Frequently Asked Questions

LoveGraft analyses your exported WhatsApp chat using AI. You export the chat from WhatsApp (without media), upload the .txt or .zip file to our site or send it to our WhatsApp bot, and receive a relationship report: a health grade, communication statistics, a red-flag checklist, psychology profiles for both people, and specific recommendations — each backed by messages from your own chat.

On iPhone: open the chat, tap the contact name at the top, scroll down to "Export Chat", and choose "Without Media". On Android: open the chat, tap the three dots in the top right, select "More", then "Export Chat", and choose "Without Media". On Desktop: open the chat, click the three dots, select "More", then "Export Chat". It takes about 20 seconds.

LoveGraft accepts .txt files (the standard WhatsApp export format) and .zip files (which WhatsApp sometimes creates automatically), up to 50MB. Both work identically — upload whichever your phone produces. If a file is over 50MB, re-export it and choose "Without Media".

Basic statistics are ready in seconds. The full AI analysis typically takes 1–3 minutes depending on the length of your chat and current load. You see progress while it runs, and the results page carries on if you return to it later.

The analysis is about two people. You can upload a group chat, but the report is built around the two most active participants, and the free preview's side-by-side comparisons only appear for two-person chats. For the clearest result, export a one-to-one chat.

Yes. We publish a complete sample report for a fictional couple, Alex and Jordan, across 15,847 messages. It is the exact report you would receive, free to explore, with nothing to upload. Open the sample report

The full AI analysis includes a relationship health grade (A–F), a red-flag checklist (gaslighting, manipulation, love bombing, silent treatment and more, each marked found or clear with the evidence), breakup probability and trajectory, attachment styles, communication styles and love languages for both people, communication analytics (response times, who initiates, effort balance, double texting, silences), the positive-to-negative ratio, language style matching, conflict patterns, the topics you talk about most, and personalised recommendations with suggested messages.

A letter grade from A to F, with + and − steps, derived by code from the evidence the AI gathered — the red flags found, the positive moments found, and communication quality — so the model never assigns the grade itself. It comes with a confidence level (a short, clean chat rates lower confidence than a long one with real conflict in it) and, where a chat has gone quiet, a note saying which period the grade mostly reflects.

AI analysis is a tool for reflection, not a definitive assessment. It is grounded in Gottman's research and can identify meaningful patterns, but it cannot understand full context, sarcasm, inside jokes, or cultural nuance. Think of it as a data-informed perspective, not a professional diagnosis. Results should never be the sole basis for a major relationship decision.

The AI looks for linguistic patterns associated with gaslighting (denying reality, dismissing feelings, rewriting events) and manipulation (guilt-tripping, emotional blackmail, controlling language). Each item on the checklist is reported as found or clear, with the message behind any finding, how often it occurred, and who did it. It flags potential concerns; it cannot confirm abuse — that requires professional assessment.

It is based on Dr. John Gottman's research into the communication patterns that predict relationship outcomes. The AI measures the ratio of positive to negative interactions, whether conflict gets repaired, any serious concerns such as control or coercion, and engagement trends over time. It is a statistical pattern match, not a prophecy.

Secure, anxious, avoidant and disorganised — for both people, with a confidence level. The AI reads texting behaviour such as response consistency, emotional expression and conflict handling to determine each person's style, and describes how the two styles interact.

Not with certainty, and LoveGraft does not claim to. What the analysis can do is measure the patterns research links to relationships under strain: a falling share of positive to negative exchanges, conflicts that get dropped rather than repaired, one person carrying most of the initiation, replies getting slower and shorter, and long silences. The report turns these into a breakup probability, a trajectory (improving, stable or declining) and a list of named risk factors, each backed by messages from your own chat. Treat it as an early-warning reading, not a forecast — people end and rescue relationships for reasons that never appear in a chat.

The advice in a LoveGraft report is specific to your conversation. It lists the strengths worth protecting, the concerns worth addressing, things to start and things to stop, and a set of action cards — each aimed at one of you, with a suggested message you could actually send. There is also a "next best message" section for common situations, such as re-opening after a silence or answering a complaint without escalating. Every recommendation points back to the evidence it came from, so you can judge whether it fits.

Reliability rises with the amount of conversation the AI can see. LoveGraft needs at least 20 messages to run, a few thousand gives a solid picture, and for very long chats the report states exactly which messages were analysed. Two safeguards are built in: the numbers behind the grade (reply times, who initiates, the positive-to-negative ratio, language style matching) are counted directly from your messages rather than estimated by the AI, and every claim in the report has to cite a message. What the analysis cannot see is everything that happened outside the chat, so read it alongside your own judgement rather than instead of it.

In two layers. First, plain statistics are computed from the export: message and word counts per person, reply times, who starts conversations, double texting, silences, activity by hour and weekday, emoji use, and how closely your writing styles match. Second, the AI reads the conversation in chronological sections and records evidence — warmth, support and appreciation on one side; tension, control, manipulation or reality-denial on the other — each with who did it, when, and a paraphrase of the message. A final pass turns that evidence into the report: grade, red-flag checklist, psychology profiles, trajectory and recommendations. The grade is derived by code from the evidence counts, never invented by the model.

It can show you the signals that usually travel with interest, and how balanced they are. The report measures who initiates, how quickly each of you replies, message length and effort, double texting, how often personal questions get asked, affectionate language and emoji, and whether the other person adopts your words and rhythm. It also gives an engagement balance and an effort asymmetry between the two of you. What it cannot do is read a mind: someone can be warm and responsive out of friendliness or habit. Use the numbers to check your read of the situation, not to replace it.

LoveGraft does not score each message as happy or sad. Instead the AI collects concrete positive moments (affection, support, appreciation, repair after a row) and concrete negative ones (tense exchanges, dismissiveness, controlling or manipulative behaviour) from your conversation, and the report counts both sides by the same rules to give a positive-to-negative ratio. Gottman's research puts stable relationships around 5:1, so the ratio is shown against that benchmark, together with an overall sentiment trend and where in the timeline the positive and negative periods fall. Emoji and affectionate wording feed the picture, but the count is of real exchanges, not of individual words.

Partly. The AI reads the whole conversation in order, so it picks up recurring jokes, nicknames and shared references, and the report lists the inside jokes, shared hobbies and future plans it found. It cannot know the history behind them, and it can misread sarcasm or teasing as tension. Two protections help: every concern in the report is paraphrased from an actual message, so you can see what was flagged and decide whether it was a joke, and one mild flag on its own is not enough to move your grade.

LoveGraft does not produce a personality test and does not claim to. What it profiles is how each person behaves in the conversation: attachment style (secure, anxious, avoidant or disorganised, with a confidence level), communication style, how each of you handles emotion and conflict, likely love language, triggers, coping habits, interests and goals — each backed by quotes from the chat. Those are observations of texting behaviour, not a validated psychological assessment, and they become more dependable the longer the chat is.

By showing you the habits you cannot see from inside the conversation. The report tells you who starts conversations and who replies faster, how balanced the effort is, how your writing styles match, how disagreements usually start and whether they get resolved or just dropped, and which patterns show up again and again. The recommendations then turn that into specific things to start, things to stop, and suggested messages for the moments that tend to go wrong for the two of you.

Only once the conversation is on WhatsApp. LoveGraft reads WhatsApp chat exports, so it cannot import Tinder, Hinge or Bumble messages directly. For an early-stage chat that has moved to WhatsApp, the same measurements apply: who initiates, reply speed and consistency, message effort, question-asking, and whether interest is rising or fading over time. The unlock screen shows a couple of these before you decide whether to buy the full report.

It measures the communication side of compatibility. LoveGraft calculates language style matching — how far two people's use of everyday function words converges, a measure from linguistic research on rapport — directly from your messages, and the AI describes how your attachment styles interact, how well your communication styles fit, and your conflict risk. It also lists the topics you discuss most and how each of you feels about them. Communication compatibility is a real signal, but it is one dimension; values, life goals and how you are together in person are outside the chat.

The report tracks engagement over the life of the chat: message frequency, who initiates, reply speed, message length, question-asking, affectionate language, and the longest silences. It reports the current cadence (daily, weekly, sporadic or dormant), the engagement trend, an emotional-distance reading, and flags breadcrumbing and ghosting patterns with the point in the timeline where they began. A drop in any of these can have innocent causes — exams, a new job, a bad month — so the report shows you the change and when it happened, and leaves the explanation to you.

Only in a limited way. The statistics show when the two of you are most active — the peak hour and weekday — and the timeline shows which stretches of the chat were calmer. That tells you when you are both likely to be present and unhurried, which is a reasonable starting point. Whether today is a good day depends on mood, privacy and what else is happening, which no chat export contains.

No, and LoveGraft does not rely on it. Frequency is one input among many: reply times, initiation balance, message length and effort, the positive-to-negative ratio, conflict and repair patterns, and the red-flag checklist all feed the report. Frequency is also read in context — the report states the chat's own rhythm and the era it was most active in, so a couple who moved most of their talking to another channel is not graded as if they had gone quiet.

It matters as a comparison, not as an absolute. Some people write in paragraphs and some in five words, so the report looks at each person's average length and effort relative to the other, and at how that changes over time. A steady shortening of one person's replies while the other's stay long is worth noticing; a chat where both people have always been brief is not a warning sign.

No. LoveGraft does not attempt lie detection, and no text-based tool can do it reliably. What the report can show is inconsistency in behaviour over time — for example a sharp change in availability, warmth or responsiveness — and it separately reports any secretive-behaviour signals it finds in the conversation. Those are observations to think about, not evidence of dishonesty.

It depends on consent. A chat has two authors, and our Terms of Service require you to have the informed consent of everyone in a conversation before you upload it. Used that way — to understand your own patterns, or together as a couple — the analysis is a reflection tool. Used as surveillance, it is not, and it can also mislead you, because it sees only the chat. If you would not be comfortable telling the other person you ran the analysis, that is a reason not to run it.

Through a set of behaviours counted over time: who initiates and how often, how quickly and fully each person replies, double texting, personal questions asked, affectionate and appreciative language, remembering details, and how each of you responds when the other reaches out. The report summarises these as an engagement balance and an effort asymmetry between the two of you, together with a commitment and trust reading. Consistently high investment on one side and low on the other is one of the clearest patterns the analysis surfaces.

A relationship AI is software that reads your actual conversation rather than answering general questions about relationships. LoveGraft takes an exported WhatsApp chat, computes the measurable facts (reply times, initiation, silences, style matching), has an AI read the messages in order to gather evidence of warmth, friction, repair and any serious concerns, and then produces a report: a health grade, a red-flag checklist, psychology profiles for both people, the trajectory of the relationship, and specific recommendations, each tied to messages from your own chat.

Not yet. LoveGraft reads WhatsApp exports (.txt or .zip) only. The parser is built around WhatsApp's export format, including its many regional date and time variants, so a file from another app is rejected rather than half-read. If the conversation you care about also lives on WhatsApp, export that one.

It depends on whether you want general advice or an analysis of your own relationship. A general chatbot can answer "is this normal?" from what you tell it, and it will be swayed by how you tell it. LoveGraft reads the conversation itself — every message from both people — and grounds the answer in counts and quotes: who initiates, how the two of you argue and repair, what the red-flag checklist actually found, how the pattern has changed over time. If you want an outside view of the record rather than a second opinion on your account of it, that is what LoveGraft is for. You can see a complete sample report before uploading anything.

Yes. From a WhatsApp export LoveGraft reads message frequency and timing, response times, who initiates, message length and effort, emoji use, affectionate versus tense language, question-asking, and how all of these change over the life of the chat. The AI layer adds the parts statistics cannot see: red flags such as guilt-tripping or gaslighting, attachment and communication styles, conflict and repair patterns, and the overall trajectory. Everything it reports is tied to messages from your own chat.

Yes, within limits we state plainly. Your chat is encrypted in transit and at rest, it is used only to produce your analysis, and only LoveGraft administrators can access stored files or reports — never for advertising, never sold, never used to train AI models. Analysing a chat means we have to hold it briefly: the uploaded file is deleted automatically within 2 days (7 days for WhatsApp bot uploads) and the finished report after 90 days. You can ask us to delete everything sooner at support@lovegraft.com. Full details are in our Privacy Policy.

In four ways. Deletion is automatic and scheduled at our hosting provider — chat files within 2 days (7 days for WhatsApp bot uploads), intermediate analysis data within 2 days, reports after 90 days, shareable links after 24 hours — so it never depends on anyone remembering. Access to stored data is limited to LoveGraft administrators and used only for support and quality checks. The AI providers we send messages to are configured not to train on your data. And the report is built to be shared safely: share cards use "Person A" and "Person B" unless you choose to show real names. Our database runs in the EU; chat files and the analysis itself are processed in the United States under the EU's Standard Contractual Clauses, which the Privacy Policy explains in detail.

Only for as long as the job needs. Your chat file is deleted within 2 days (7 days for WhatsApp bot uploads). Your finished report is kept for 90 days for support and quality purposes, and only LoveGraft administrators can see it. Shareable links expire after 24 hours. If you are signed in and saved the analysis to your account, it stays until you delete it. Payment records are kept as the law requires; they contain no chat content.

You, and — while it is stored — LoveGraft administrators, who use that access only to check what was delivered, help if you contact support, and improve the analysis. Nobody else: not the other person in the chat, not advertisers, and not the AI providers beyond producing the analysis. If you create a share card, you decide what goes on it, and names are anonymised by default.

Yes. To produce the analysis, your chat content is sent to third-party AI providers — currently Google's Gemini models do the main work. The providers we use are configured not to use your data for model training, and may retain requests for up to 30 days for abuse monitoring before deleting them. The current list of providers is in our Privacy Policy.

Basic statistics (message counts, response times, activity patterns) are completely free. The full AI analysis is a one-time payment — €4.99, or a lower local price in many countries, with the exact amount shown in your local currency at checkout — and no account is needed. A 5-credit pack is €8.99 (€1.80 per analysis) for people who want to analyse several chats or track one over time; packs need a free account so your analyses can be saved.

Yes. The basic statistics are free and include message counts, response times, who initiates, ghosting and silence detection, conversation energy, and activity patterns. The free version does not include the AI insights: the grade, red flags, psychology profiles, trajectory or recommendations.

All purchases are final and non-refundable. The AI analysis is delivered instantly and our costs are incurred at the time of processing, so we cannot offer refunds. Please review our Terms of Service for full details. If an analysis fails to deliver, contact support@lovegraft.com — a failed job is resumed, not resold.

No. LoveGraft uses one-time payments only. There are no subscriptions, recurring charges, or hidden fees. You pay once per analysis, or buy a credit pack, and that is it.

LoveGraft works on all modern browsers including Chrome, Safari, Firefox, and Edge. For the best experience, use the latest version of your browser. Internet Explorer is not supported.

Yes. LoveGraft is fully responsive and works on phones and tablets. You can export your WhatsApp chat and upload it directly from your phone, or share the export straight to our WhatsApp bot, without needing a computer.

Yes. The AI reads the conversation in whatever language it is written in, and the word statistics are tuned for English, German, Dutch, Afrikaans, French, Spanish, Portuguese and Italian, with script detection for Japanese, Korean, Chinese, Arabic, Russian, Hindi and Thai. The report itself is written in English.

Yes. Download your analysis as shareable images from the results page, or sign in with a credit pack and it is saved to your account. Otherwise, save what you want to keep before you leave: we retain reports for 90 days for support, but that retention is not a backup you can retrieve from.

Check that the file is a WhatsApp export in .txt or .zip form, under 50MB (re-export with "Without Media" if not), and that it contains at least 20 messages. If WhatsApp produced an unusual date format, the site will say so rather than guess — send the first few lines to support@lovegraft.com and we will add it. If the problem persists, try another browser or contact support.

No. LoveGraft is an informational tool, not a decision-maker. It can help you see patterns you might have missed, but major relationship decisions should involve self-reflection, honest conversations with your partner, and ideally guidance from a licensed therapist or counsellor.

You can, but use discretion. The analysis is an AI perspective that may highlight sensitive issues. Sharing is most productive when done in the spirit of improving communication, ideally with a therapist to guide the conversation. The share cards anonymise names by default, so you can also share a result without identifying anyone.

They do different jobs. LoveGraft is fast, private and objective about the record: it counts who texts first, shows the positive-to-negative balance and quotes the messages behind every concern, with no agenda. A therapist brings training, empathy and the context of your whole life, and can work with trauma or abuse — which a report cannot. If the analysis surfaces something painful, treat that as a good reason to talk to a professional, not as a substitute for one.

If the AI identifies concerning patterns, take them seriously but do not panic. AI can produce false positives, which is why every flag shows the message behind it. Use the findings as a starting point for reflection or a conversation with a professional. If you feel unsafe, contact local support services or emergency services.