Are Fertility Apps Reliable?

Not to the day, if the app works from cycle length. In a study where 949 volunteers collected urine samples across a full menstrual cycle so that the luteinising hormone surge could establish when ovulation actually occurred, the accuracy of ovulation prediction across those 949 participants was no better than 21% for the cycle-tracking apps tested (Johnson, Marriott and Zinaman, Current Medical Research and Opinion 2018;34(9):1587-94). A separate assessment compared 20 websites and 33 apps against a standardised 28-day cycle with ovulation on day 15 and a fertile window of cycle days 10 to 15: one website and three apps predicted the precise fertile window (Setton, Tierney and Tsai, Obstetrics and Gynecology 2016;128(1):58-63).

That is a statement about calendar prediction, not about every app. Tools that read signs from the cycle you are currently in — temperature, cervical fluid, a hormone test — perform differently, and are treated separately below.

Six days. In 221 women contributing 625 cycles in which the day of ovulation could be estimated from urinary hormones, conception occurred only when intercourse took place during a six-day period ending on the estimated day of ovulation; the probability of conception rose from 0.10 when intercourse happened five days before ovulation to 0.33 on the day of ovulation itself (Wilcox, Weinberg and Baird, New England Journal of Medicine 1995;333(23):1517-21). A later re-analysis that corrected for error in identifying ovulation found the same six-day interval in two independent datasets, with the highest probability on the day before ovulation and a probability close to zero after ovulation (Dunson et al, Human Reproduction 1999;14(7):1835-9).

The ASRM and SREI Practice Committees define the fertile window the same way — the six-day interval ending on the day of ovulation, with peak fecundability in the two days before ovulation (Fertility and Sterility 2022;117(1):53-63). Two things follow. The window opens before ovulation and shuts on it, so an app that flags fertile days after ovulation is flagging days with close to zero probability. And because the window is six days wide, being wrong by three days can place it entirely outside the real one.

Because it assumes the stable part of the cycle is the part that moves. Among 612,613 ovulatory cycles logged by 124,648 users of one app, mean cycle length was 29.3 days, the mean follicular phase was 16.9 days with a 95% confidence interval of 10 to 30 days, and the mean luteal phase was 12.4 days with an interval of 7 to 17 days (Bull et al, npj Digital Medicine 2019;2:83). The follicular phase — the stretch before ovulation that calendar maths fixes at 14 days — is where almost all the variation lives.

The Johnson data show the same thing from the user's side. Of 949 volunteers, 34% believed they had a 28-day cycle and only 15% actually did. Even within genuine 28-day cycles the most likely day of ovulation was day 16, and that day accounted for only 21% of those cycles among the 949 participants. There is no single day to predict from cycle length, which is why an app that only knows your period dates cannot supply one.

The warning attached to this is not academic. Freis and colleagues note that a fertile window wrong by a few days may lead a couple to concentrate intercourse on less fertile or non-fertile days, which is worse than not timing it at all (Frontiers in Public Health 2018;6:98).

Freis and colleagues built a 30-point scoring system covering which parameters an app uses to set the fertile window, what evidence exists for that method, what evidence exists for the app itself, and whether qualified counselling is offered. They then scored 12 apps available in German and English:

• All six calendar-based apps scored 0 of 30. They set the fertile days from previous cycles and met none of the criteria.
• The two calculothermal apps scored 3 of 30 and 2 of 30.
• The four symptothermal apps, which read the current cycle through temperature and cervical fluid, scored 20, 20, 19 and 11 of 30.

The authors judged three of the twelve eligible for further study and called for prospective evaluation by independent investigators free of commercial interest. That caveat is worth carrying into every figure on this page: several of the largest datasets in this field were produced by the companies selling the products, and are marked as such in the source list below.

Three references, in ascending order of rigour, and knowing which one a claim rests on tells you how much it is worth:

• Urinary LH. Cheap and widely used, and the benchmark in the Johnson study. ASRM notes that urinary LH devices carry approximately 7% false-positive results, versus a surge that may be followed by ovulation at any point in the next 2 days.
• Basal body temperature. Confirms ovulation after the fact through the post-ovulatory rise, which makes it useful for learning a pattern and useless for acting on the current day.
• Serial ultrasound. The reference standard for the actual day of follicle rupture, and the comparator a protocol published in Medicina 2023;59(9):1513 sets out for validating at-home urine hormone monitoring in regular cycles, PCOS and athletes. That paper is a study protocol, so it reports a design rather than results.

An app validated against its own algorithm has been validated against nothing. The question to ask of any accuracy claim is what it was measured against, and in how many cycles.

On the evidence retrieved, yes, though the margin is measured in days rather than orders of magnitude. In 1,155 ovulatory cycles from 964 participants, with ovulation prediction kits as the benchmark, a wearable-physiology method detected 1,113 of 1,155 ovulations with a mean error of 1.26 days, against 3.44 days for the calendar method applied to the same cycles (Thigpen, Patel and Zhang, Journal of Medical Internet Research 2025;27:e60667). Accuracy fell for abnormally long cycles, to a mean absolute error of 1.7 days from 1.18.

The finding that matters most in that study is about who the calendar method fails: it performed significantly worse in participants with irregular cycles, while the physiology method did not differ by cycle variability. Participants were recruited from the manufacturer's own commercial database, which is a limitation the authors state.

This is where calendar prediction stops being merely imprecise. Three separate failure modes are documented:

• No surge to find. In the Johnson cohort, no LH surge was detected at all for 99 of the 949 women during the cycle studied. An app extrapolating from period dates will still produce a confident fertile window for such a cycle.
• Normal-looking length, abnormal luteal phase. In women with polycystic ovaries who reported regular cycles, median cycle length was 28 days and cycle-length variation was no greater than in fertile women with normal ovaries — but early luteal phase progesterone was significantly lower (Joseph-Horne et al, Human Reproduction 2002;17(6):1459-63). A cycle length an app is happy with does not confirm that ovulation went well.
• Screening features that have not been tested on people. The irregular-cycle feature built into one large tracking app to flag PCOS risk was piloted on 9 virtual test subjects, over-predicted risk compared with a physician, and was reported explicitly as work done before empirical testing on human subjects (Rodriguez et al, JMIR Formative Research 2020;4(5):e15094).

Persistently irregular cycles are a reason to be assessed rather than to track harder. How PCOS affects conception covers what the diagnosis does and does not change, and when ovulation tests read positive without ovulation covers a failure mode no app can see.

The honest answer is that regular intercourse removes most of the need for precision. The ASRM and SREI Practice Committees state that intercourse every 1 to 2 days during the fertile window yields the highest pregnancy rates, and that results with less frequent intercourse — two to three times weekly — are nearly equivalent; they advise couples not to restrict frequency when attempting conception (Fertility and Sterility 2022;117(1):53-63).

Read alongside a six-day window, that is the practical conclusion. Intercourse every two to three days across the cycle cannot miss a six-day window, and needs no prediction at all. An app earns its place if it helps you notice a pattern, prompts a hormone test at roughly the right time, or surfaces cycles that look nothing like each other — not by naming a day.

The underlying biology of that window is set out in what the fertile window is and how to track it, and the hormone results an app is indirectly guessing at in what FSH, LH and estradiol actually measure.

Several claims made for and against these apps are not supported by the sources here.

• That using a fertility app raises the chance of conceiving. No retrieved trial randomised couples to an app against no app and measured pregnancy. The accuracy studies measure prediction, not outcome.
• That any app is accurate enough to avoid pregnancy on the same footing as contraception. The one large contraceptive dataset retrieved covers a temperature-based app used for that purpose and reports a typical-use Pearl Index of 6.9 pregnancies per 100 woman-years against 1.0 for perfect use, in a cohort of 22,785 users — with 12-month discontinuation of 54%. That is a different product used for a different purpose, and it was authored by the company that sells it.
• That the 21% figure applies to every app. It was measured against the cycles of those 949 participants, on the cycle-tracking apps available at the time, none of which published a methodology, which is itself part of the finding.
• That sensor-based apps are accurate in PCOS. The wearable study reported no loss of accuracy with cycle variability, but did not enrol a diagnosed PCOS group, and the ultrasound-validated PCOS comparison exists so far only as a protocol.
• That app-recorded data are private. No source retrieved for this article assessed the data-handling practices of these products, so nothing here supports a claim in either direction.

What is established is narrow and usable: the fertile window is six days and ends on the day of ovulation, cycle length alone cannot locate it, apps reading current-cycle signs outperform apps reading past cycle lengths, and intercourse every two to three days makes the prediction question largely moot.