Frequently asked questions

A number from 0 to 100 for how well linked a domain is next to every other domain we know about. A site nothing links to sits near 0. The few sites almost everything links to sit near 100. Most real, working sites sit between 10 and 40, so that is where most of the useful comparisons happen.

It measures one thing: the links pointing at a domain. It is not a quality score, a traffic estimate, or a guess at how a search engine will treat you.

It is an independent 0 to 100 logarithmic rating based on the number and authority of unique linking domains, computed from our own web-scale link index and refreshed monthly.

Here is how. We start with a short list of sites we know are what they say they are: national broadcasters, universities, governments, standards bodies, old and trusted reference works. Each one gets a starting score. That score is then passed out across the web in rounds. A domain linked from a scored domain takes a share of what that domain holds, then passes on a share of what it now holds to the domains it links to. The rounds repeat until the numbers stop moving. Stopping there is what makes the result a fact about the whole web, not just about the few steps around one site.

All the way through we count separate domains, not pages. A site linking to you from ten thousand pages counts once, and what that link is worth depends on what the domain giving it holds by then. One link from a much linked domain moves the number more than a hundred links from domains nothing points at.

Out of that comes one score per domain, over a huge range. We sort every domain in the release by that score, and its place in that order is what we put on the 0 to 100 scale. That is the number you see.

Why this makes a high rating hard to fake. Score only ever enters at the hand picked sites. So the only way to score well is to be reachable, in few steps, from sites that were trusted to begin with. Linking to yourself gets you nothing, because a link makes no score, it only passes some on. A thousand domains built to link to each other share out whatever reached them from outside, and if nothing reached them they share out nothing. Buying links from domains that the starting list cannot reach is buying a share of zero.

Why it holds up as AI makes the web bigger. Making a site, or ten thousand sites, is close to free now, and the amount of content on the web no longer tells you much. What is still expensive is a link from a national newspaper, a university department or a standards body, because people with something to lose give those. We measure from that end of the web, so a flood of new domains barely moves the old ones: the new arrivals land at the bottom and stay there until somebody who counts points at them.

Equal steps up the scale mean multiplying, not adding. Going from 20 to 30 might mean a few more linking domains. Going from 70 to 80 means many times more. The gap between 90 and 100 is bigger than the whole gap between 0 and 50.

So read it as a ranking, not a percentage. Compare domains with each other rather than averaging a list, and remember that five points near the top is worth far more than five points near the bottom.

One whole domain name, cut back to its main part. Links from blog.example.com and shop.example.com are one linking domain, example.com. Endings made of more than one part are handled properly, so bbc.co.uk is one domain and not a piece of co.uk.

Good is relative, and that is what makes the number useful. Most domains that have a rating at all sit at the low end, so the marks to go by are: under 30 means few links from well known sites, which covers most new and small sites; 30 to 59 means a fair spread of other sites link to it; 60 and above means links from all over the web, which in practice means large publishers, platforms and institutions.

The useful question is almost never “is 34 good”. It is “how does this domain compare with the other two hundred on my list”, which is what the bulk dataset is for.

Every month. Each release scores every domain we cover again from scratch.

Because the rating is relative and the whole web moves. Your number can change because sites linking to you gained or lost links of their own, because links to you went when pages came down, because what we cover shifted, or because other domains moved around you. A one or two point drift between months is normal and usually means nothing.

The latest release rates over 200 million domains, which is most of the web that anybody links to. If a site has links pointing at it, it is very likely in there.

The exact count moves month to month, and every domain in a release is rated in that release. You can see the count for each one in the table on the data page.

Because it has not earned the links the rating measures. Hardly anything links to it, and what does link is not much linked itself, so it sits at the bottom of the scale. A domain can be real, live and busy and still be weakly linked: brand new sites, sites behind a login, sites that get their visitors from ads or social platforms, and company internal domains all sit there.

Every domain starts at 0 and climbs as well linked sites begin to point at it.

No. Ratings are per main domain. Ask about blog.example.com and you get the rating for example.com.

Judge it yourself before you pay. The free sample is 10,000 domains picked at random from the current release, as a CSV file with the same columns as the paid files, so you can put our numbers next to whatever you already use and see how they line up. Random rather than our best rows on purpose: a sample of the top of the table would tell you nothing about the smaller sites you will actually be sorting. No account needed.

The rating is worked out over a link index of more than 200 million domains and redone from scratch every month, which is what makes it steady enough to sort and filter a large list by. It describes the links around a domain and nothing else: it is not a traffic estimate, a quality score or a guess at how a search engine will treat you. It is provided as is, with no promise of accuracy, and nothing on this site is advice about the domains it describes.

Three a day without an account, twenty five a day with a free one. Each check takes a list of up to 10,000 domains, so three checks a day is thirty thousand domains a day. Without an account the limit resets 24 hours after you use it. Signed in, it resets at midnight UTC.

No, and there is not one to create. You enter your email address, we send you a link, and following it signs you in. The link works once and runs out after 15 minutes. We keep no password, so there is nothing to leak, reset or reuse.

No. This part is built, not just promised. A lookup is a single indexed read; the domain is never written to a log file, a database table, an analytics tool, a cache or an error report. Rate limits count against your session, never against a domain. The browser extension keeps answers in your browser's memory only, and they are gone when the browser restarts. The full policy is here.

Because it uses Chrome's tabs permission, the only way to see the address of a tab you just switched to without making you click the icon every time. Chrome calls that permission reading browsing history whatever an extension does with it. Ours takes the host name, cuts it back to the main domain, asks for that domain's rating, and forgets it.

Paste it into the box, one per line or split by commas, spaces, tabs or semicolons, or upload a .csv or .txt file. A pasted web address, a host name with a port and a trailing dot are all understood, and every entry is cut back to its main domain before it is looked up. Repeats are answered once, so the answer can be shorter than the list. Anything past 10,000 entries is left off rather than the whole list being refused, and the answer says so.

Monthly releases back to 2017, plus every new one published while your licence runs. Every domain is scored again in every release, so a domain is a line you can follow rather than a single number: compare any two months and you can see whether a site is climbing, sliding or standing still.

The free checker and the API give you the current release only. The history comes with the bulk data licence.

One parquet file per monthly release, plus every release before it back to 2017, plus every release published while your licence runs. Files are named after their month, for example 2025-12.parquet.

Each row is one domain, with these columns:

• domain (string) – the main domain
• rating (int16) – the 0 to 100 rating
• score (double) – the raw score before scaling, if you want to scale it yourself
• inbound_count (int64) – separate domains linking in
• outbound_count (int64) – separate domains linked out to

Rows are sorted by domain and the files are ZSTD compressed, so a sorted merge or a binary search over them is cheap.

Up to about 25 GB each, unpacked on disk. Plan for that when you write the download step: write to disk rather than into memory.

Anything in the Arrow family will open one. DuckDB needs no setup at all:

SELECT domain, rating
FROM read_parquet('2025-12.parquet')
WHERE domain IN ('example.com', 'bbc.co.uk');

Or in Python:

import pandas as pd
frame = pd.read_parquet("2025-12.parquet", columns=["domain", "rating"])

Because rows are sorted by domain, DuckDB and Arrow can skip most of the file when you filter on domain, so a narrow query does not read 25 GB.

The full guide has working code for nine languages, a list of programs that open parquet without any code, and how to load a release into your own database.

Yes, one address to call. It lists every file your licence covers, each with a signed download link, so a scheduled job can pick up each new release without anyone clicking anything. It is not a per domain lookup API and is not charged per domain. The documentation is here.

Yes. The sample file on the data page is 10,000 domains picked at random from the current release, in a CSV file with the same columns as the paid files. It is a fair slice rather than our best rows, so what you see is what the whole file looks like. No account needed.

Inside your own organisation, almost anything: store it, query it, join it to your own data, build products on it, publish work based on it. There is no limit on how many people use it and no limit on where you run it.

You may also show single ratings to your own customers in your product or your reports, and you do not have to credit us.

What you may not do is pass the dataset on: selling, republishing or handing over the files, or a large enough part of them to stand in for the whole, to anyone outside your organisation. The full terms are here.

By bank transfer. Subscribe and we send you an invoice with our bank details on it, payable within 14 days. Your access starts the day the payment reaches us and runs for a full year from that day.

No. All fees are final once paid. That is why the free sample exists, and why the columns and file sizes are published in full before you buy: so you can check the data is what you need first.

We send a renewal invoice 30 days before your year ends and email you about it. Paying it adds another year with no gap. Ignoring it lets the licence run out on its end date, and the files you have already downloaded are yours to keep using under the licence terms.

Your email address, when you first and last signed in, and, if you licence the data, your company name and billing address so the invoice is right. Free checker counts are held against your account or IP address and expire within 24 hours. Each lookup also leaves a short record of when it happened, whether it came from the checker or the extension, how many domains it covered, and the IP address, country and browser it came from. That record never includes the domains themselves and is not kept against your account. That is the whole list.

Email [email protected] and we delete the account and everything attached to it. Paid invoices stay in our invoicing provider's records for as long as tax law requires.