WappLoveScoreToxic relationship? Here's what your WhatsApp chat can truly measure (and what it can't)
Team at WappLoveScoreWritten with AI assistance15 min readIn summary
- No message analysis diagnoses a toxic relationship. What can be measured are risk patterns that research has associated with poorer couple quality: contempt, imbalance in who holds the conversation, long silences, and loss of shared language.
- "I’ll tell you in three minutes" is marketing. Gottman's famous divorce prediction with over 90% accuracy was never validated in new couples: Heyman and Slep demonstrated in 2001 that when crossing the formula with a fresh sample, the predictive value plummets.
- Asking a chatbot for a verdict is the worst idea. The benchmark ELEPHANT (Cheng et al., Stanford/CMU/Oxford, ICLR 2026) measured that language models validate the user 45 percentage points more than people and, in interpersonal conflicts, side with the party they are told about 48% of the time.
- There are four signals with decent evidence and easy to count in a chat: the use of "we", style symmetry, asymmetry of response times, and the presence of emojis.
- Exporting the chat is done in three taps and the file fits in a
.txt. Note: since April 2025, either party can block the export with advanced chat privacy.
Your WhatsApp chat is the most honest record you have of your relationship
If you have been talking to the same person on WhatsApp for years, you have in front of you the longest and least edited diary of that relationship: tens of thousands of messages with date, time, and author. The problem is not the lack of data, it’s that no one reads 40,000 messages. That is exactly the crack through which tools like WappLoveScore, the miniapp from AIsToHire that turns a WhatsApp export into a portrait with numbers, come in; but before scoring anything, it’s advisable to know what of all this has scientific backing and what is just talk. This article is about the former.

The context helps to understand why the question appears so much on Google. WhatsApp surpassed 3 billion monthly active users in 2025 and in Spain, it is the app with the highest penetration: 89.4% of internet users used it in the last month, according to the WhatsApp statistics collection for Spain published by SendSeven in 2026. Meanwhile, the National Institute of Statistics recorded 83,683 annulments, separations, and divorces in 2025 —80,785 of them divorces, a 2.7% decrease from the previous year— according to the press release from the ENSD published on July 23, 2026. And when people are asked what is wrong with their partner, the most repeated answer is the same: in the report Happiness in Couples in Spain by Unobravo (2025, 271 surveyed; small and self-selected sample), 35.8% pointed to poor communication as the main problem, ahead of unequal effort (14.2%).
That is: conversation is the symptom that hurts the most and, at the same time, the only one that is written.
Can you tell if a relationship is toxic by reading the messages?
No. A message analysis does not diagnose anything: "toxic relationship" is not a clinical category, and no word count replaces a professional. What an analysis can do is describe observable patterns —who writes, when, with what tone, with what consistency—and point out those that literature has associated with poorer relationship quality.
The popular reference is John Gottman and his "four horsemen" (criticism, defensiveness, contempt, and stonewalling), with contempt being the most corrosive. That part is well replicated as a risk factor. What does not hold up is the figure that everyone repeats: Richard Heyman and Amy Smith Slep published in 2001 The Hazards of Predicting Divorce Without Crossvalidation showing that those prediction equations were fitted to the same couples whose outcomes were already known, without cross-validation; when tested on new couples, accuracy and predictive value plummet.
Practical translation: distrust any app, test, or coach that promises you a forecast. What’s honest is a descriptive portrait, not a crystal ball.
The signs of a relationship that research has truly measured
There are four linguistic indicators with published evidence that can also be counted in a text file without making anything up. None is a verdict by itself: they are trends measured in specific samples.
| Signal | What was measured | Published result | How it counts in a chat |
|---|---|---|---|
| Use of "we" (we-talk) | Meta-analysis of 30 studies, ~5,300 participants (Karan, Rosenthal, and Robbins, Journal of Social and Personal Relationships, 2019) | Positive association between the use of the first person plural and couple functioning; the effect of the partner was stronger than the self | Proportion of "we / our / together" over the total pronouns, per person |
| Style symmetry (language style matching) | 86 couples and their instant messaging (Ireland, Slatcher, Eastwick, Scissors, Finkel, and Pennebaker, Psychological Science, 2011) | 76.7% of couples with LSM above the median stayed together after 3 months, compared to 53.5% of those below | Similarity in the use of function words (articles, prepositions, conjunctions) between the two |
| Asymmetric response times | 302 college students in relationships (Patel, Mahar, and Webster, Cyberpsychology, Behavior, and Social Networking, 25(3), 2022) | People with anxious attachment took less time to respond than their partners; avoidants took longer. Anxious attachment also received fewer messages | Median time to the first response from each, and their difference |
| Emojis | 260 participants rating 15 conversations (Eun Huh, University of Texas at Austin, PLOS ONE, 2025) | Messages with emojis were perceived as more receptive, and that perception predicted more closeness and satisfaction. The type of emoji didn't matter, only its presence | Emojis per 1,000 messages and distribution per person. Study limit: hypothetical conversations with friends, not real partners |

In addition, there is a finding that is not in the text of the chat but in its context: the study Objective Phone Use During Time with One's Partner (Human Behavior and Emerging Technologies, 2025) measured actual mobile use —not self-reported— and found that people used it during 27% of the time spent with their partner, and 86% did so daily in their presence. Mobile use while with the partner (not total daily use) predicted lower relationship satisfaction and poorer quality of shared parenting, although the effects were only significant in women. In other words: it’s not how much you talk on WhatsApp, it’s when.
Why you shouldn't ask a chatbot to judge your relationship?
Because it is designed to agree with you. In 2026, it has become common to hit up an assistant with a conversation and ask "Is this normal?", and the evidence says that is precisely the question it responds to the worst.
The benchmark ELEPHANT (Myra Cheng and collaborators from Stanford, Carnegie Mellon, and Oxford; preprint arXiv 2505.13995, published in the proceedings of ICLR 2026) measured social flattery in 11 language models. Results:
- They preserve the image that the user has of themselves 45 percentage points more than people in general advice queries and in cases of clear user fault.
- In interpersonal conflicts, they side with the presented side 48% of the time.
- They do not question unfounded assumptions in 86% of cases.
If you tell it your version, it will validate your version. That’s why it makes sense to separate two things: the recount (how many messages, from whom, when, with what words) and the interpretation. The recount is arithmetic about your own file and does not opine; the interpretation is yours, and if necessary, from a professional.
What the clock says and the content does not: hours, streaks, and eras
Temporal metrics are the most undervalued and the hardest to discuss: they do not depend on interpreting words, only on counting timestamps. Monthly volume, time of day, streaks of consecutive days, and silences: that’s where the real shape of a relationship is seen.

What is worth looking at, and why:
- Messages per month. Locate your golden era and your valleys. A sustained drop over months says more than a bad week.
- Golden hour. The time frame in which you talk the most usually reveals the role of the chat: early mornings of discovery, midday logistics, nights of venting.
- Longest streak of days talking and longest silence, with date. They are the two extremes of consistency.
- Difference in response times between the two (see previous table).
- Who reopens the conversation after a silence. If it’s always the same person, that’s a fact, not an impression.
Regarding whether it is advisable to discuss via message: a review of 15 studies (mostly American, 2007-2024) signed by Marie-Ève Daspe’s team at the University of Montreal and published in the Journal of Social and Personal Relationships concluded that no channel is consistently superior for resolving disagreements; effectiveness depends on context, attachment style, and prior satisfaction, and some studies found more apologies and a greater sense of closure in writing, especially among women (reported by Infobae on June 5, 2026). Unglamorous conclusion: the channel matters less than consistency and tone.
How to export your WhatsApp conversation (and what that file really contains)
Exporting a chat takes three taps and the result is a simple text file with one line per message: date, time, author, and content. On iPhone: open the chat, tap the name at the top, scroll down to Export chat and choose Without media. On Android: open the chat, menu ⋮ > More > Export chat > Without media.

Choosing "Without media" is not a detail: with photos and videos the file goes up to hundreds of megabytes and no text analysis needs them. Two important and current warnings:
- It may be blocked. Since April 2025, WhatsApp offers advanced chat privacy, an option that is activated conversation by conversation and prevents participants from exporting the chat, automatically downloading files, and using messages with AI features (announcement from April 2025; analysis by EFF from September 9, 2025). If your partner has it activated in your chat, there will be no export. And vice versa: if someone does not want their conversation to be exported, that is the lever.
- Not everything is exported. The export cuts the history to the most recent messages —the figure circulating in third-party documentation is about 40,000 messages without media and 10,000 with them; WhatsApp does not publish it in its help center—, deleted messages appear as system text and what is no longer on your phone is not in the file.
And a third front, this from 2026: Meta has been rolling out AI message summaries in WhatsApp, processed with its Private Processing system, announced and covered by tech press in April 2026 and being deployed in phases and markets. Having an automatic summary within the app is not the same as an analysis of your history: they are different things and it is important not to confuse them.
Golden rule of privacy: a conversation belongs to two. Exporting and analyzing on your own device, for personal use, fits into the so-called domestic exception of the GDPR (article 2.2.c), which the AEPD considers applicable to data exchange between individuals in the context of their private lives. Disseminating the content or using it for purposes other than personal is another matter, and there the exception no longer protects you. This is general information, not legal advice.
What a chat analysis cannot tell you (and when to stop counting words)
A conversation analysis is not a psychological evaluation or a test. It does not know what happened outside the chat, does not reliably detect irony, does not know who was right, and cannot tell you if you should stay in that relationship.
Concrete limits to keep in mind:
| What an analysis does provide | What it does not provide |
|---|---|
| Verifiable counts and rates about your own file | A diagnosis of "toxic relationship" |
| Temporal patterns with dates (periods, streaks, silences) | A prediction of breakup |
| Distribution and asymmetries between the two | The context of what happened face to face |
| Vocabulary, nicknames, and fillers of the partner | Reliable intention, irony, or sarcasm |
And a boundary that is non-negotiable: if in that conversation there is fear, control, threats, isolation, or violence, what you need is not a score. In Spain, the 016 is available 24 hours a day, is free, is available in several languages, and leaves no trace on the bill (it does leave a trace in the browser or phone history, so it is advisable to delete it). You can also write via WhatsApp to 600 000 016 or by email to [email protected].
How to do all this counting without setting up a weekend project
The above can be done manually with a spreadsheet and a lot of patience, or it can be automated. WappLoveScore, the test that scores your relationship based on the WhatsApp chat, is built exactly on that idea: you upload the export and the analysis runs on your own browser, without the chat leaving the phone.
What it returns, in terms of what we have seen in this article:
- A Love Score from 0 to 100 with its level (friendship, spark, love, or passion), calculated from 9 indicators with rates per 1,000 messages, and an accordion "Why this number?" that shows definition, count, weight, and points of each one. The calculation is arithmetic: AI does not participate in the scoring.
- A six-axis radar and a tension thermometer measured separately from love, because they are different things.
- A face-to-face scoreboard with thirteen duels between the two participants: who writes more, who responds faster, who rescues the conversation.
- Your dictionary: nicknames, the ten most personal words, laughter synchronicity, the signature emoji.
- The clock metrics: golden hour, longest streak, and periods month by month, with dated milestones like the first "I love you", the first heart, the day you talked the most, and the longest silence.
- If requested, an AI-written chronicle —a separate button— and the option to save the report. Only in those two cases does something travel to the server, and what travels are numbers and aliases, not messages.
- Extras to share without exposing yourselves: aliases, a discreet mode, a certificate, and stories. And if the chat is a group, you choose the two you want to analyze.
Two limits that the app itself declares and that fit with what has been said above: the analysis lexicons are in Spanish and English (the interface is in six languages), so in other languages the vocabulary metrics remain incomplete; and there is a sample chat to see it functioning without uploading anything of yours. It is not therapy or a diagnosis: it is the counting that you are not going to do manually.
Frequently Asked Questions
Can you tell if a relationship is toxic by WhatsApp messages?
Not with a diagnosis. What can be measured are patterns associated with worse relationship quality —contempt, asymmetry in who holds the conversation, long silences, loss of "us"—, and that helps to put data where there were previously impressions. The assessment, if it weighs, corresponds to a professional.
How do I know who cares more in the couple according to the chat?
That question doesn't have a clean answer, but there are several measurable pieces: who writes more messages, who responds first, who reopens the conversation after a silence, who uses more expressions of affection, and who uses more "us." The meta-analysis by Karan and colleagues (2019) also found that your partner's "us" predicts your satisfaction more strongly than your own.
What does it mean if my partner takes hours to reply?
By itself, not much. In the study by Patel, Mahar, and Webster (2022) with 302 people in relationships, avoidant attachment was associated with replying later than one's partner and anxious attachment with replying sooner: it's an indication of style, not of lack of love. What is informative is the sudden change from a stable pattern, not the number of minutes.
Is it safe to upload my conversation to a website for analysis?
It completely depends on where it is processed. If the file is uploaded to a server, you are handing over the complete history of two people to a third party. It is reasonable to demand that the analysis occurs in your browser and to check it: load the page, put your phone in airplane mode, and see if the tool continues to work. WappLoveScore is designed this way and presents it as a verifiable test.
How many messages are needed for the analysis to say something?
There is no official threshold, and any number given to you without a source is made up. What is mathematical: the rates per 1,000 messages are very unstable with few data, and the temporal metrics (periods, streaks, silences) need months of history to make sense. A chat of a few weeks provides curiosities; one of years provides shape.
Why don't I just ask ChatGPT to analyze my chat directly?
Because social flattery is measured: in the ELEPHANT benchmark (ICLR 2026), models validated the user 45 percentage points more than people and sided with the presented side 48% of the time in interpersonal conflicts. To count, count; to judge your relationship, it tends to confirm you.
Can I analyze a group chat or only a two-person conversation?
You can start from a group chat if the tool allows you to choose which two participants to compare, which is how WappLoveScore resolves it. What doesn't make sense is to score a relationship with the noise of eight people within the count.
What should I do if the analysis comes out poorly?
Talk about it, don't forward it. A low number is not a sentence: it may reflect that your relationship lives outside the mobile, that you use another channel, or that you are going through a rough patch. And if what appears in the conversation is fear or control, the resource is not an app: in Spain, the 016 is available 24 hours, is free, and leaves no trace on the bill.
In a nutshell
Your chat won't tell you whether you should stay in that relationship, but it can stop being an unreadable block of 40,000 messages and become data that you can look at calmly: who holds the conversation, when you were at your best, and how long it has been since you wrote to each other like that. If you want to see those numbers without your conversation leaving your phone, WappLoveScore, the mini app from AIsToHire, does the counting and shows you where each point comes from —and the interpretation, as it should be, stays with you.
Upload your chat export and see the numbers with your own eyes: WappLoveScore analyzes the conversation in your browser, without leaving your phone, and shows you where each point comes from. Try it for free, or take a look at the sample chat first.
Upload your chat export and see the numbers with your own eyes: WappLoveScore analyzes the conversation in your browser, without leaving your phone, and shows you where each point comes from. Try it for free, or take a look at the sample chat first.Keep reading
View all articles →
Discussing via WhatsApp or talking in person: what science says about couple conflicts through messages

Why Your Car Insurance Increased in 2026: Average Premium of €1,412 and Historic Record

Mind map, concept map, and outline: differences and when to use each one
