Methodology

Comparable prices

We convert all prices to an effective monthly price, so subscriptions with different billing periods (monthly, annual, two years) can be compared fairly. We calculate in cents, without rounding errors.

Promo price and renewal price

We clearly separate the introductory (promo) price from the renewal price. We show how much you pay in the first year and which price applies once the promotion ends — especially important for VPNs and similar services, where the renewal price is often much higher.

Total contract cost

For carrier bundles we calculate the cost of the whole contract — with activation fees, equipment fees and the price after the promotion; the first promo month is only part of it.

Check status

• Checked — there’s an official source and a check date.
• Likely current — checked recently, but without a complete official source.
• Not confirmed — no check yet.
• May be out of date — set automatically once a certain number of days has passed (depends on the category).
• Not available right now — the provider doesn’t offer the service or plan for a US account, so the price isn’t missing because of us.

How the order is decided

We don’t hide how the order is decided. It’s determined only by the criteria below — affiliate links or partnerships don’t affect placement, and we don’t charge for a spot in the list.

• Plans within a service — sorted by lowest effective monthly price (including free months).
• Services in lists — within a category by our score: higher scores higher up, services without a score (data not collected yet) after them, and services without plans last. The home page cards follow the same order. You can sort by price directly in the list. Commercial motives play no part in this order.
• Comparison table with competitors — the column order follows the subject of the review: first the product being described, then the competitors chosen for comparison. We use no “winner badges” and don’t write that any VPN is best for everyone.
• The plan with an accent-colored border — that’s the plan the provider itself has marked as most popular on its official page. It’s not our recommendation.

How we score

A score should only be shown together with what it’s made of. That’s why the criteria and their weights are listed below for every category we score — VPNs, music, cloud storage, AI services, online courses, streaming services, fitness and health, gaming subscriptions and antivirus software. Each category has its own criteria: asking a VPN and a music service the same questions would be meaningless, and one shared list would mean half the rows were always empty.

The score is calculated for each market from that market’s own data: servers in the country, student plans in the country, adaptation to the market. If such data is missing, the criterion isn’t counted in the average. We don’t take it from another market: a translated score would look just as convincing but measure the wrong market. For the US, the score is based only on data from the US market, and where that isn’t enough, no score is shown.

The score runs from 1 to 5 and is a weighted average of the criteria below. It’s calculated for each plan separately, and next to the service the average of its plans is shown: NordVPN Basic and NordVPN Prime are neither the same product nor the same price, and one shared score would be misleading for both.

How the score is calculated

Each criterion gets a value from 1 to 5, which is multiplied by the criterion’s weight. The sum is divided by the total weight of the criteria we have data for. If, for example, a criterion weighing 10% is missing, the sum is divided by 90 instead of 100 — that’s how much weight remains with the criteria we know.

Two things follow from that. First, a low price doesn’t lift the score on its own: price is one criterion among several, and a plan can get 5 for price and still land in the middle. Second, a gap in our data doesn’t penalize the plan; it only reduces the number of criteria counted.

Criteria for VPN services

Criteria for music services

A different list and a different most important criterion. Price weighs 10% here: where prices are close together, a higher weight would turn the score into a contest over a few dollars. Audio quality weighs the most, because it’s the only place where you get a different product for the same money: from 24-bit/192 kHz down to the 256 kbps ceiling.

Catalog size is deliberately left out. The big providers list catalogs of “over 100 million songs”, but the number comes from them and can’t be compared. A criterion where everyone would get 5 doesn’t separate the scores, it only dilutes them.

Criteria for cloud storage

Here price is measured per terabyte: an amount without storage space says nothing. A plan with 100 GB and one with 2 TB are simply different sizes; comparing their prices would be like comparing the price of one apple with the price of a whole sack. Encryption weighs the most: it’s the only difference money can’t make up for.

For Microsoft 365 Family we count the storage as 2 TB: that’s the shared storage new subscriptions get for up to six people. The six people appear on their own row, where they belong.

Criteria for AI services

Here the score measures the purchase terms and what the service can do; it doesn’t judge model quality, and we say so openly. Three of five providers describe how their limits work, but not the number of messages: Claude Pro and Google AI Pro — five-hour windows and weekly limits, Midjourney — hours of fast GPU time per tier. ChatGPT only says the limits are dynamic and depend on load, and Perplexity Pro doesn’t publish some of them. On top of that, the published limits are expressed in different units, so “how much you get for the money” can’t be compared between these services to our checking standard. A score pretending to measure that would be worse than no score at all. So we assess what can be checked and what costs money: what the service can do weighs the most (20%), and whether the provider publishes any limits at all is assessed separately (10%).

Business tiers aren’t part of the scale — that’s a different product for a different buyer.

Criteria for online courses

In this category one criterion is deliberately removed — refund terms: for half of the providers we didn’t find them, and we don’t hide that. The certificate is a criterion (20%), but the question is recognition: almost everyone issues certificates, the difference is whether anyone outside the platform recognizes them.

Criteria for streaming services

Content you can’t get for free weighs the most (18%). With a movie and TV subscription you’re buying content: originals from Netflix, HBO Max or Disney+ can’t be watched without a subscription, and TV providers add premium channels and libraries. YouTube Premium is a different case: the same videos are also available for free, and the subscription removes ads and adds features — background play, downloads, YouTube Music Premium. That’s useful, but it’s convenience, not content, so YouTube Premium gets 1 on this criterion and the other services 5. Without this criterion the score measured only features, and YouTube Premium ended up above the movie platforms. The other streaming criteria were reduced proportionally; market adaptation and the price criteria are unchanged.

The hardest decision in this category is adaptation to the local market, and it weighs 15% — more than price. The first version of the score didn’t have it at all, and the result was illogical: the local provider with its own channels and library ended up behind the international catalogs. The score measured only how the international catalogs differ from each other, and not at all what makes viewers choose a local provider.

We don’t state the share of localized content and won’t — no provider publishes it, and our own estimate would be a guess. The level, however, can be described, and where we have it, it’s shown next to the product with a source; where the data is missing, the criterion isn’t counted in the score.

We dropped two criteria. Catalog size we don’t assess, for the same reason as for music: the numbers come from the providers themselves and can’t be compared. Restriction to one household is confirmed by only half of the providers, and silence doesn’t let you conclude that “there’s no such requirement”: silence isn’t a denial. Download limits are measured in different units — “up to 100 titles” and “up to 10 devices” can’t go on the same scale — so the “Watch offline” criterion (7%) doesn’t count them: it only answers whether content can be watched offline at all and whether only on a phone or tablet.

Price gets 10% here — no more than in other categories, and that’s deliberate: the criterion would compare a bundle with TV channels, sports and other subscriptions with a single catalog as if they were the same product at a different price. A high weight for price would penalize the bundle for containing more.

Criteria for fitness and health

The only category where the subscription price isn’t the whole price. Garmin Connect+ without a Garmin watch and Google Health Premium without a Fitbit or Pixel Watch have nothing to read data from, while with WHOOP it’s the other way around — the device is included in the membership. That’s why the “Subscription is enough” criterion weighs 25% — more than the two price criteria combined (20%): it’s an extension of price.

There’s deliberately nothing here about measurement accuracy or health benefits. Those are medical claims we can’t back up, so we don’t make them on this website — neither positive nor negative.

Criteria for gaming subscriptions

The only category where a subscription can be a requirement. Without PlayStation Plus or Nintendo Switch Online, some games you already bought simply don’t work online. For someone choosing, that’s more important information than catalog size: you pay for what you already have to work; the content is secondary here. That’s why online play with others weighs as much as access to the newest games — 25% each.

We don’t assess catalog size, even though it’s the first thing that comes to mind. There simply is no number that can be compared honestly: providers don’t list it consistently, and catalogs change every month. If we wrote “about 400” next to “about 100”, the score would rest on numbers nobody has verified. So we chose the criteria we have data for.

Where it works weighs 20%, because it decides whether the subscription works with the equipment you already have. This criterion really separates the scores: the services land on different levels — from “console only” to “console, PC and cloud”. A criterion everyone would answer the same way we don’t include at all.

Criteria for antivirus software

Antivirus protection is measured by an independent lab. In the other categories we have to go by what the provider sells — how many devices, whether there’s encryption — because there’s no independent measurement. For antivirus software, AV-Comparatives does that: three criteria come from the lab’s latest tests — offline protection, online protection and false positives.

Offline protection weighs the most (25%), online protection 15%. They’re two separate criteria, because a single number would be misleading here: the same program can be among the best when there’s a connection and let almost one in seven threats through without one. A laptop on a flight or during an internet outage has only what’s installed, so a program that protects equally well in both cases should rank ahead of one that’s only strong when connected.

False positives (10%) aren’t a minor thing: either they block a clean program, or they teach you to click “allow anyway” — and then a real warning doesn’t work either. The thresholds are the groups the lab itself published in the same test; we don’t make up our own.

Price is compared here per device: the same provider’s plans for one and for ten devices differ several times over purely because of how many devices they cover, so a median of plan prices would measure plan size. Separately we assess the second-year price increase (10%) and the number of protected devices (10%).

What’s not included: the AV-TEST scores for performance and usability — for the products we assess they’re already at the ceiling (in the June 2026 round, all six got 18 out of 18 points), so they wouldn’t separate anything. Add-ons — VPN, password manager, parental controls — we also don’t assess as a separate criterion: they appear in the plan comparison table, and they don’t measure protection.

And five rules without which the weights wouldn’t mean anything:

• A criterion without data drops out — it doesn’t become zero. Zero means “checked, doesn’t exist”, and we don’t know that — those are two different things. Such a criterion is removed, and its weight is spread over the others. Otherwise the score would measure the gaps in our data: the provider would get a lower score because we didn’t find something about it. That doesn’t mean it’s worse.
• If data is missing for more than four criteria, we don’t show a score at all, and for VPNs the limit is stricter: more than two. Past a certain share of gaps, the number stops being a score and becomes a guess with decimals.
• Speed isn’t part of the score. It’s stated by the providers themselves, and independent measurements change every month. Putting it into the number would mean putting in a guess, so speed stays in the review text, as a named claim with a source.
• Streaming via VPN is assessed by what the provider offers. We check two things: whether there’s Smart DNS (then the service can be used on a TV or console where there’s no app) and whether there are servers meant for streaming. Whether a specific platform can be unblocked right now isn’t part of the score — that would require a subscription with every provider, and the answer would be valid for one day.
• This score goes into structured data — but only it, and only where we show it. We publish it as reviewRating with a clearly stated 1–5 scale: it’s our own rating of another party’s product, with a named author. App store ratings aren’t in there and won’t be — Google’s guidelines explicitly forbid it: “Don't aggregate reviews or ratings from other websites”. On the page they’re still visible, with the store name, number of ratings and check date.

Every criterion in the score is shown on the review page with what we actually found, with a link to the official source. If you disagree with the weights, you’ll see all the parts there and can do the math your own way.

The score is determined only by the published criteria: partners don’t pay for placement, and providers don’t see the text before publication. The reviews are written by a named person — the author is listed with every text.

Data from app stores

We use public data from the Apple App Store and Google Play: user ratings together with the number of ratings. We state which store the data comes from. We don’t show the rank in the store’s category list — we haven’t collected that data, and writing about something we don’t collect would be an empty promise.

Important: this is not market share and not a general top list of the “most popular” services. We don’t use the data to decide the order of services — it’s shown only as extra context. We don’t write reviews ourselves and don’t change them. When we summarize them, we do it in our own words and state the store and date.

They do go into our score — and more than once: the app rating is a criterion in seven of the nine categories on this page (not for gaming and antivirus). The weight differs several times over between them, so we don’t repeat the exact number here — it’s in each category’s criteria table, and that’s the only place it’s worth reading. It’s the only part of the score where users themselves have a say; they rate the app, and their rating says nothing about whether the subscription price is fair.

When there are two stores, we show no combined value: each store’s row already has its own number of ratings, and a number merging them adds nothing to the decision — it only hides which of the two opinions carries more weight. In the score, the app rating is weighted by the number of ratings: a plain average of, say, 4.8 (600 ratings) and 4.5 (300,000 ratings) would give 4.65 — a number where six hundred people weigh as much as three hundred thousand. And we don’t put these ratings into the structured data search engines read: Google’s guidelines explicitly forbid it, because it’s data from other websites.

Where the price is missing

Every price shown has been checked for the US — on the providers’ official pages or at checkout. Where we show no amount, there’s a specific reason, and it’s stated on the page itself: either the provider doesn’t publish the price (then we say where you can see it), or the service isn’t offered for US accounts at all. We never allow ourselves to guess. If the data temporarily can’t be loaded, no prices are shown at all — better nothing than a wrong price.

Report an error

Found an inaccuracy? You can report an error on every service page. See also our corrections policy or write directly via the contact page.

What doesn’t affect the methodology

Nobody can buy the order or the scores — neither providers nor partners. How that’s ensured and what the editors don’t allow themselves is described in the editorial policy. What the amounts shown mean legally and what we’re not responsible for is in the terms of use.