Type any address you like onto clinicaltrials.gov and the server says yes. A real trial record, a page that never existed, a string you invented on the spot — all of them come back as HTTP 200, and all of them come back as the same file, byte for byte.
That is not a fault. It is an Angular application, and the application decides after the fact whether what you asked for exists. But it does mean the usual way of poking at a site from outside tells you almost nothing here.
So we went to the sitemap instead, and took all of it: the root index, the nested index hiding inside it, the third index hiding inside that, and every leaf file at the bottom. Plain XML the whole way down, none of it sampled.
What came back is the most lopsided site we have measured in this series. ClinicalTrials.gov publishes 600,762 trial records. It publishes 69 other URLs. That, as the site's own sitemap tells it, is the entire shape of the thing.
Results at a glance
ClinicalTrials.gov is a register with a small building attached. For every page that explains, guides or searches, there are roughly 8,707 trial records. The whole non-record side — homepage, search, policy, API documentation, patient guidance — fits in one sitemap file of about 8.5 KB.
And that small file has gone stale. 3 of its 69 URLs no longer resolve to a real page, while the site's own site-map page links to a live page the sitemap does not contain. The vault is immaculate. The lobby has not been swept in a while.
A register with a small lobby
Most large sites in this series are built the other way round. Marketplaces grow listings and wrap them in category pages, filters, hubs and guides, so the navigational furniture scales alongside the inventory.
ClinicalTrials.gov does not scale its furniture at all. The records run into the hundreds of thousands; the rest of the site is a homepage, a search, a handful of policy pages, some documentation for people submitting studies, and a short shelf of patient-facing explainers.
You can see the whole of it in one picture, because the whole of it is one small file.
The map that points at ghosts
A sitemap that small should be the easiest file on the internet to keep correct. You could check it by hand over a cup of tea. It is not correct.
Because every path here returns the same shell, a plain fetch cannot tell a live page from a dead one. You have to render each address and see what the application draws. Do that across the small file and 3 of its 69 entries turn out to be empty rooms: a review page, a videos page, and a page about how to join a study.
The last one is the telling case. The sitemap ships that address with an underscore in it. The site's own site-map page links the same address with a hyphen, and the hyphenated version is a real, working page that does not appear in the sitemap at all. The site's JavaScript bundle knows only the hyphen. The sitemap kept a spelling the site itself had moved on from.
So the small file fails in both directions: it advertises pages that are gone and omits a page that is there. Nobody notices, because nobody audits 69 URLs when there are 600,762 next door.
What a trial page actually serves
The records themselves are the reason anyone comes. They are also, to a crawler making a plain request, invisible.
We sampled study pages from the oldest identifier in the corpus to the newest. All returned 200, all byte-identical to each other, all carrying the same generic title and an empty meta description. A deliberately invented identifier returned exactly the same bytes, so a 200 here does not prove a trial exists.
This is worth stating carefully. It does not mean the records are unindexable; search engines render JavaScript. It does mean the discoverability of 600,762 records rests on rendering going well every time, and that anyone auditing this site with a fetch-only tool would conclude, wrongly, that every page on it is identical.
Links nobody had to go earn
Run the domain through our Backlink Analyzer and the picture inverts. The site whose navigation fits in 8.5 KB has 121,556 referring domains.
Divide those domains by the pages and you get 0.2 referring domains per page, a middling number in this series. It sits below itch.io at 0.318, well below GOV.UK at 0.906 and far below MDN at 3.9, but comfortably above Eventbrite at 0.148, PyPI at 0.082 and Rotten Tomatoes at 0.063. Calculator.net, with 222 pages and 143 referring domains each, remains from another planet.
The interesting part is where those links come from. This is not a site that publishes link bait; the entire publishable surface is 69 pages, 3 of them dead. The links attach to the records. Every trial written about, cited or argued over acquires a permanent address here, and the citation follows. The link is a by-product of the record existing.
A perfect score, honestly read
Our Website Audit returned 100. No critical issues, no warnings, no notices, 123 checks passed. It is the first perfect score in this series — ahead of PyPI at 91, MDN at 88, itch.io at 88 and GOV.UK at 86.
Read that as a fact about how the site is built rather than as a report card. A single application shell, served fast, with nothing bolted on, has almost no surface on which to fail a technical check. The score is real. It is also a measure of restraint more than of effort.
Copy this in an afternoon
You almost certainly have a lopsided site too. Here is how to find out whether your small side is rotting.
1. Pull your sitemap all the way down. Follow every nested index to its leaves. This site's root index lists 2 children, which looks flat; one of them is another index pointing at a third index listing 121 leaf files. Read the root only and you miss the site.
2. Split the corpus into records and rooms. Count the machine-generated URLs and the hand-made ones separately. The ratio tells you which half of your site has an owner and which half runs on momentum.
3. Render every hand-made URL, do not just fetch it. On any single-page application a fetch returns 200 for everything, including addresses deleted years ago. Load each one and check that something real draws.
4. Diff the sitemap against your own navigation. List the URLs your header, footer and site-map page link to, and compare both ways. Pages in the sitemap but not the nav are usually dead. Pages in the nav but not the sitemap are usually invisible.
What you cannot copy
The records arrive. The unfair advantage here is that the content is submitted to the site rather than commissioned by it, and every submission comes with a permanent identifier attached. That is why 15 URLs of submission documentation can sustain 600,762 records, and why the referring domains accumulate without anything resembling a content strategy. If your inventory does not walk in on its own, none of the economics on this page transfer to you.
And here is what we could not check:
We never saw a server-rendered trial page. Every study URL we sampled returned an identical shell with no record-specific content, and an invented identifier returned the same bytes. We cannot tell you what a search engine's renderer ultimately sees, how much of the corpus is indexed, or how any of it performs. We measured structure, not outcomes.
Our own first pass got the composition wrong. The total was right and independently reconfirmed — 3 separate counting methods agree on 600,831 URLs with no duplicates — but the initial breakdown of the 69 non-trial pages covered only 55 of them and silently dropped a whole section, the patient-facing study-basics shelf. We publish the corrected breakdown above. We also dropped an earlier claim that those 69 URLs were the complete browsable site, because rendering them showed it was not true in either direction.
The dead-page test depends on rendering, not on status codes. Every path on this domain returns 200, so "dead" here means the application drew nothing recognisable when we loaded the address. That is a judgement call made from the outside, and the site's own team would have better evidence than we do.
Our backlink tool's numbers do not perfectly reconcile. The dofollow and nofollow figures it reports sum to about 5% more than the total backlinks it reports on the same screen. The gap is small enough that the shape of the finding holds, but it is our defect and you should treat the absolute totals as approximate.
We measured on one day. The lastmod dates in the sitemap are self-reported by the site, the current year is partial, and none of this tells you how quickly a newly registered trial appears in the index.
The pages nobody owns are the ones that quietly break. Run your domain through the UNmiss Website Audit and see what the crawler finds first.
- Technical, on-page, speed and mobile checks in one pass
- Issues ranked by severity, not by volume
- Works on any domain, no signup
Frequently asked questions
How many URLs does ClinicalTrials.gov actually publish?
600,831 in total: 600,762 individual trial records and 69 non-trial URLs. Nothing was estimated. All 122 leaf sitemap files were fetched, and the total was confirmed by an independent second count using 3 different methods, with zero duplicate URLs across the corpus.
Why not just sample the sitemap and multiply?
Because it would have produced a wrong answer. The 121 study shards are not uniform slices — they range from 4,434 to 5,119 URLs, with 54 distinct sizes, and only 2 of the 121 hold 5,000 or more. Estimating from the first, middle and last file would have been off by thousands.
What is the trial-to-page ratio, exactly?
600,762 divided by 69 is about 8,707 trial records for every non-trial URL. If you exclude the 3 sitemap entries that no longer render a real page, the ratio against the 66 that still render rises to roughly 9,102 to 1.
How fresh is the archive?
99,122 URLs, or 16.5%, carry a lastmod date in the current year, and 73,071 carry the previous year. Together that is 28.7% of the archive touched in roughly the last 20 months. The current year is partial, measured on 29 August 2026, so it is not comparable with a full prior year.
Can you tell how many trials exist from the NCT numbers?
No. The identifiers in the sitemap run from NCT00000102 to NCT07793721, a numeric range of about 7.79 million holding 600,762 live records. The identifier space is sparse, so the highest number is not a count of anything.
What did the backlink tool report?
A domain rank of 93, 121,556 referring domains, and roughly 17.2 million backlinks, with the dofollow and nofollow halves splitting about 54.7% dofollow. Those two halves sum to slightly more than the stated total, which is a reconciliation defect in our tool rather than in the data about the site.
How does the link density compare?
It is mid-table for this series. Healthline measured 14.4 referring domains per page, MDN 3.9, GOV.UK 0.906, GeeksforGeeks 0.6, itch.io 0.318, Eventbrite 0.148, PyPI 0.082, Meetup 0.069, Rotten Tomatoes 0.063 and Levels.fyi 0.026. Calculator.net, with just 222 pages, measured 143.
Is a perfect audit score unusual?
Yes. It is the first 100 in this series, with 0 critical issues, 0 warnings, 0 notices and 123 checks passed across on-page, technical, speed and mobile. It reflects a site with a single consistent shell and very little bolted on, rather than unusual remedial work.
The lesson here is not that ClinicalTrials.gov did something wrong. A register whose job is to hold records permanently is entitled to be almost entirely records. The proportions follow from the purpose.
The lesson is what happens to the other side. When one part of a site grows into the hundreds of thousands and the other stays at 69, the small part stops having a caretaker. It is too small to move any metric, and small enough that a broken address can sit in the sitemap indefinitely while the working version of the same page goes unlisted. Neglect at that scale is invisible from the inside.
Which is the argument for looking from the outside occasionally. Pull your own sitemap to the leaves, render the pages a human would actually browse, and check that the map still matches the building. The UNmiss Website Audit will do the crawling half of that for you.
Measured on 29 August 2026. URL counts come from a full fetch of the clinicaltrials.gov sitemap tree — the root index, both nested indexes and all 122 leaf files, roughly 63 MB of uncompressed XML — with no sampling and no failed requests. A second pass recounted the total using 3 separate parsing methods, all agreeing at 600,831.
Backlink and audit figures come from the UNmiss Backlink Analyzer and UNmiss Website Audit, run the same day against clinicaltrials.gov. Series comparisons use the same tools on earlier subjects. Page liveness was determined by rendering each non-trial URL, because every path on this domain returns HTTP 200 whether or not the page exists.