About Bitch Please

Reading a conversation you are emotionally involved in is genuinely hard, and not because people are stupid about it. Two specific things go wrong.

The first is that you cannot hold a timeline in your head while you are upset. A message at 11:04 and a message at 11:05 are one minute apart. If those two messages contradict each other, the contradiction is sitting in plain sight — but when you are scrolling at 1am looking for reassurance, you read for tone rather than for sequence. You come away with a feeling instead of a fact.

The second is that everyone you ask is on your side. Your friends love you. They will confirm the thing you already suspect, because that is what love looks like at 1am. That is valuable and it is not analysis. What is missing is somebody who will read the same screenshots and say “actually, that checks out” when it does.

Bitch Please is built to be that second thing. It is not designed to agree with you. It is designed to be specific.

When you submit a conversation, the system does four things in order. Each one is constrained by the one before it, which is the whole design.

The score is a measure of how much fails to line up. It is not a measure of how likely somebody is to be lying, and it is not a probability of anything.

There is a hard rule behind the number: a score above 40 must be backed by either a specific contradiction or a statement traced to actual evidence. If the system produces a high score with nothing underneath it, the result is rejected and regenerated rather than shown to you. A confident number over an empty analysis is the single most dangerous thing a product like this can produce, so it is blocked in code rather than discouraged in a prompt.

A case is not a single verdict. It is a question that stays open until you close it. When you come back with new evidence, the system does not start over. It receives everything already on record — the exact statements, with their dates — and works out what the new material does to them.

Sometimes that raises the score. Often it lowers it. In one real case the score moved from 46 to 27 because a carrier outage notice supported the explanation that had looked weak, and a story that appeared incriminating turned out to be a repost from the previous year.

The system did not fully capitulate, though, and this is the part worth understanding: it kept one detail flagged, because the outage ended at 11:03 and he said he saw the call at 11:05. It moved on the evidence that moved and held the two minutes that did not.

A tool that can only accuse is not measuring anything. Being able to say “that actually checks out” is what makes the accusations worth taking seriously.

Here is a real case, with the numbers it actually produced. A woman uploads one screenshot of a conversation with her boyfriend and writes: he said he was home by 11, but he posted a story at 1am with music playing.

It will sometimes be wrong, and the honest thing is to design for that rather than pretend otherwise. It can misread a low-quality screenshot, mistake a quoted reply for an original message, or weigh something more heavily than it deserves.

Three things limit the damage. First, every finding is traceable: if the system claims something was said, you can see which message it came from and check it yourself. A wrong conclusion built on a real quote is arguable; a wrong conclusion built on a quote that never existed is not, which is why fabricated citations are discarded rather than displayed.

Second, you can correct it, and corrections stick. Tell it that it misread something, or that a person is no longer who it thinks they are, and the earlier belief is marked superseded rather than silently overwritten.

Third, a result that cannot be supported is rejected before you see it. If the system generates a high score with no contradiction and no evidence-traced statement behind it, that output is thrown away and regenerated. This happens invisibly and it is the single most important guard in the product.

A friend gives you loyalty. This gives you a timeline. Both are useful and they are not substitutes for each other.

The last row is the important one. This product is built to hand you a better question, not a decision. It can tell you that the outage ended at 11:03 and the call was seen at 11:05. What that means about your relationship is not something a machine gets to say.

On the paid plan, Bitch Please keeps a memory of your situation across months. It does not keep a transcript. Storing everything would be both expensive and useless — recall works better when there is less noise in it.

What gets kept:

• Relationship facts, such as who someone is to you and when that changed.
• Important people and how they connect to the situation.
• Significant events with dates.
• Questions that were never answered.
• Corrections you made to something the system believed.
• Resolutions, including the ones where the explanation checked out.

What gets discarded: jokes, filler, phrasing, emotional weather, speculation, and anything already obvious from the verdict. Most exchanges produce nothing worth remembering, and the system is explicitly instructed that returning nothing is the correct and common answer.

Screenshots go directly to private storage and are never publicly readable. They do not pass through a public endpoint, and the storage path is chosen by the server rather than by the browser, so one person's upload cannot land in another person's space.

Possessing a case link does not grant access to the case. Ownership is checked on the server for every read, for signed-in accounts and anonymous sessions alike. A case that is not yours returns the same response as a case that does not exist, because confirming that a case exists is itself a disclosure.

Sharing is a separate, explicit act. When you share a result, the system creates a new public object with its own identifier. It contains a score and one line, with phone numbers, emails, handles, links, addresses and the other person's name removed. It never contains the screenshots, the messages, the conversation, or the case identifier. You see exactly what will be published before it is published, and you can revoke it at any time.

Deleting a case deletes the evidence objects, the memories derived from it, and any share card, and cancels queued processing. See the privacy policy for the specifics.

These are refusals, not features that have not shipped yet.

• It will not tell you whether someone is lying. It has no way to know that, and any product claiming otherwise is selling you a feeling.
• It will not diagnose anyone. Not narcissism, not any other condition. It has met neither of you.
• It will not help you monitor a person. No advice on accessing somebody's phone, accounts, messages or location without their knowledge.
• It will not stay in character where that would be harmful. If a conversation shows signs of violence, threats, stalking, coercive control or self-harm, the tone changes completely — no score talk, no jokes — and it points toward real help.
• It will not pretend to have evidence it does not have. Ask it for advice with no screenshots and it will give you a genuinely useful read, but it cannot produce a single evidence-backed statement, because there is no evidence to point at. That is enforced by the same rule that governs everything else.

You can describe a situation instead of uploading anything. The system gives you a read on how much the situation actually warrants your worry, the strongest fair version of the other person, what you might not be seeing, and what you would need to know to settle it.

The same anti-validation rule applies. Advice that only agrees with you is the same failure as an analysis that only accuses — it just feels nicer on the way out. So the generous reading of the other person is mandatory there too, and the section telling you what you may be missing is the point of the whole thing rather than a footnote.

It is for the person who has read the same four messages nine times and cannot tell any more whether they are seeing a pattern or inventing one. It is for the moment when the story changed slightly and you are not sure whether you are remembering the first version correctly.

It is not for building a case against somebody. If you are looking for a tool to confirm a decision you have already made, this one will keep disagreeing with you, and it will do it with specific quotes.

What is Bitch Please?

Bitch Please is an AI text message and relationship analyzer. You upload screenshots of a conversation or paste the messages, and it reads them, reconstructs the timeline, and shows you which parts do not line up. Every finding is tied to a specific message it can quote, and every verdict includes the strongest innocent explanation alongside the suspicious one.

Is Bitch Please a lie detector?

No. Bitch Please is not a lie detector and cannot determine whether someone is lying. It measures how much of an account fails to line up internally — a timestamp that contradicts a claim, a story that changed between messages. Inconsistency is not proof of deception, and people contradict themselves for ordinary reasons including bad memory, imprecise language and embarrassment.

How does the Bitch Please score work?

The Bitch Please score runs from 0 to 100 and measures how much of the material fails to line up, not how likely someone is to be lying. Scores of 0 to 20 mean things look fine, 21 to 40 mean something is slightly off, 41 to 60 mean the situation is genuinely unresolved, 61 to 80 mean there is a real problem in the material, and 81 to 100 mean the account contradicts itself in a way that is hard to explain innocently.

Can the score go down?

Yes. When you add new evidence to an existing case, the score can move in either direction. A case scored 46 that drops to 27 because a carrier outage notice confirmed the innocent explanation is the product working correctly. A tool that can only accuse is not measuring anything, and being able to clear someone is what makes the accusations worth taking seriously.

Does Bitch Please keep my screenshots private?

Yes. Screenshots are uploaded directly to private storage that is not publicly readable, and a case can only be opened by the session or account that created it. Possessing a case link is not enough to read it. Sharing is a separate, explicit action that creates a scrubbed card containing only a score and one redacted line — never the screenshots, the messages, or the conversation.

What does Bitch Please remember between sessions?

On the paid plan, Bitch Please stores durable facts about a situation: who the people are, what was claimed and when, which questions were never answered, and any corrections you made. It does not store every message. When you correct something, the earlier version is marked superseded rather than deleted, so the history of what was believed at each point stays intact.

How much does Bitch Please cost?

Bitch Please is free to try without an account, with a limited number of analyses. Creating a free account raises that allowance. Bitch Please Unlimited costs $9.99 per month and adds persistent cases, memory across months, ongoing conversation, Reply For Me and Recheck Everything.

Can Bitch Please tell me if my partner is cheating?

No. Bitch Please cannot tell you whether someone is cheating and does not try to. It can show you that a specific claim in a specific message does not fit another specific claim, and it can tell you which question would settle the discrepancy. The conclusion is yours to draw, and the product is deliberately built to hand you a better question rather than a verdict about a person.

What will Bitch Please refuse to do?

Bitch Please will not diagnose a mental illness, claim to detect lies, or help you access someone's phone, accounts, messages or location without their knowledge. If a conversation shows signs of violence, threats, stalking, coercive control or self-harm, it drops the entertainment tone entirely and responds plainly, including pointing toward real help.