How AI Analyzes Text for Dating Compatibility
Yes, and the science is robust. A landmark study by Schwartz et al. (2013) at the University of Pennsylvania analyzed 700 million words from 75,000 Facebook users and demonstrated that language patterns predict Big Five personality traits — openness, conscientiousness, extraversion, agreeableness, and neuroticism — with statistically significant accuracy. Extraverts used more social words and exclamation marks. People high in openness used more complex sentence structures and abstract vocabulary. The correlations held across age, gender, and cultural backgrounds.
"Language is the most reliable behavioral signal we have. People can curate a photo, but their writing style reveals patterns they don't consciously control." — Dr. Martin Seligman, University of Pennsylvania, founder of positive psychology
This is the foundation of text-based compatibility analysis. When you write your manuscript on Anketta, the AI isn't just reading your words — it's building a semantic meaning vector that gets compared against the vectors of other members.
Think of it this way: every word you write occupies a position in a vast mathematical space. Words with similar meanings sit close together. "Cozy" and "warm" are neighbors. "Adventure" and "exploration" are neighbors. This is what researchers call a vector embedding — a way to convert language into numbers that preserve meaning.
Modern embedding models (like those behind ChatGPT) use transformer architecture to process entire paragraphs at once, capturing context and nuance. The sentence "I love quiet evenings at home" and "I need space to recharge" are superficially different — they share no keywords — but their vector representations are remarkably close because they express similar personality needs. A 2022 study in Nature Machine Intelligence found that transformer-based embeddings captured semantic similarity with 89% accuracy compared to human raters (Bommasani et al., 2022).
Traditional keyword matching would miss this connection entirely. It would treat "quiet evenings" and "space to recharge" as unrelated because the words don't overlap. Semantic analysis understands that both describe an introverted preference for low-stimulation environments.
This is the academic backdrop the field stands on — not a description of what Anketta specifically computes. Beyond the Big Five, modern NLP identifies emotional tone, attachment style indicators, communication preferences, and value hierarchies. A 2020 study in the Journal of Research in Personality found that text analysis predicted attachment styles — secure, anxious, avoidant — with 71% accuracy from as few as 500 words (Park et al., 2020). For context, that's roughly the length of a substantive manuscript on Anketta.
Across the literature, these signals consistently emerge from free-form text:
- Emotional expressiveness. How readily someone shares feelings, measured by the ratio of emotional to factual language. Research shows compatible couples typically have similar expressiveness levels (Pennebaker, 2011).
- Communication depth. Whether someone gravitates toward surface-level observations or introspective analysis. A 2019 study in Personality and Social Psychology Bulletin found depth-matching predicted relationship satisfaction better than shared interests.
- Humor style. Affiliative humor (bringing people together) vs. self-deprecating humor vs. observational wit. Each style signals different social orientation and coping mechanisms.
- Value signaling. The themes people return to — family, career, creativity, independence — reveal prioritized values even when not explicitly stated.
"We found that 500 words of free-form text contain more psychologically predictive information than a 300-item personality questionnaire." — Dr. Gregory Park, University of Pennsylvania, computational social scientist
Anketta does not lean on these specific classifiers. Instead of building separate "attachment" or "humor" detectors, the product takes a more modest, transparent route — embed the manuscript and compare meaning vectors between people. Details below.
Most dating apps don't analyze text at all. Tinder's algorithm primarily optimizes for engagement — it shows you profiles likely to keep you swiping based on your past behavior and the Elo-like desirability rating system (Tinder Engineering Blog, 2019). Mutual attraction is inferred from swipe patterns, not personality compatibility.
Hinge's "Most Compatible" feature uses the Gale-Shapley algorithm — originally designed for matching medical residents to hospitals — combined with behavioral signals like who you like, comment on, and skip (Hinge, 2021). It's more sophisticated than Tinder but still relies on behavioral proxies rather than understanding who people actually are.
Bumble's algorithm weighs recency, activity, and profile completeness. It rewards users who log in frequently and respond quickly — good for engagement metrics, less meaningful for compatibility prediction. For detailed comparisons, see our analyses of Anketta vs. Twinby and Anketta vs. Hinge.
Anketta's approach is fundamentally different — and more modest in what it claims. The model builds a semantic vector for fragments of your manuscript and finds people whose fragments sit nearby in that meaning space. On top of that, the system uses your explicit highlights ("like" / "dislike" on lines) and your individual word preferences. No swipe data, no popularity scores, no engagement optimization, no claim to recover your attachment style from text.
Transparency and consent are non-negotiable. According to a 2023 Pew Research survey, 67% of dating app users expressed concern about how their data is used for matching, and 53% wanted to understand the algorithms behind their matches. The black-box approach — where users have no idea why they received a particular match — erodes trust.
Ethical AI matching requires three pillars:
- Informed consent. Users know their text is being analyzed and understand what's being measured. No hidden profiling.
- Data control. Users can view, modify, or delete their manuscripts and the analysis derived from them. GDPR and its equivalents aren't just legal requirements — they're ethical minimums.
- Explainability. When Anketta suggests a match, the reasoning should be understandable: "You both write extensively about creative pursuits and value personal growth." Not a score. Not a percentage. An explanation.
A 2024 study in AI & Society found that users who understood how AI matching worked reported 34% higher trust in matches and were 28% more likely to engage in meaningful conversation (Zhang et al., 2024).
The same transformer technology powering ChatGPT is being applied to understand human compatibility in ways that were impossible five years ago. Global investment in AI-powered dating features reached $1.2 billion in 2025 (PitchBook, 2025), signaling an industry shift away from photo-first, swipe-based models.
The key insight is simple: compatibility lives in language. How you describe your weekend tells more about partnership potential than what you look like doing it. AI text analysis makes that insight scalable — and when paired with the depth of an essay-based profile, it produces matches grounded in genuine understanding rather than algorithmic guesswork.