Wise Built the Internet's Most Boring Page

The currency converter is the most copied programmatic template on the internet, and almost every copy earns nothing.

Wise's version is credited with 37 million monthly visits — a Semrush figure published by Anatolii Ulitovskyi in November 2025, which we have not re-verified. What we could verify is the machine underneath it, and it is not the converter. On 16 August 2026 we counted Wise's sitemaps and found 14 separate programmatic page factories holding roughly 13 million URLs.

Then we opened the biggest one and found that 77% of it is the same page with a different query string.

That gap — between the URL count and the page count — is the most useful thing here, because it is the mistake you will make first. Below: what produces the result, what transfers, what does not, and the census you can run on a competitor this afternoon.

Results at a glance

One disclosure before the findings. Anatolii Ulitovskyi fact-checks this blog, and his article on Wise cites unmiss.com as one of its sources. His figures below are labelled as his and dated. Everything else we measured ourselves on 16 August 2026 with free tools, and each measurement is in a screenshot you can open at full size.

The product is the answer

Wise moves money between currencies. Someone typing "usd to eur" is, more often than not, minutes away from needing exactly that.

This is the structural advantage and it is worth naming before anything else. RTINGS had to build a laboratory to earn the right to publish numbers nobody else had. Wise already had the numbers, because running the exchange is the business. The rate feed is not content marketing — it is the product, rendered as a page.

That is why the converter template works for them and fails for the affiliate sites copying it. The template is identical. What sits behind it is not.

One rate feed, many factories

The central mechanic is not the converter. It is that a single live data source can be projected into a dozen unrelated page types.

An exchange rate feed and a bank-identifier dataset will generate: converter pages, SWIFT code lookups, IBAN examples, routing numbers, cost-of-living comparisons, provider comparisons, currency account pages. Each factory answers a different question. All of them read from the same two sources.

So adding a currency does not add a page. It adds a row to every factory at once. That is the compounding, and it is why the total reaches eight figures.

Your version: before you design a template, list every question your existing data can answer, not just the obvious one. The obvious one is the one your competitors already built.

Two factories carry nearly everything

Wise's sitemap index lists 27 entries. Stripping the blog and twelve help-centre locale feeds leaves 14 programmatic systems. They are wildly uneven.

The long tail of that list is the real lesson. airport-in-city is 176 URLs. exchange-rate-alerts is 19. Those are not failures — they are what it looks like when adding a new page type costs almost nothing, so you try things that might not work.

Your version: judge a programmatic system by how cheap its fifteenth page type is, not how big its first one got. If launching a new template needs a quarter of engineering time, you have built a page, not a factory.

Most of it is one page repeated

This is where a URL count stops being a page count. We opened the first shard of the currency-converter sitemap and counted what was in it.

Here is one of those pages, so you can see what the amount parameter actually changes.

Whether that is smart is genuinely arguable. Those URLs match real queries — people do search "1000 usd to eur" — and the page answers with the number already filled in. But they are near-duplicates competing with each other, and Google has been consolidating parameter variants for years.

Your version: count your programmatic layer as distinct templates × distinct data rows, and treat parameters as a separate line item. If your 500,000-page plan is really 40,000 pages and a query string, you want to know that before you build it, not after.

Write meta to fit the data

We sampled 20 pages across five factories and measured every title and description. Titles averaged 59 characters and descriptions 155 — both sitting close to what Google displays. But the averages hide the interesting part.

The longest of both came from the IBAN bank pages: a 104-character title and a 309-character description, on pages like /iban/united-arab-emirates/habib-bank. The shortest title was 31. Nobody chose those lengths. The template interpolates a bank name and a country name, and some banks have long names.

That matters because of what our own previous case study found. RTINGS averaged 259-character descriptions, with 19 of 20 over the display limit — deliberately oversized. Wise averages 155, with only 8 of 20 over. Two very large, very successful programmatic sites, opposite strategies.

Your version: "write descriptions long" is not a rule, it is one site's choice. What both sites actually do is let the template render whatever the data produces and not hand-tune 13 million tags. Decide the pattern once, then check the extremes — the longest and shortest your data can generate — because those are the ones that will look broken.

At this scale, architecture beats copy

We pointed our free Website Audit at wise.com, and it scores lower than you would expect.

79/100, with one critical failure, five warnings and seven notices. For comparison, RTINGS scored 90/100 with zero critical issues on a site roughly one four-hundredth the size.

The critical issue is the one worth staring at. Wise's converter sitemap is split into 170 shards by language. Locale is the organising principle of the entire system — and the reciprocal hreflang links, the ones that tell Google these pages are versions of each other, are not all coming back. That is the specific failure mode of locale-scaled programmatic SEO, found on the site that does it best.

The rest reads like scale, too: outbound links to dead pages, images missing alt text, page titles too long — which our own 20-page sample independently confirmed at a maximum of 104 characters.

Your version: the failures that matter at scale are relational, not textual. A missing alt tag is one page. A broken hreflang return link is a whole locale disconnected from its siblings. Audit for the second kind before you multiply.

What you can copy this week

• Project one dataset into many page types instead of one page type into many rows.
• Make new templates cheap. The 19-URL factory is the evidence that they are.
• Count honestly: distinct templates × distinct rows, with parameters listed separately.
• Let the template set meta length, then inspect the longest and shortest cases your data can produce.
• Audit for relational failures — hreflang reciprocity, internal links, canonicals — before adding a zero to your page count.

What you cannot copy

You probably do not own the data. Wise publishes exchange rates because it sets them. If you are rendering someone else's API into pages, every competitor can render the same API, and the page is not defensible.

The product is the answer. Someone searching "usd to eur" wants a conversion, and Wise sells conversions. That alignment cannot be manufactured by a content strategy. It is the business model doing the work.

Regulatory surface is a moat and a cost. SWIFT codes, IBANs and routing numbers are publishable because Wise operates inside that system. It is also why those pages must be right.

What we could not measure, stated plainly. We have no traffic, revenue or conversion data of our own for Wise. The 37 million is one third-party tool's estimate, on one date, reported by someone else, and we did not re-verify it. Our own figures are URL counts, meta lengths and a site audit — none of which is traffic. Sitemap URLs are not indexed pages, indexed pages are not visits, and visits are not revenue. Most of the factory totals in this article are estimates from a first-and-last-shard sample rather than full counts, and they are marked with a tilde wherever that is true.

Two things we deliberately did not test. We did not examine Wise's backlink profile, so we can neither support nor challenge the claim that its links were earned through PR rather than campaigns — treat that as his assertion, not a finding. And we did not measure where Wise ranks for converter terms, so nothing here should be read as a ranking result. This study is a structural census, not a performance report.

Run this census yourself

1. Fetch /robots.txt and follow every Sitemap: line. Wise's index points at 27 separate feeds. The ones you have never heard of are the interesting ones.
2. Count occurrences, not lines. Sitemap files are frequently a single line, so grep -c returns 1 and you will conclude a 170-shard index has one shard. Use grep -o '<loc>' | wc -l. This caught us out twice.
3. Sample the first and last shard. A first shard of exactly 50,000 means the generator fills to the cap in order, so the total is (shards − 1) × 50,000 plus the last one. Say "estimated" when you do this.
4. Break one shard down by URL shape. Query strings, sub-pages, and distinct entities. This is where an eight-figure number turns into a four-figure one.
5. Sample 20 pages for meta length across different factories, and look for the extremes rather than the average.
6. Audit one page and read which group loses points, not the headline score.

Frequently asked questions

The one thing to take away

The 13 million URLs are not the achievement, and treating them as one is how you get the lesson backwards. Two factories hold 94.5% of that number, and three quarters of the largest is a query string. The honest inventory is much smaller and much more interesting: a live dataset that a dozen different page types can read from.

So the first question about your own programmatic build is not "how many pages can we generate". It is "what do we measure, own, or set that nobody else can render" — and then, how many different questions does that one thing answer.

Start where we started — audit one page free — and you will know in ninety seconds whether your architecture can carry what you are planning to put on it.

Measured on 16 August 2026 against wise.com by direct sitemap counting and a 20-page meta sample, plus free-tier runs of our own tools. Factory totals marked with a tilde are estimated from a first-and-last-shard sample, not counted in full; sitemap URLs are not indexed pages and not traffic. Figures credited to Anatolii Ulitovskyi come from his November 2025 LinkedIn analysis, which cites Semrush and unmiss.com, and were not independently re-verified. He also fact-checks this blog. No traffic or revenue data for Wise was available to us. Third-party metrics change; re-run anything here before you rely on it.