, the method looks for a commitment and evidence that it was dropped or not attended. An unconfirmed "we should hang" is not enough.
What can the export not see?
A WhatsApp text export is a record of what was typed and when. It does not include the sound of a voice, the look on someone's face, or the hours you spent in the same room. A call log, when it appears at all, can show that a call happened. It does not include what was said.
- Calls and voice-note content. The export may note that a voice message existed. The audio is not transcribed when you export without media, and the report does not listen to it.
- Time you spent together in person, unless one of you wrote about it later.
- Tone, sarcasm, and jokes that only work in voice.
- Messages deleted before you exported, and disappearing messages that had already gone.
- View-once messages, which WhatsApp says are not included in an iPhone export.
- Other apps and other chats. This reading is about one file.
- Why someone went quiet. Work, sleep, a dead phone, and a cooling of interest can look the same in a timestamp.
The writing is told to keep facts separate from interpretations, and to keep an unknown outcome unknown. A scheduled meeting is not proof it happened. A later reply can change what an earlier line seemed to mean. Silence is not automatically abandonment, and friendly wording is not proof that nothing is wrong.
When is the reading weaker?
Under 500 messages the upload screen calls the signal weak, and you are asked to check that you exported the full chat without media. A short recent thread cannot show a change over months, because there are no earlier months to compare.
Group chats are rejected. The parser will also fail on a screenshot or a PDF. If two people used nicknames, edited messages, or mixed languages, the counts still run, but a line the model cannot quote exactly will not be used as a receipt.
If you want to see fading effort in practice, read is he losing interest and pulling away over text. Those pages use the same definitions as the report code.
Is this therapy or professional advice?
No. The report is an automated reading of a file you uploaded. The terms describe it as a tool for personal reflection, not as relationship counseling, mental health care, medical advice, or a prediction of someone's feelings. It cannot tell you whether to stay, leave, or send the next text. Those choices stay yours.
If the thread shows harm, or you feel unsafe, do not treat a chapter as a plan. Talk to someone qualified, or use a local emergency service. A calmer reading of a chat can help you see a pattern. It cannot stand in for care.
The clear next step is a What Are We? report on your one-to-one WhatsApp chat. You pay once per report, with no subscription. The amount is shown at checkout. Counts such as who starts a conversation and typical reply time come from code over the whole export, and quotes are taken word for word. Upload your chat.
Questions about the method
Does the AI invent statistics?
No. Message counts, reply times, and the other figures listed on this page are computed from the export. A signal with too little data is left out. A quote that is not in the file, word for word, is removed.
What counts as ghosting in the report?
A real attempt to reach the person, no later response, and a change from how you usually talk. A quiet weekend, on its own, does not meet that test.
Why exclude gaps of 6 hours from reply time?
A night of sleep, a workday, or a phone left in another room can open a long gap. The median reply time ignores those gaps so they are not stored as a slow reply. A new conversation is still counted when someone writes after more than 6 hours.
Which chat file can I upload?
The uploader accepts a WhatsApp chat export (.zip or .txt) today. It does not accept iMessage, Instagram DMs, Facebook Messenger, KakaoTalk or LINE yet. Group chats, screenshots and PDF files do not work.
Is the report free?
No. The full report is a one-time payment, not a subscription. The price is shown at checkout. There is no monthly plan.
Is this a substitute for therapy?
No. The report is not professional advice. If you need mental health care or you are unsafe, talk to a qualified person or a local emergency service.
Last updated: November 17, 2026.