How the six related cities are chosen
At the bottom of every one of our 500 city pages there is a block headed “if this city is not the one, look at these six”. This page is the method behind those six, including the measurements that show it working and the one measurement that got worse on purpose.
It is worth stating why this needed a method at all. The obvious approach — show the six most similar cities — produces six versions of the same place, and it quietly buries everything unpopular. We measured both problems on our own site before building anything.
Six cities, chosen to be close to the one you are reading about on cost, connectivity, safety, climate and latitude — and deliberately unlike each other, so you get six real alternatives instead of six near-duplicates.
Six different countries every time. On average the six span 3.9 of the six world regions we cover. Every one of our 500 cities appears in at least one list, so nothing is unreachable. No city pays to be there.
The problem, measured first
Before building anything we counted the internal links across the whole site to see which cities were actually reachable by a reader browsing normally. We excluded the exhaustive machine-generated grids — the full city index, every comparison pair and every clock page — because those link to all 500 cities by construction and would have produced a flattering, meaningless 100%. What is left is editorial linking: the links a human put there because the city was worth mentioning.
| Measure | Before | What it meant |
|---|---|---|
| Cities with at least one editorial link | 388 of 500 (77.6%) | 112 cities were never mentioned anywhere except the exhaustive grids. |
| Links from one city page to another | 0 | City pages were dead ends. There was no way to move sideways at all. |
| Mean editorial links per city | 1.78 | And a median of 1 — most cities had exactly one mention or none. |
| Most-linked city | 10 links | Against a median of 1. Attention was extremely concentrated. |
| Share of all mentions taken by the top 50 cities | 29.8% | A tenth of the catalogue absorbed nearly a third of the attention. |
| Gini coefficient of link distribution | 0.478 | Textbook popularity bias: the cities everyone already knows got linked, and the rest were invisible. |
Measured across all pages in the export on 1 August 2026. These are our own numbers about our own site, and they were not good.
What the list optimises for
Four things at once, which is the whole difficulty. Optimise for the first alone and you get duplicates; optimise for the rest alone and you get irrelevant noise.
- Relevance. How close a candidate is to the city you are reading about, across nine attributes: cost (on a log scale, because the difference between $600 and $900 matters far more than between $3,600 and $3,900), remote-work score, internet, safety, walkability, air quality, nightlife, climate and absolute latitude. Cost and internet carry the most weight; air quality the least.
- Diversity. Each pick is penalised by how similar it is to the picks already made — the standard maximal-marginal-relevance trick. This is what stops the block being six variations on one theme.
- Novelty. Each city carries a bonus of −log₂ of its share of existing editorial links, so the cities nobody was linking to are cheaper to include than the famous ones.
- Coverage. A running usage penalty makes a city less attractive the more lists it is already in, and a final repair pass places any city still at zero into the list where it fits best.
Two hard constraints sit on top of all four. No two cities in a list may share a country, so the six are always six different countries — this prevents the common failure where “alternatives to Lisbon” turns out to be four other Portuguese cities. One of those six may be the anchor's own country, because the nearest domestic alternative is often the most useful single entry in the list, but only ever one. And the anchor city itself can never appear in its own list.
What it measured out at
We built the naive version too — plain most-similar-six, no diversity, no novelty, no coverage term — and measured both against the same metrics, so the comparison is real rather than asserted.
| Metric | Most-similar-6 | What we ship | Change |
|---|---|---|---|
| Catalogue coverage | 493 of 500 (98.6%) | 500 of 500 (100.0%) | every city reachable |
| Intra-list diversity | 0.096 | 0.158 | +65% |
| Mean novelty | 9.30 | 9.48 | higher is rarer |
| World regions per list | 1.92 | 3.90 | roughly double |
| Most-used city appears in | 16 lists | 11 lists | less concentrated |
| Least-used city appears in | 0 lists | 1 lists | no orphans |
| Lists containing two cities in one country | many | 0 | hard constraint |
Intra-list diversity is mean pairwise dissimilarity within a list; novelty is the mean of −log₂ P(city) over each list. Both computed over all 500 lists.
An example
Here is the list that comes out for Bled, a small Slovenian lake town, to show the shape of the thing rather than to argue it is perfect:
| City | Country | Region | Cost | Score |
|---|---|---|---|---|
| Maribor | Slovenia | Europe | $1,600 | 3.7 |
| Innsbruck | Austria | Europe | $2,400 | 3.8 |
| Flic en Flac | Mauritius | Africa | $1,800 | 3.6 |
| Ras Al Khaimah | UAE | Middle East | $2,000 | 3.5 |
| Nicosia | Cyprus | Europe | $2,000 | 3.7 |
| Bruges | Belgium | Europe | $2,200 | 3.6 |
Six countries, several regions, a spread of costs around the anchor. Two of the six are the obvious nearby answers; the others are places a reader comparing lake towns would almost certainly never have typed into a search box.
What the block does not know
It only knows the attributes listed on the catalogue method page, which means it does not know your passport, your budget, your clients' time zone, whether you need a hospital nearby, or that you have already been to three of the six. It is a starting point for looking sideways, not a recommendation about your life.
It also has no memory. It does not know who you are, it does not track what you clicked, and the six cities under a given city page are the same six for everybody. That is a deliberate choice: the lists are fixed, published, and identical for every reader, which is what makes it possible to write this page and be held to it.
Money
Nobody pays to be in this list. No city, tourist board, coworking space, visa consultancy, relocation agency or property developer has paid for inclusion, for a position, for a score, or to be excluded. We have never been offered money for any of those things and we would refuse it, because a ranking that can be bought is worth nothing to the person reading it. Where we earn anything at all it is disclosed on the affiliate disclosure page, and no affiliate relationship touches any number on this site.
In this specific case that means the six cities under any page are produced by the procedure above and nothing else. There is no paid placement slot, no promoted entry, and no way to buy an appearance in another city's list.
Common follow-up questions
How are the six related cities on each city page chosen?
By balancing four things: similarity across nine attributes including cost, internet, safety, climate and latitude; deliberate dissimilarity between the six picks; a novelty bonus for cities that were rarely linked before; and a coverage term that ensures every one of our cities appears in at least one list. No two cities in a list may share a country.
Why is the most similar city sometimes not in the list?
Because six near-identical cities are less useful than six genuinely different ones. Average similarity is slightly lower in the lists we ship, and that is the deliberate trade. The comparison pages give the raw nearest matches without diversification.
Do all cities appear in at least one related list?
Yes — all 500. Before the module existed, 112 of them had no editorial link anywhere on the site outside the exhaustive index grids.
Are the related cities personalised to me?
No. There is no tracking and no personalisation. The six cities under a given city page are identical for every reader, which is what makes the method publishable and checkable.
Can a city pay to appear in these lists?
No. There is no paid placement, no promoted slot, and no commercial route into the selection at all.