Method · Updated October 2026
How we test and measure AI girlfriend sites
Every brand profile and the site ranking follow the same three steps: a paid, hands-on test; a full reading of what users publish; and the same set of numbers for every site, refreshed monthly. This page lists each step, each indicator, where it comes from, and the rules we read it by.
Step 1
We buy a membership and test it ourselves
Every profile starts with a paid membership. We use the platform for weeks and work through it feature by feature: chat and roleplay, memory across sessions, image generation, voice, video where it exists, the character library and customisation, the free tier and its limits, the pricing page, the token or credit system, billing, and the cancellation flow. That tells us what the product does.
What the test is not. It is not the verdict. One reviewer’s taste is a thin basis for judging a platform used by millions, so the hands-on test is used to agree or disagree with what users say, feature by feature, and the page says which it is each time.
Step 2
We read what users publish, in full
Sources. The most-watched YouTube videos about the brand, read through their transcripts rather than their titles; the Reddit threads Google surfaces for the brand, read with their top comments, upvotes and comment counts as they stand that day; and the press articles Google News lists, read in full. Google’s own AI Overview of the brand and the searches people type next to its name are read as well.
Classification. Each video, thread and article is classified after reading as positive, mixed, neutral or negative. A tutorial, a walkthrough or a passing mention counts as neutral. Affiliate content is read and labelled as such. Threads that are not about the brand are dropped. The counts on each page (so many positive of so many read) come from this classification and nothing else.
Weighting. Upvotes and dated, specific complaints count for more than reply counts, because this category’s Reddit threads attract promotion. A verdict that only appears in affiliate content is discounted until it shows up somewhere less tidy.
Features, one by one. From these sources each profile records what people say about each feature, and whether the hands-on test agrees, partly disagrees or disagrees. Disagreements are stated plainly, with the reason.
Step 3
We measure every site with the same indicators
The same indicators, in the same order, from the same sources, so profiles can be compared side by side and the ranking is never a matter of taste. Every figure on a profile names its source and the month it covers.
| Indicator | What it tells us | Source | Refreshed |
|---|---|---|---|
| Organic search traffic, worldwide and by country | Size, and the direction of travel. Estimated visits from Google results each month, 24 months back. | DataForSEO Labs | Monthly |
| Brand search volume, US and worldwide | Demand: how many people type the brand's name into Google each month, 12 months back, plus cost per click. | Google Ads keyword data via DataForSEO | Monthly |
| Google US position for "ai girlfriend" | Where the site stands for the category's own search in United States results; one of the figures the site ranking is built on, read together with the all-traffic rank and the rest of the data. | DataForSEO ranked keywords | Monthly |
| All-traffic popularity rank | Reach beyond search: a rank for every domain on the internet built from DNS and browser data. The site ranking uses it. | Tranco | Daily; shown monthly |
| Popularity tier | A cross-check on Tranco from real Chrome usage: top 1,000 to top 1,000,000. | Google Chrome UX Report | Monthly |
| Position for the category's own search | Whether the site is found by people who have not chosen a brand yet. We check “ai girlfriend”, United States, desktop. | Live Google results via DataForSEO | Monthly |
| Ranking keywords and branded share | How much of the site's traffic comes from its own name versus generic searches. | DataForSEO Labs | Monthly |
| Referring domains and backlinks | Staying power. We weight referring domains over raw backlink counts, which affiliate networks inflate. | DataForSEO Backlinks | Monthly |
| Domain Rating | A 100-point measure of backlink strength, shown as “Domain Rating by Ahrefs” under its public licence. | Ahrefs (public endpoint) | Monthly |
| Google Trends interest | Relative interest in the brand name over 12 months and 5 years, plus the rival names rising next to it. | SerpApi | Monthly |
| Videos, press, AI Overview, search suggestions | What is being published about the brand, and what people type next to its name before they sign up. | SerpApi (YouTube, Google News, Google) | Monthly |
| Reddit threads | The threads Google surfaces for the brand, with upvotes, comment counts and top comments as they stand that day. | Reddit, read directly in a browser | Monthly |
| Community size | Members and followers of the official accounts, where they can be read. | Discord, TikTok, Facebook public pages | Monthly |
Reading the numbers
The rules we read them by
Trend before level. A site at 500K visits and falling is a different story from one at 300K and doubling. Each profile reports the month-on-month, quarter-on-quarter and year-on-year change, and says when a year-on-year figure is flattered by a low base.
Demand against traffic. When brand searches keep growing while search visits fall, the site has a ranking problem, not an audience problem. When both rise together, that is the healthiest pattern there is.
Popularity is a rank, not a visit count. Tranco measures reach across every channel, including direct, apps and social. We never convert it into visits and we compare it only with the other AI girlfriend sites we track.
Referring domains over backlinks. A backlink count that multiplies while referring domains stay flat means a few domains linking from every page, the signature of affiliate networks. Referring domains are the staying-power indicator.
The category search is the comparison that matters. Shared-keyword lists mix in generic terms, so the one head-to-head we trust is the brand’s position for “ai girlfriend” itself, checked live.
Reputation explains the shape of the curve. What users praise drives brand searches; what they criticise drives churn, which shows up after the visit. Each profile ends with a reading of the numbers as a story, and a final verdict with the condition that would change it.
Independence and limits
What to know before you rely on it
Estimates. Organic traffic and search volume are estimates modelled from ranking data; they are consistent across sites, which is what a ranking needs, but they are not analytics from the site itself. Google rounds search volume into bands, which is why months often repeat the same figure.
Licences. Ahrefs data on this site is limited to Domain Rating from its public endpoint, shown as “Domain Rating by Ahrefs”. We do not publish SimilarWeb figures because its estimates for the same site vary widely between articles.
Money. Memberships are paid for by us. Some outbound links are affiliate links, disclosed on the disclosure page; they never change a figure, a classification or a verdict, and no platform can pay for a profile or a position in the ranking.
Corrections. If a figure is wrong or stale, tell us. Each profile shows the date every figure was pulled.