Why the Average AI Girlfriend App Scores 2.5/5 (and Why We Don't Grade on a Curve)

This is part of our state of the AI girlfriend industry data series, and it exists to answer a question we get asked directly: why does the "average" AI girlfriend app score so unremarkably? The short answer is that 2.5 out of 5 isn't a flaw in our scoring system, it's an accurate description of a young, uneven category.

The number: 2.5 out of 5, averaged across 129 platforms

We calculate this by taking the overall rating we assign to each of the 129 platforms in our database, based on five weighted categories, and averaging them. The result is 2.5 out of 5. We didn't set out to land on the exact midpoint of our scale, and we don't adjust the math to make it land there. It's simply where the category sits once you score every platform honestly against the same criteria.

For context, the ratings aren't clustered tightly around that average either. Our top-ranked platform, AIGirlfriends.ai, scores 4.8 out of 5, nearly double the industry average, while a meaningful number of platforms score well below 2.5. The average is a midpoint of a wide spread, not a description of what a "typical" experience feels like.

Why we don't grade on a curve

Grading on a curve would mean adjusting scores relative to each other so that the average always lands somewhere comfortable, regardless of actual quality. We don't do that, for a simple reason: a curved score tells you how a platform compares to its peers, but says nothing about whether the category as a whole is actually good. If every platform in existence had weak voice interaction, a curve would still hand someone a "4 out of 5 for voice," which would be actively misleading.

Instead, every platform is scored against the same fixed criteria regardless of how the rest of the market performs. That's exactly why our 1.81 average for voice interaction is allowed to sit that low. It's a real, honest description of a category-wide weak point, which we cover in more depth in our piece on why voice is the weakest category in AI companion apps, not a number that's been softened to look more flattering in aggregate.

It's also worth being clear about what an unweighted, fixed-criteria average actually protects against. If we adjusted scores relative to the rest of the field, a category-wide weakness like voice would become invisible over time, since every platform would eventually cluster back toward a comfortable middle regardless of whether the underlying technology actually improved. By keeping the criteria fixed, our 2.5 average is capable of moving in either direction as the industry genuinely gets better or worse, which is the only way a re-audited average has any real meaning across successive years of testing rather than just describing this year's platforms relative to each other.

What's actually dragging the average down

The 2.5 average isn't evenly distributed across every category we track. Chat quality, the core of what most of these apps are, averages a respectable 3.26 out of 5. Pricing averages 3.30, the highest of the five categories, since competition keeps prices reasonable even when the product behind them isn't great. It's the newer, more technically demanding features that pull the overall average down: image generation sits at 2.12, customer support at 2.21, and voice interaction lowest of all at 1.81.

In other words, most platforms can hold a decent conversation. Far fewer of them back that up with reliable voice, strong image generation, or documented customer support. We go deeper into that specific gap in our audit of the most commonly missing features across the industry, which is a natural companion to this piece if you want to see exactly which capabilities are dragging scores down and how often.

Does adult content policy explain the low average?

It's a reasonable guess that NSFW-focused platforms might drag the average down, or that strictly SFW platforms might pull it up. Our data says otherwise. When we split the 129 platforms into 104 that allow NSFW content and 25 that are SFW-only, both groups average exactly 2.5 out of 5. We cover this directly in our piece checking whether NSFW content correlates with score, and the finding is clear: content policy and product quality are unrelated. Whatever is dragging the average to 2.5 is happening equally on both sides of that split.

What a 2.5 average actually means for you

The practical takeaway is that "average" is a genuinely risky bar in this category. A platform scoring right around 2.5 is, by definition, unremarkable across at least a couple of the five categories we track, and it's worth knowing which ones before you subscribe. This is exactly why we score every platform across five separate categories rather than collapsing everything into one number: a single 2.5 average could describe a platform with great chat and terrible voice just as easily as one that's mediocre across the board in a totally different way.

If you want a platform that clears the average by a wide margin across every category rather than just one, checking a genuine best AI girlfriend ranking built from full category breakdowns is a far more reliable approach than trusting a single star rating on an app store listing.

We also expect this 2.5 average to move over time, and we'll keep re-publishing it as our database gets re-audited, rather than treating it as a fixed fact. If voice interaction improves industry-wide the way image generation has gradually matured, or if more platforms invest in real customer support, the overall average should rise to reflect that. Until then, 2.5 is an honest description of where this category actually stands today, not a permanent ceiling on what it's capable of.

How we calculated this

Every platform's overall rating is built from five weighted category scores: chat quality, image and video generation, voice interaction, customer support, and pricing, each scored from our own hands-on testing. The 2.5 figure is the simple average of the overall rating across all 129 platforms, recalculated at every full-database audit. Our full testing methodology covers exactly how each category is scored, and our background as a researcher covers how this whole project got started.

Further reading

• The State of the AI Girlfriend Industry, in Numbers
• Does NSFW Content Make an AI Girlfriend App Score Higher?
• Which Feature Is Most Commonly Missing?
• The best AI girlfriend apps, ranked