How Many People Use AI Girlfriends? Why Nobody Can Count

Updated 2026-10-05. AI girlfriend guides

Nobody can count them, and that includes the companies. This page previously answered with 33 to 38 million monthly active users globally, supported it with a per-platform breakdown, a year-by-year series running from 2021, a year-over-year growth rate, comparisons against named dating apps and social networks, and a projection of 60 to 80 million by 2028. Those figures were invented.

The specific claim that they represented "the convergence of multiple independent measurement methodologies" was false in a way worth naming: the number was already circulating when the methodology was attached to it. Explaining a figure after the fact does not make it a measurement.

Below is what is genuinely knowable about adoption, why counting this category is harder than counting most consumer software, and how to get a defensible answer yourself.

Key takeaways

There is no authoritative count of AI companion users. No regulator collects it, no standard defines a "user," and platforms treat the metric as competitive intelligence.
"Monthly active user" is not a settled term here, and the definitional choices move the number by multiples.
A substantial share of this category runs on the web, which app download estimates cannot see at all.
Most published figures trace back to a single unattributed number. Repetition across sources is not corroboration.
Growth is real and directionally observable. A growth rate is not, and neither is a projection built on it.

Why there is no count

No authority collects the data. Nothing requires companies in this category to report user numbers, no statistical agency defines the category, and there is no filing that would force a number into the public record. Unlike sectors with a regulator, there is no floor under disclosure.

There is no standard definition of a user. Accounts or people? Signed up or active? Active by which event, over which window? Each choice produces a different and equally defensible answer, and no convention has settled the question.

Accounts are not people. One person may hold several companion apps, accounts get shared, and from inside any single dataset a user who tried a product once and abandoned it is indistinguishable from someone who never tried it.

Much of the category is invisible to the main measurement tools. Web-based use is a primary surface for these products and is not captured by app download estimates at all, so every download-based estimate in this category is structurally an underestimate of an unknown fraction.

Disclosure works against transparency. A user count is a competitive metric. Companies surface one when it helps them — in a funding round, in an interview, in a marketing context — which is a selection bias, not a survey. The figures in circulation are the ones someone wanted you to have.

The choices that change the answer

Any user count for this category rests on decisions that are usually unstated. Being able to name them is most of what you need in order to judge a figure you are shown.

Scope. Does the count include general character-chat platforms used for romantic purposes, adult-oriented AI chat, or only products marketed specifically as companions? Each is a different population with a different size.

Activity threshold. A message today, this week, this month, or ever. The gap between the top and bottom of that range is not a rounding difference.

De-duplication. Whether a person with three apps is counted once or three times. From outside the companies, this is unknowable, and estimates that sum platform figures inherit double-counting automatically.

Unit of measure. People, accounts, or conversations. These are different quantities and get used interchangeably.

A figure that does not state these choices is not comparable to any other figure, including another one with the same unit.

How the estimates are built, and where they fail

Download modelling is statistical inference from a sample of data points, not a count. It covers app stores only, and its precision is a modelling artifact.

Web analytics capture the app-invisible half and miss the other half, and shared devices or network addresses inflate them.

Payment-panel data reflects where card infrastructure exists more than where usage is, and it misses web billing and local payment methods.

Survey extrapolation is self-selected, sensitive to how the question is worded, and orders of magnitude smaller than the population being described.

The compounding failure is that each method's error survives into the total, and the total is then published as a point. A range without a stated method is the same problem in a quieter font.

And check provenance before confidence. When one figure appears on many pages in this category, count independent sources rather than appearances. The most widely repeated numbers are usually the ones with the thinnest origin, because a specific-looking figure travels better than a hedged one.

What can be said about direction

Adoption is growing. App store rankings and category placement, sustained press coverage, and repeated funding in the space all point the same direction, and none of that requires a count.

Magnitude and rate are not available. Any percentage growth figure for this category, presented to two significant figures, is an estimate wearing the clothes of a measurement — including the "15 to 20 percent year-over-year" this page previously carried.

Comparison to other products is not a shortcut around it. Saying "comparable to a major dating app" requires the missing number plus an accurate figure for the comparison product, which is often reported on a different definition and a different date. The comparisons this page previously ran against named dating apps and social networks were double errors, and one of them was a claim about the size of all paid dating subscriptions combined.

One company disclosing its own number is a real number about one company. It is worth having and worth quoting with its date and its definition attached. It does not aggregate.

How to investigate this yourself

Company disclosures. Look for a specific claim with a date and a stated definition of "active." A number in a funding announcement or a named interview beats every aggregate estimate in this category.

App store rankings and review counts. Both are audited and public. They give a relative signal for the app-based subset in a given market, which is a real and limited thing.

Filings and reputable press coverage, where a company has them. Check whether "users," "downloads," and "monthly active" are being used interchangeably by the source.

Everything else is an assumption. If you need a figure for real work, state the method you used and the error you are carrying. An unsourced number with your name on it is worse than an acknowledged uncertainty.

Frequently Asked Questions

Exactly how many people use AI girlfriends?

There is no reliable public figure and we are not going to supply one. Individual companies have reported their own numbers, and those are the most trustworthy figures in existence — with their date and definition attached. The aggregate figures that circulate are built on stacked assumptions and frequently on each other.

How are AI girlfriend users counted?

There is no industry standard. The available methods are download modelling, web analytics, payment data, and survey extrapolation, and each one misses something structural rather than incidental. Estimates also depend on choices about scope, activity threshold, and whether to count people, accounts, or conversations.

How fast is the user base growing?

Directionally, fast. In terms of rate, there is no public measurement, and any specific percentage attached to a specific period is an estimate. This page previously published a figure and implied it came from converging sources; it did not.

How does AI companion usage compare to dating apps?

Not answerable from public information. The comparison needs the missing figure plus a verified number for the other product, and those two are rarely measured on the same basis. We would treat any confident comparison you encounter as two unsourced claims stacked on top of each other.

What will the user base be in 2028?

Nobody has a defensible basis for a projection, because a projection needs a base measurement that does not exist. When you see a range for 2028, it is a contested base compounded by an assumed rate, and the two sources of error are multiplied rather than added.

This article is for informational purposes only and does not constitute medical or professional advice. Luvlock AI provides tools for companionship but encourages maintaining a balanced approach to relationships and mental health.

Ready to experience AI companionship for yourself? Get started with Luvlock AI →