Do AI assistants favour businesses near the city centre?
On 200 random AIGROW Index pages, we found the local businesses that ChatGPT, Claude and Gemini name on OpenStreetMap, measured how far each lies from the city centre, and compared them with every business of the same trade around the same city: 20 countries, 22 kinds of business, 88 cities.
Published by Guglielmo Paesano, founder of AIGROW · working paper, 5 November 2026
- Pages
- 200
- Countries
- 20
- Kinds of business
- 22
- Named / expected
- 1.40
- Central vs farther out, same page
- 2.41
Working paper, not peer reviewed
On 200 random AIGROW Index pages, a business within 2 km of the city centre was 2.41 times as likely to be named by ChatGPT, Claude or Gemini as one of the same trade around the same city farther out. Of the 863 businesses they named, 57.0% were central, against 40.7% expected from the map.
Findings
Central means within 2 km of the city centre. Figures are for the businesses named by at least one of the three assistants unless it says otherwise, and the tables below give each with its 95% interval.
A business within 2 km of the city centre was 2.41 times as likely to be named as one of the same trade around the same city farther out.
The 95% interval runs from 1.96 to 2.94. Each comparison is between businesses of the same trade around the same city, so trade and city are held fixed. Each assistant on its own shows the same pull: ChatGPT 2.54, Claude 2.66 and Gemini 2.71. Comparing only businesses alike in chain status, own website and number of sites, the ratio stays at 2.35 (1.90 to 2.96). 125 of the 200 pages have a named business found on the map; the others add nothing to either side.
The share named falls with every step away from the centre: 8.5% of the businesses under 1 km from it were named, against 2.6% of those 5 km or more out.
A business under 1 km from the centre was 5.05 times as likely to be named as one of the same trade and city 5 km or more out (3.26 to 8.25). Seen from the other side, 57.0% of the 863 businesses the assistants named lay within 2 km of the centre, against 40.7% if each page’s had been drawn from its map with no regard to distance: 1.40 times the expected share (1.30 to 1.50).
Without hotel and restaurant pages, the trades visitors from elsewhere ask about, a central business was still 1.98 times as likely to be named.
This check was added after the first results and was not part of the method fixed in advance. It keeps 182 pages and 7,960 businesses, with an interval from 1.53 to 2.51. By kind of business, 8 of the 22 have a same-page interval wholly above 1: architects, dentists, driving schools, gyms, hair salons, hotels, insurance brokers and restaurants. Real estate agencies, with 55 named businesses, show no clear pull: 0.83 (0.52 to 1.32). Each kind rests on 9 or 10 pages, so a single row shows a direction, not a size.
What it means for a business farther from the centre
When someone asks an assistant for the best of a trade in a city, the answer leans toward the centre, and a business on the edge of town starts behind, even when it is closer to the person asking. The study does not show that location causes the gap, so it gives no reason to move. What an owner farther out can do is describe the business clearly enough for an assistant to place it.
- Say where you are and where you work
- Name your neighbourhood and the areas you serve on your own website and on your Google Business Profile, so an assistant reading about you can place you in the city.
- Write for the districts you serve
- Give each district or nearby town you serve a page of its own that says what you do there and how customers reach you.
- Be the same business everywhere
- Use one name, address and phone number on your site, your Google Business Profile and the directories, so every mention an assistant reads points to the same place.
- Measure it
- Check who the assistants name for your trade in your city now, and again after any change. AI Local Fix does this groundwork for a local business, from listings and the Google profile to page markup and a review kit, then asks the same questions again and counts your closest rivals in the answers.
Distance and the answers
The first chart gives the share of the businesses on the map that at least one assistant named, by distance from the centre. The second sets the central share among the businesses each assistant named against the share the map predicts. Bars in one chart share a scale.
- Central share, named
- Central share expected
By assistant
Named / expected divides the central share among the named businesses by the expected share. The last two columns compare a central business with one farther out on the same page, the second also within the same chain status, website and number of sites.
| Assistant | Named businesses | Central share, named | Central share expected | Named / expected | Central vs farther out, same page | Adjusted for chain, website and branches |
|---|---|---|---|---|---|---|
| ChatGPT | 531 | 58.8% (54.5% to 62.9%) | 41.6% | 1.41 (1.31 to 1.53) | 2.54 (2.00 to 3.24) | 2.44 (1.88 to 3.21) |
| Claude | 509 | 59.5% (55.2% to 63.7%) | 41.6% | 1.43 (1.32 to 1.58) | 2.66 (2.03 to 3.62) | 2.60 (1.93 to 3.58) |
| Gemini | 350 | 60.6% (55.4% to 65.5%) | 42.1% | 1.44 (1.30 to 1.58) | 2.71 (2.03 to 3.65) | 2.52 (1.83 to 3.48) |
| Any of the three | 863 | 57.0% (53.7% to 60.3%) | 40.7% | 1.40 (1.30 to 1.50) | 2.41 (1.96 to 2.94) | 2.35 (1.90 to 2.96) |
Counting every linked name, so a business named in ten answers counts ten times, the central share is 60.6% against 42.2% expected: 1.44 times (1.31 to 1.58).
By distance from the centre
Each band is compared with businesses 5 km or more from the centre on the same page.
| Distance from the centre | Businesses on the map | Named by any assistant | Share named | Against the outer band, same page |
|---|---|---|---|---|
| 0 to 1 km | 3,354 | 286 | 8.5% (6.8% to 11.5%) | 5.05 (3.26 to 8.25) |
| 1 to 2 km | 3,567 | 206 | 5.8% (4.5% to 7.4%) | 2.84 (1.90 to 4.47) |
| 2 to 3 km | 2,770 | 112 | 4.0% (3.1% to 5.4%) | 2.36 (1.57 to 3.55) |
| 3 to 5 km | 3,645 | 131 | 3.6% (2.7% to 4.7%) | 1.56 (1.12 to 2.17) |
| 5 km to the edge | 4,862 | 128 | 2.6% (1.8% to 3.9%) | reference |
Other ways to draw the line
Every variant keeps the same-page ratio between 1.96 and 2.52, and every interval stays above 1.
| Variant | Businesses on the map | Named businesses | Central share, map | Central share, named | Central share expected | Named / expected | Central vs farther out, same page |
|---|---|---|---|---|---|---|---|
| Main rule: central means under 2 km | 18,198 | 863 | 38.0% | 57.0% | 40.7% | 1.40 (1.30 to 1.50) | 2.41 (1.96 to 2.94) |
| Central means under 1 km | 18,198 | 863 | 18.4% | 33.1% | 20.1% | 1.65 (1.46 to 1.87) | 2.42 (1.89 to 3.10) |
| Central means under 3 km | 18,198 | 863 | 53.3% | 70.0% | 54.2% | 1.29 (1.22 to 1.37) | 2.52 (1.95 to 3.22) |
| Inside 5 km only, under 2 km against 2 to 5 km | 13,336 | 735 | 51.9% | 66.9% | 54.3% | 1.23 (1.17 to 1.30) | 1.96 (1.64 to 2.35) |
| Single-site independents only | 16,530 | 632 | 37.6% | 57.1% | 40.8% | 1.40 (1.29 to 1.52) | 2.37 (1.90 to 2.96) |
| Without addresses in another town | 16,824 | 832 | 40.7% | 58.4% | 43.5% | 1.34 (1.25 to 1.44) | 2.25 (1.84 to 2.77) |
| Without hotel and restaurant pages, added after the first results | 7,960 | 451 | 27.9% | 44.3% | 32.0% | 1.39 (1.23 to 1.54) | 1.98 (1.53 to 2.51) |
By kind of business
Most named first. With 9 or 10 pages each, read a row as a direction; where a kind has few named businesses there is little to compare.
| Kind of business | Pages | Businesses on the map | Central share, map | Named businesses | Central share, named | Central share expected | Named / expected | Central vs farther out, same page |
|---|---|---|---|---|---|---|---|---|
| Restaurants | 9 | 8,276 | 45.4% | 223 | 71.3% (65.0% to 76.8%) | 46.8% | 1.52 (1.37 to 1.71) | 3.16 (2.28 to 4.63) |
| Hotels | 9 | 1,962 | 48.0% | 189 | 70.4% (63.5% to 76.4%) | 54.3% | 1.30 (1.15 to 1.49) | 2.69 (1.61 to 7.32) |
| Hair salons | 9 | 2,718 | 30.6% | 62 | 59.7% (47.3% to 71.0%) | 33.3% | 1.79 (1.40 to 2.26) | 3.46 (2.00 to 6.90) |
| Gyms | 9 | 584 | 29.3% | 56 | 48.2% (35.7% to 61.0%) | 31.6% | 1.52 (1.15 to 1.95) | 2.20 (1.32 to 3.12) |
| Real estate agencies | 9 | 328 | 48.2% | 55 | 43.6% (31.4% to 56.7%) | 47.3% | 0.92 (0.69 to 1.11) | 0.83 (0.52 to 1.32) |
| Dentists | 9 | 1,056 | 31.4% | 54 | 38.9% (27.0% to 52.2%) | 23.8% | 1.63 (1.22 to 1.88) | 2.25 (1.33 to 3.30) |
| Veterinarians | 10 | 240 | 11.7% | 40 | 12.5% (5.5% to 26.1%) | 11.3% | 1.10 (0.00 to 1.84) | 1.14 (0.00 to 2.58) |
| Car repair shops | 9 | 1,126 | 7.0% | 39 | 12.8% (5.6% to 26.7%) | 10.2% | 1.26 (0.59 to 2.38) | 1.32 (0.54 to 3.04) |
| Driving schools | 9 | 368 | 17.4% | 34 | 44.1% (28.9% to 60.5%) | 31.1% | 1.42 (1.05 to 1.79) | 2.18 (1.15 to 3.47) |
| Lawyers | 9 | 345 | 51.6% | 30 | 80.0% (62.7% to 90.5%) | 59.1% | 1.35 (0.69 to 1.92) | 2.88 (0.28 to 14.54) |
| Architects | 9 | 188 | 50.5% | 24 | 66.7% (46.7% to 82.0%) | 54.0% | 1.24 (1.02 to 1.79) | 1.79 (1.07 to 5.14) |
| Insurance brokers | 9 | 151 | 49.7% | 16 | 68.8% (44.4% to 85.8%) | 48.4% | 1.42 (1.01 to 1.64) | 2.36 (1.02 to 3.77) |
| Dermatologists | 10 | 87 | 35.6% | 8 | 50.0% (21.5% to 78.5%) | 26.3% | 1.90 (0.50 to 3.00) | 3.11 |
| Electricians | 9 | 170 | 15.9% | 7 | 28.6% (8.2% to 64.1%) | 19.2% | 1.49 (0.00 to 2.07) | 2.52 |
| Accountants | 9 | 80 | 31.2% | 7 | 14.3% (2.6% to 51.3%) | 19.8% | 0.72 (0.00 to 1.50) | 0.63 (0.00 to 2.00) |
| Physiotherapists | 9 | 252 | 26.6% | 6 | 66.7% (30.0% to 90.3%) | 28.7% | 2.32 (1.50 to 3.76) | 10.30 |
| Plumbers | 9 | 64 | 18.8% | 5 | 40.0% (11.8% to 76.9%) | 22.7% | 1.76 (0.00 to 3.33) | 2.20 (0.00 to 10.30) |
| Moving companies | 9 | 65 | 4.6% | 3 | 0.0% (0.0% to 56.1%) | 16.7% | 0.00 (0.00 to 0.00) | 0.00 (0.00 to 0.00) |
| Digital marketing agencies | 9 | 63 | 36.5% | 2 | 50.0% (9.5% to 90.5%) | 40.8% | 1.23 (0.00 to 2.00) | 1.58 |
| Financial advisors | 9 | 44 | 38.6% | 1 | 100.0% (20.7% to 100.0%) | 25.0% | 4.00 | n/a |
| Cleaning services | 9 | 16 | 18.8% | 1 | 0.0% (0.0% to 79.3%) | 16.7% | 0.00 | 0.00 |
| Roofers | 9 | 15 | 6.7% | 1 | 0.0% (0.0% to 79.3%) | 0.0% | n/a | n/a |
Big cities and small maps
The data holds no population figures, so city size is approached two ways: by the radius each city’s map was read to, and by the number of businesses on a page’s map, split at the median page of 15.
| Group | Pages | Businesses on the map | Central share, map | Named businesses | Central share, named | Central share expected | Named / expected | Central vs farther out, same page |
|---|---|---|---|---|---|---|---|---|
| Cities read to 10 km | 17 | 1,759 | 13.1% | 87 | 19.5% (12.6% to 29.1%) | 15.6% | 1.25 (0.92 to 2.91) | 1.33 (0.90 to 3.43) |
| Cities read to 8 km | 183 | 16,439 | 40.7% | 776 | 61.2% (57.7% to 64.6%) | 43.5% | 1.41 (1.31 to 1.51) | 2.53 (2.07 to 3.10) |
| Pages above the median map size | 97 | 17,678 | 38.2% | 782 | 58.7% (55.2% to 62.1%) | 41.4% | 1.42 (1.32 to 1.53) | 2.50 (2.02 to 3.10) |
| Pages at or below the median map size | 103 | 520 | 33.7% | 81 | 40.7% (30.7% to 51.6%) | 33.9% | 1.20 (0.99 to 1.45) | 1.57 (0.96 to 2.28) |
How the study was made
- The answers
- The November 2026 edition of the AIGROW AI Visibility Index asked ChatGPT, Claude and Gemini the same questions about one trade in one city, several times over. 7,198 answers on the pages studied came back. No assistant was asked anything new for this study.
- From memory
- All 7,221 answer records of these pages in the Index data are marked as answered without web search, and no assistant was told where the person asking was. The questions name the city only, never a district or a street.
- The pages
- 200 local Index pages drawn at random, 10 in each of 20 countries, covering 22 kinds of business in 88 cities.
- The centre
- Each city’s map was read around one point, the reference coordinates AIGROW Labs uses for that city: to 8 km on 183 pages, and to 10 km on 17 pages in Berlin, London, Los Angeles, Madrid, Mexico City, New York, Paris, Rome and São Paulo. Distance runs from that point to the nearest map object of the business.
- The map
- Every OpenStreetMap business of the page’s trade around its city: 18,198 businesses on the 200 pages, 6,921 of them within 2 km of the centre. A business is one distinctive name on one page; names made only of trade and city words are left out, since an answer cannot single them out.
- From a name to a business
- A name is linked when it fits exactly one business on the page’s map, with the same links as the chains study. Names that fit no business or several are left out of every count. A named business counts once, however often it is named.
- Central or farther out
- Central means under 2 km from the centre, and farther out means between 2 km and the edge of the map. A business is named when at least one answer of the assistant on that page names it. The method was fixed on 5 November 2026 at 11:18 CET, before any figure about named businesses was computed.
- The comparisons
- The same-page ratio is a Mantel-Haenszel risk ratio with one stratum per page, so each comparison is between businesses of the same trade around the same city. Named / expected divides the central share among the named businesses by the share expected if each page’s named businesses were drawn from its map with no regard to distance. The adjusted ratio adds chain status, own website and the number of map objects with the same name as strata.
- Statistics
- Shares of businesses carry 95% Wilson intervals. The share named in each distance band and every ratio carry a bootstrap of whole pages (2,000 resamples), which keeps the mix of pages in the uncertainty; when more than one resample in ten has no defined value, the interval is left out.
Limits
- A pattern, not its cause
- Central businesses may be older, better known, more reviewed or more written about, and any of these could be why the models name them. How well known a business is may explain the pattern as much as distance does, and this study cannot separate the two.
- Answers from memory
- The assistants had no web search and no location. An assistant that knows where the person is, or that searches the web, may answer differently.
- Big cities and small maps cannot tell
- In the cities read to 10 km the same-page ratio is 1.33 (0.90 to 3.43) on 17 pages, and on the 103 pages with 15 businesses or fewer on the map it is 1.57 (0.96 to 2.28). Both intervals reach 1, so for these pages this sample cannot tell either way. The pull is clear where a city and trade have many businesses on the map.
- The centre is a reference point
- It is the coordinate AIGROW Labs uses for the city, not a measured downtown. In a city with several centres, or one whose businesses cluster away from that point, distance from it means less, and that may be why the cities read to 10 km show a weaker result.
- A check added later
- The check without hotel and restaurant pages was added on 5 November 2026 at 11:20 CET, after the first results. It gives 1.98 (1.53 to 2.51), but it was not part of the method fixed in advance.
- Names not on the map
- Names that cannot be tied to one business on the map are left out of every count. If they were mostly far from the centre, the ratios here would be overstated; where they are was not measured.
- The map itself
- OpenStreetMap may describe central businesses more fully, and businesses beyond the radius are outside the comparison. 1,374 businesses whose address names another town or district were kept; without them the same-page ratio is 2.25 (1.84 to 2.77).
- No second sample
- 200 pages, each kind of business on 9 or 10 of them. The larger sample of the chains study cannot be reused, because its map data holds no coordinates, so there is no second sample to repeat the test on.
- One edition
- All answers come from the November 2026 edition of the Index.
- Competing interests
- The author founded AIGROW, which publishes the Index and sells services that aim to improve how businesses appear in AI answers. The study had no external funding.
Rankings from the AI Visibility Index
See which businesses and products ChatGPT, Claude, Gemini and Grok named for the categories and places below.
How to cite
Guglielmo Paesano, Centre or edge of town: distance from the city centre in AI answers, AIGROW Labs working paper, 5 November 2026, aigrow.gg/research/ai-city-centre-bias-2026.
A working paper, not yet peer reviewed. The figures on this page are the paper’s.
Answers: AIGROW AI Visibility Index, November 2026 edition, DOI 10.5281/zenodo.23934866.
Map data © OpenStreetMap contributors, ODbL.
The study names no business. The files that link names to businesses stay with AIGROW Labs.
Do assistants name businesses in the centre instead of you?
The free scan asks ChatGPT, Claude, Gemini and Grok your own customers' questions, shows who they name instead of you, and checks whether their crawlers can read your site.
AIGROW Labs · 5 November 2026