Your trade decides, not your city
300 measurements, 16 markets: two trades in the same city share almost no sources, while one trade across two cities shares five times more. What it changes for your listings.
3%
shared sources between two trades in the same city
In this article
Short answer: the sites an AI reads to answer about you depend on your trade far more than on your city. We measured 300 trade-city combinations across 16 markets. Two trades on the same street share almost nothing - half the pairs have no source in common. The same trade in two different cities shares five times more. The consequence is direct: look at what your peers elsewhere are doing, not at the shop next door.
17%
between the same trade in two different cities
9 of 9
markets where the trade outweighs the city
In London, an architect, a dentist and a plumber share not one source. A dentist in London and a dentist in Manchester do.
What we measured
A cell is a triplet: a market, a trade, a city. For each one we ask three unbranded questions - « what are the best X in Y? » - in the market's language, and we record the addresses the AI assistant actually consulted to answer. Perplexity is the only one of the three AI assistants we query that publishes its sources: this measurement therefore rests on it alone, and we say so before the numbers rather than in a footnote.
On 5 September 2026 that gives 300 cells, 16 markets, 95 trades, 115 cities and 896 answers read, for 2,299 distinct domains cited. From each cell we keep the twelve most-cited domains: twelve is our ceiling, not a finding.
The question asked here was never asked in our earlier measurements. On 29 August we counted how many doors each city has. Today we ask what decides those doors: where you work, or what you do.
The business next door does not have your doors
Across 898 pairs of trades sharing a city, the average source overlap is 3%. In 442 of those pairs, nearly one in two, there is no shared source at all. London shows it at its sharpest: four trades, four watertight source worlds.
| Trade in London | Most-cited sources |
|---|---|
| Architect | londonbest.uk, houzz.com, architectsup.co.uk |
| Dentist | doctify.com, topdoctors.co.uk, doctory.uk |
| Plumber | checkatrade.com, mybuilder.com, ratedpeople.com |
| Veterinarian | aboutdogs.co.uk, allvets.co.uk, dnavetcare.co.uk |
This contradicts a widespread habit: asking a local shop owner which directory « works around here ». Their answer is true for them and worthless for you. The London plumber lives on tradesperson marketplaces; the dentist lives on medical directories. They meet nowhere.
The same trade elsewhere does look like you
Across 446 pairs made of the same trade in two different cities of the same market, average overlap rises to 17%, and only 7% of pairs share nothing. Five to six times better than geographic proximity.
| Trade and market | City A | City B | Shared sources out of 12 |
|---|---|---|---|
| Dentist, United States | San Antonio: reddit.com, doctor.webmd.com, healthgrades.com | Dallas: yellowpages.com, doctor.webmd.com, healthgrades.com | 5 |
| Lawyer, France | Rennes: starofservice.com, annuaire-avocats-france.fr, alexia.fr | Auvergne-Rhone-Alpes: annuaire-avocats-france.fr, alexia.fr, juritravail.com | 5 |
| Lawyer, Poland | Warsaw: oferteo.pl, tujestprawnik.pl, ratingcaptain.com | Krakow: oferteo.pl, tujestprawnik.pl, orlyprawa.pl | 3 |
Nine markets out of nine, without exception
We ran the calculation market by market, keeping only those with at least three cities and three trades. The trade beats the city everywhere, from the Gulf to Brazil, and the gap ranges from threefold to twenty-fivefold.
| Market | Cells | Same city, two trades | Same trade, two cities |
|---|---|---|---|
| Poland | 49 | 6% | 17% |
| Brazil | 42 | 3% | 21% |
| United States | 40 | 1% | 15% |
| United Kingdom | 31 | 2% | 15% |
| France | 30 | 4% | 20% |
| United Arab Emirates | 22 | 3% | 11% |
| Saudi Arabia | 16 | 1% | 25% |
| Switzerland | 14 | 5% | 14% |
| Belgium | 7 | 4% | 29% |
And no door is worldwide
If the trade decides, a given source should follow a trade from country to country. That holds, but weakly: of 2,299 domains cited, only 75 appear in more than one market, and none appears in all sixteen. Two cells drawn from two different markets share no source at all in 97% of cases.
The few domains that cross borders do so by trade, never by geography. Lawzana is cited in eleven markets, exclusively for lawyers. Booking.com in five, exclusively for hotels. Clutch.co in eight, for business services. The international door exists, but it is vertical.
| Domain | Markets | Cited for |
|---|---|---|
| lawzana.com | 11 | lawyers, only |
| clutch.co | 8 | business services |
| reddit.com | 7 | all trades |
| booking.com | 5 | hotels, only |
| bestlawyers.com | 5 | lawyers, only |
One last figure puts directories back in their place: 79% of the domains cited appear in a single cell, and only 4% of those domains are directories in the classic sense. What the AI assistant reads is overwhelmingly an editorial page, a local ranking, a forum - not a trade directory you sign up to.
What this measurement does not prove
- It rests on a single AI assistant, the only one that publishes its sources. We do not know whether ChatGPT and Gemini read the same sites, and we do not claim they do.
- The 95 trades and 115 cities come from our own campaigns, not from a random draw of the world economy. Seven of the sixteen markets have too few cities to be compared, and are left out of the per-market table.
- Being cited as a source does not mean being recommended. A source can be read and then ignored in the final answer.
- Twelve sources per cell is a ceiling we chose; a cell may well hold more.
What it changes for you
Look for your sources by trade, not by city.
Ask your trade's question in another city of the same country: the sites that come back are your real doors.
Stop copying the shop next door.
One source in two has nothing to do with yours, and half the time there is no shared source at all.
Think vertically when you go international.
A stronghold of your trade can follow you across ten countries; no general-purpose one does.
Do not buy a listing before seeing the source cited.
Nine in ten of the sites the AI assistant reads are not directories and have no signup form: you get in through a mention, an article, a ranking.
You can run this measurement on your own trade and city in a minute, free and with no account, and see the sources cited in your place.