Should you track local rankings separately on mobile and desktop? It is an expensive question to get wrong. If the two surfaces really rank differently, every grid, every report and every competitor set has to be built twice.
TL;DR
- We measured 703 locations across 40 keywords and 25 US metros. Mobile and desktop returned the same businesses in the same order on 99.85% of ranked pairs.
- The number that makes that meaningful: desktop compared against a second desktop reading of the same location scored 99.88%. Mobile sits inside the noise.
- The real difference is depth. Desktop returns 20 results every time. Mobile stops at 10, and comes up short of 10 on about one location in twenty.
- Nothing diverges by vertical or by market size. Every keyword category and every metro tier lands inside the control’s own margin.
- So you do not need a separate mobile dataset. And any tool showing a mobile local position past 10 is inferring it, not measuring it.
What we measured, and why the control is the whole story
Every location was fetched on mobile and on desktop, from the same place, within the same minute. A share of locations were then fetched a third time on desktop again.
That third fetch is the part that makes the rest meaningful. Google’s local results are not deterministic: ask the same question twice from the same place and you can get slightly different answers. So “mobile agreed with desktop 99.85% of the time” means nothing on its own. The only useful question is whether that is better or worse than desktop agreeing with itself. Without a control, a study like this is just a number.
One technical note, because it trips people up: mobile and desktop local results are not the same page with a device flag. They are two different Google surfaces, reached different ways, and a request built for one returns nothing on the other. This compares two surfaces, not one surface twice.
Finding 1: Inside the results they share, the order is the same
Scored on the businesses both surfaces returned, mobile placed them in desktop’s order on 676 of 703 locations. Across all 31,124 ranked pairs, 46 were inverted. That is 99.85% concordant.
The control, on 752 locations, came in at 99.88%. Three hundredths of a percentage point apart.
Per-location rank agreement, against the noise floor
Share of locations at each exact-agreement score · equal-depth locations only
Desktop vs desktop, the control n = 752
coffee shop in Philadelphia, 0% exact agreement, five inversions) scored exactly as badly on the desktop control. The most scrambled result anywhere is a control result: veterinarian in Nashville, at twelve inversions, more than double the worst mobile ever produced. Desktop reproduced itself perfectly on 97.1% of locations; mobile reproduced desktop perfectly on 95.1%.Put plainly: the difference between mobile and desktop is smaller than the difference between desktop and a second look at desktop. On this evidence there is no separate mobile ranking to track. There is one ranking.
Finding 2: The real difference is depth (20 versus 10)
Desktop returned exactly twenty organic local results on all 703 locations — in New York and Los Angeles, and equally in Fort Wayne, Bakersfield, Des Moines and Spokane. Twenty is not a description of how many businesses exist in a market. It is simply what the surface returns.
Mobile behaves differently in kind. It returned ten on 668 locations and fewer than ten on 35: nine on 29, eight on 6. Ten is a ceiling, and about one location in twenty falls short of it.
Organic local results returned per location
703 locations · 40 keywords · 25 US metros
The short lists are not thin supply. They are spread evenly across market sizes (9 in the largest metros, 13 in mid-size, 13 in the smallest), and desktop found twenty businesses for those same queries in those same places. Mobile simply stops sooner, and what decides where is not visible from the outside.
For rank tracking this is the entire practical difference between the two surfaces. Mobile can answer “am I in the top ten?” It cannot fill a 20-deep grid at all.
No divergence by vertical, and none by market size
The study was deliberately built wide enough to break the finding if it was going to break: 40 keywords across seven ranking categories, 25 metros in three size tiers across all four census regions, every keyword-metro pairing distinct.
| Ranking category | Locations | Inversions | Concordance |
|---|---|---|---|
| Food & retail | 86 | 12 | 99.69% |
| Beauty & fitness | 75 | 10 | 99.70% |
| Professional | 86 | 8 | 99.79% |
| Automotive | 72 | 6 | 99.81% |
| Contractor | 79 | 5 | 99.86% |
| Service-area trade | 217 | 5 | 99.95% |
| Health | 88 | 0 | 100% |
| Desktop control | 752 | 40 | 99.88% |
| Market tier | Locations | Concordance | Desktop depth | Mobile short lists |
|---|---|---|---|---|
| Tier 1 — New York, LA, Chicago… | 221 | 99.82% | 20 always | 9 |
| Tier 2 — Seattle, Denver, Boston… | 255 | 99.86% | 20 always | 13 |
| Tier 3 — Tulsa, Boise, Fort Wayne… | 227 | 99.88% | 20 always | 13 |
The divergence that turned out to be a measurement error
This is the part to read if you ever run a study like this yourself.
Early on, one result stood out. Four locations on ac repair in Seattle scored 33.3%, 55.6%, 55.6% and 66.7% exact agreement, while the desktop control on those same locations scored 100% every time. A failing arm beside a perfect control is normally the strongest signal a study can produce, and it very nearly became the headline.
Those four locations were also the only four in the entire sweep where mobile returned nine results instead of ten. That was not a coincidence. It was the whole explanation.
One Seattle location, scored 55.6% exact
The same ranking, one listing shorter
- 1One Hour Heating & Air Conditioning
- 2Owens & Sons Plumbing & HVAC
- 3Seattle AC And Heating Contractors
- 4Smart Residential Services Inc.
- 5HotWire Furnace, Air Conditioning & Appliance
- 6Seattle Heating & Air Conditioning
- 7Fuse Service
- 8Leschi Top HVAC Specialist
- 9Seattle Heating And Air Conditioning Repair
- 10Mark’s Mini Splits
- 1One Hour Heating & Air Conditioning
- 2Owens & Sons Plumbing & HVAC
- 3Seattle AC And Heating Contractors
- 4Smart Residential Services Inc.
- 5HotWire Furnace, Air Conditioning & Appliance
- 6Fuse Service
- 7Leschi Top HVAC Specialist
- 8Seattle Heating And Air Conditioning Repair
- 9Mark’s Mini Splits
We then re-ran those locations six times, keeping the full page each round. Mobile returned ten organic results on all six rounds, in the same order every round, matching desktop position-for-position. Nothing had been omitted. The listing our first pass recorded as missing, Seattle Heating & Air Conditioning, is in fact ranked first on both devices.
The wider claim built on those four locations failed too. There is no HVAC effect, and no vertical effect: ac repair contributes two inversions across five locations, the same as electrician across two and dentist across six.
Three measurement traps, if you run this yourself. All three produced the same shape of error: a false divergence that looked like a real finding.
1. Parsing before the page settles. Our first pass scrolled a fixed number of times and then read the list, which is a timing assumption rather than a settle condition. On four locations it read the list one row early. Wait for the result count to hold steady instead.
2. Comparing lists of different lengths slot by slot. When mobile is one result shorter, every slot below the omission holds a different business by construction, and a single missing listing reads as total disagreement. Compare the order of the results both surfaces returned.
3. Matching businesses on their display name. Nine of eighty-one desktop top-tens contained the same business name twice — two Planet Fitness in one Philadelphia list, two locations of The Joint in Atlanta. Chains break name-keyed comparison, so match on stable identifiers.
What this means for your rank tracking
- You do not need a separate mobile local dataset. It is one ranking, shown ten deep on one surface and twenty deep on the other.
- Any tool reporting a mobile local position of 11 or worse is reporting something Google did not return. Mobile stops at ten. A position of 14 on mobile is an inference, not a measurement.
- Use desktop for anything that needs depth: grids, share of voice, and competitor sets that run past the top ten.
- Use mobile only to answer “top ten or not”, and expect it to come up short of ten on roughly one location in twenty.
- Treat small position differences as noise unless you have a control. Desktop disagrees with itself often enough that a single-place move, on its own, means very little.
What this study does not show
- Why mobile stops short on about one location in twenty. Confirmed real by re-probing, and not explained by market size. This is the only genuine mobile-specific behaviour we found, and we cannot yet explain it.
- Whether mobile carries a different ad load. Our desktop and mobile ad detection worked differently, so reporting a gap between them as a device difference would repeat exactly the error described above. We are not claiming it.
- Anything outside the US. US metros only.
Method, in short
Concordance is measured over the businesses both surfaces returned: each mobile result is matched to the first unused desktop slot holding that business, and the resulting sequence is counted for inversions. Matching greedily to the first unused slot is what keeps duplicate chain names safe. A location is “clean” when its sequence has zero inversions.
Membership figures are reported on equal-depth locations only. A mobile list shorter than ten cannot match a ten-item desktop set, and including the short lists would measure depth while calling it disagreement.
One bug found after the run and corrected in every number above: results were originally keyed on business name plus review count, and the two surfaces format that count differently (1100 against 1.1K), so the same business could read as two. Everything here is rescored on a normalised match.
If you want to see where a business actually stands across a 20-deep local grid, and across the AI engines that increasingly answer these questions instead of Google, that is what Local Dominator is built to measure.
