AI Girlfriend Market Size: Why the Numbers Disagree

Updated 2026-10-05. AI girlfriend guides

There is no defensible market size for AI companions, and this page no longer pretends there is one. It previously reported a market of 1.2 billion dollars in 2026, up from about 200 million in 2022 and projected to 2.5 billion by 2028; a top-five revenue concentration of 75 percent; roughly 30 percent share for a single platform; 250 million dollars of venture investment across 40 deals in 2025; subscriptions at 80 percent of industry revenue; and monthly-active-user bands for three tiers of platform. Every one of those figures was invented. They have been removed rather than replaced with a fresher set of guesses.

What is left is the part that is actually useful: what a market-size number would have to mean to be worth quoting, why the estimates in this category disagree by an order of magnitude, which parts of the picture you can verify by using the products, and what the one funding figure we can genuinely source does and does not tell you.

Key takeaways

What a market size number would have to mean

The scope problem is upstream of everything else. "AI companion market" can mean companion-specific apps, all character-chat platforms, all consumer generative-AI spending, or adult-oriented AI chat. Those are four defensible readings of the same phrase and they produce four completely different numbers, and headline figures routinely mix them without saying which they used.

A dollar figure requires a revenue definition. Gross consumer billings or net of app store fees? Subscriptions only, or subscriptions plus in-app purchases, advertising on free tiers, and any licensing revenue? Include one-time purchases? Define "paying user" as someone who has ever paid or someone paying this month? Each of these is a modelling choice, and each moves the total substantially.

A user figure requires a user definition, which is a different problem, covered on our page on how many people use AI companions. A market size in users and a market size in dollars are separate claims, and a document that gives you both without definitions for either is presenting arithmetic as research.

A number without a stated scope is not a measurement. It is a number with a hidden assumption in it, and the assumption is where the disagreement lives.

Why the published estimates disagree

Firm estimates for this category do exist. They disagree by roughly an order of magnitude, and they disagree because they are built on different methods measuring different populations.

Download modelling infers installs and revenue from sampled data points. It excludes web entirely, which matters more here than in most consumer categories.

Web analytics capture the app-invisible half of the category and miss the app half, and they are distorted by shared devices and shared network addresses.

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

Survey extrapolation is self-selected, sensitive to question wording, and small relative to the size of the claim built on it.

The structural problem is compounding. Each method carries an error, the errors survive into the total, and the total is then quoted without error bars — which is why a range printed without its method reads exactly like a point estimate.

And the sources are often not independent. The same unattributed figure appears across many pages in this category. When several sources agree, the first question is whether they are measuring separately or restating each other. Repetition is not corroboration, and the market-size claims you will encounter most confidently online are frequently the ones with the least provenance.

Why market share is the weakest claim of all

A market-share figure needs two things: total category revenue, and one company's revenue. Neither is audited, and most of the significant players are private.

So a share like "the top five platforms capture 75 percent of revenue" is not a measurement. It is a narrative about consolidation, worked backwards from partial data, partial company disclosures, and an assumed category total. The "30 percent" attached to a single platform on this page previously was the same invention wearing a company name.

A company's own monthly-active-user number is a marketing figure until a filing or a disclosure defines "active." It is the only number available for that company, and it is not the same thing as its share of a category nobody has measured.

Revenue model: the part you can actually check

The dominant model is freemium subscription, and this one is verifiable rather than estimated, because you can read the pricing page of any product in the category. A limited free tier, paid tiers above it, and an annual option is the pattern across the category.

Prices are published. What a specific product charges is a fact with no interpretation layer between it and reality. What an "industry average price" means is not, because what sits inside each tier differs between products, and the tiers are not aligned across them.

Premium messaging limits are rate limits, not unlimited access. This page previously described premium tiers as including "unlimited messaging." That is not accurate for these products, and we would not state it for the category either: paid tiers characteristically impose message caps, with higher or removed caps at higher tiers, and free tiers are more restricted still. Caps change, so read the current pricing page rather than trusting any description of it, including this one.

The other revenue lines exist in some products — in-app purchases, advertising on free tiers, and in some cases claimed licensing of anonymised conversation data for model training. Nobody publishes what share of revenue each represents. Any percentage attached to them in this category should be treated as invented, including the "80 percent from subscriptions" this page previously asserted.

Conversion, revenue per user, and lifetime value are internal metrics at every company in the category and are not disclosed on a comparable basis.

Investment: the one thing that is genuinely reported

Funding rounds are real events with public documentation, which makes them the exception in this page. One is worth stating precisely, because this page previously misstated it.

Character.AI announced a 150 million dollar funding round in March 2023 at a 1 billion dollar post-money valuation, led by Andreessen Horowitz. The company described it as a Series A, and it was reported at the time by Reuters, the Wall Street Journal, the New York Times, and Bloomberg. This page previously described it as a Series B at a 2.5 billion dollar valuation; both the round label and the valuation were wrong.

Two things that figure is not. It is not market size — a valuation prices one company's equity at one moment, and the company reported little or no revenue at the time, which the contemporaneous coverage noted explicitly. And it is not a measure of demand. It is a fact about 2023 funding markets and about one company's private round, and it should not appear in an argument about category revenue.

Category-wide funding totals are not available on a consistent basis. A figure like "250 million dollars across 40 deals in 2025" depends entirely on which startups the counter classifies as companions, and that classification is a moving definition rather than a stable one. We are not going to produce one.

We are also not going to enumerate patents or imply market entry from them. Patent filings are inexpensive, their scope is ambiguous, and a filing is not a product decision. The previous version of this page cited filings as evidence of corporate interest; we have removed it because it is not evidence of anything a reader could act on.

Competitive structure

Platforms in this category differentiate along a small number of dimensions, and all of them are observable by using two products side by side: model quality and coherence, memory depth and reliability across sessions, content-policy flexibility, feature breadth across voice, images, and video, whether the product works on web as well as in an app, and price. This is the part of the competitive picture that is genuinely knowable.

The tier structure with per-tier monthly-active bands that this page previously carried is gone. Those bands were invented, and there is no equivalent published figure to replace them with.

Compute cost is a real and directional pressure. Per-message inference cost scales with usage, which is one reason paid tiers carry caps at all and one reason the economics of a heavy user are harder than those of a light one. The direction is knowable from published pricing structures; the magnitude is not published.

Consolidation claims require data that does not exist. Whether larger players are acquiring smaller ones at any particular rate is not measurable from outside, and the previous version of this page asserted a trend toward consolidation without a source.

What you can do instead of quoting a market size

For revenue model and pricing, use the products. Published price pages and tier contents are the only figures in this category with no interpretation layer.

For adoption, use app store rankings and review volume. Both are audited and public, and both give a defensible relative signal for the app-based subset only, with web usage entirely outside the frame.

For investment, read the announcement and the reporting. Round label, amount, and valuation should come from the same report, and the date is part of the fact.

For category size, assume nobody knows. If a number is load-bearing in your work, it is an assumption you have chosen, not a measurement you have found, and it should be labelled that way.

Frequently Asked Questions

How much is the AI girlfriend market worth?

We cannot tell you, and the disagreement between published estimates is the most informative fact available. They differ by an order of magnitude because they are measuring different things. The section above sets out what a defensible figure would need to state; this page previously provided one that stated nothing.

How do AI girlfriend platforms make money?

Predominantly through freemium subscriptions, which you can verify by reading any product's pricing page. Some products add in-app purchases, advertising on free tiers, or claimed data licensing. What share of revenue each represents is not published by anyone, so any percentage you see attached to those lines in this category is unsourced.

Who are the major investors in AI companion companies?

We are not going to give a category-wide investment total, because the set of startups counted as companion companies is a moving definition and no consistent series exists. The one round we can source is Character.AI's March 2023 raise of 150 million dollars at a 1 billion dollar valuation, and it tells you about that company's financing rather than about the size of this market.

Which platforms dominate the market?

Nobody can answer this from public information. A share requires both a category total and per-company revenue, and neither is audited. The "top five control 75 percent" and single-platform share claims previously on this page were invented, and the plausible direction of travel toward consolidation is a hypothesis rather than a measurement.

What are the biggest risks to market growth?

Structurally: regulatory attention to consumer AI systems generally, reputational backlash that can move quickly, the difficulty of retaining users once novelty decays, compute costs that scale with usage, and competition from general-purpose assistants that add companion features. We are not going to forecast which of these binds or when, because that would require the market and revenue figures we do not have.

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.

Related Articles

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