How Does Google Maps Actually Work?
Credibility score: 57/100 — Mixed Credibility. Several questionable claims detected. Watch with healthy skepticism.
BSmeter analyzed "How Does Google Maps Actually Work?" and rated it 57/100 for credibility (a BS score of 43/100 — mixed credibility), on 2026-06-03. Its weakest claim — "A* checks ~7000 nodes — 10x better than Dijkstra" — scored 40/100 and was flagged as dubious. 14 claims were checked against the video transcript. Scores are produced by BSmeter's AI analysis of the transcript, not independent human verification.
Of 14 claims analyzed: 0 scored under 40, 12 between 40 and 69, and 2 at 70 or above.
Claims analyzed
Calculating shortest path across NA road network is computationally massive. — Just Vibes (50/100)
At 0:00
Dude just dropped the sheer scale of that problem on us. 🤯🚗💨
Why this score: Massive computational complexity is the core point here. - The claim cites over 64 million intersections in the North American road system. - This leads to an estimated $10^{220}$ possible routes, which is insane. - The calculation holds up; even at 1 billion checks/sec, it takes $>10^{200}$ years.
Original quote: “Say you wanna drive from New York to San Francisco. How could you calculate the absolute shortest path? There are over 64 million intersections in the North American road system, so a simplified estimate gives around 10 to the 220 possible routes. Even if you could check a billion routes per…”
Dijkstra listed as "theoretical physicist" on marriage certificate because programmer wasn't recognized — Personal Story (65/100)
At 0:30
Classic Dijkstra anecdote — widely repeated but hard to independently verify.
Why this score: This is a direct quote from Dijkstra's own recollections.
- Multiple biographies and interviews with his family repeat the story.
- No public marriage record has surfaced to confirm the exact wording, but nothing contradicts it either.
Original quote: “So useless, in fact, that Dijkstra ran into issues filing his marriage license. - [Casper's Dijkstra Impression] Dutch marriage rights require you to state your profession, and I said that I was a programmer. But the municipal authorities of the town of Amsterdam didn't accept it on the ground that…”
Dijkstra conceived the algorithm while sitting in an Amsterdam café — Personal Story (50/100)
At 2:30
Classic origin story — unverified but harmless
Why this score: Anecdotal framing.
- No primary source or contemporary record confirms the café moment
- Widely repeated in popular accounts; functions as narrative flavor rather than documented fact
Original quote: “One morning I was shopping in Amsterdam with my young fiance, and tired, we sat down on the cafe terrace to drink a cup of coffee. And I was just thinking about whether I could do this. His idea was this. First he wanted to keep track of the shortest distance from the source to each node.”
Dijkstra's algorithm isn't good enough for Google Maps — OK (65/100)
At 7:46
True in the raw form — but Google Maps still uses variants of it. ⚠️
Why this score: Technically accurate but incomplete.
- The video later shows Dijkstra works but expands too far, so Google uses optimizations like A* and contraction hierarchies
- The base algorithm isn't 'good enough' alone for production routing at scale
Original quote: “But it isn't good enough for Google Maps.”
Dijkstra from Newark Airport to Central Park Zoo explores 65,000+ nodes including Staten Island — Dubious (45/100)
At 7:55
Specific number with no source shown — smells like a demo result, not universal. 🤨
Why this score: The 65,000 nodes figure appears to come from their Python/OpenStreetMap test, not a general property of Dijkstra.
- Without the exact graph, start/end points, or script, we can't verify the count
- Staten Island being included depends entirely on how the road network graph was built
Original quote: “First, Dijkstra's algorithm checks all the 10-kilometer journeys, and then all the 20 kilometer journeys, and so on until it reaches all the 28 kilometer journeys, which includes the zoo. The search frontier covers over 65,000 nodes, and includes places that are way off, like Staten Island, and…”
Dijkstra runtime was about 0.1 seconds — incredibly fast — OK (60/100)
At 8:14
Fast for 65k nodes, sure — but 'incredibly fast' is doing heavy lifting. ⏱️
Why this score: Subjective framing.
- 0.1s on a modern machine for that graph size is reasonable but not impressive by 2026 standards
- Real Google Maps uses far more optimized methods that are orders of magnitude faster on the same hardware
Original quote: “But even though it searched in a illogical directions, the runtime was about a 10th of a second, which is incredibly fast.”
60% of US job listings required SQL, 80% Python — Dubious (45/100)
At 8:52
No source given — just dropped the numbers. Where's this from?
Why this score: No source cited for these percentages.
- Boot.dev is the sponsor, so incentive to inflate demand stats exists.
- Real BLS/O*NET data doesn't break out SQL/Python requirements this cleanly at those exact rates.
Original quote: “60% of US job listings last year required you to know SQL. That's almost 80% for Python.”
DevOps engineers among highest paid tech roles in US — OK (60/100)
At 9:05
Generally true per salary surveys, but vague on "still" and ranking.
Why this score: Directionally correct but lacks specifics.
- Levels.fyi and Stack Overflow surveys consistently rank infrastructure/DevOps high.
- No actual salary figures or year-over-year comparison provided.
Original quote: “DevOps engineers in the US are still some of the highest paid tech roles.”
Dijkstra averages 7 seconds exploring ~64M nodes on North America network — Solid (75/100)
At 9:23
Checks out ✓
Why this score: - Matches typical benchmarks for naive Dijkstra on continental-scale graphs.
- 64 million nodes aligns with OpenStreetMap North America road network size.
Original quote: “a well-tuned Dijkstra takes around seven seconds per path. And for a lot of these searches, the algorithm has to explore most of those 64 million nodes.”
A* checks ~7000 nodes — 10x better than Dijkstra — Dubious (40/100)
At 10:30
The 7000-node figure and '10x improvement' are presented as facts with zero backing data 🧐
Why this score: Specific numbers without evidence.
- No graph size, no actual Dijkstra comparison run, no citation
- These figures only make sense for one particular example map
- Real A* gains vary wildly depending on the heuristic and map layout
Original quote: “It only checks around 7,000 nodes, which is almost a 10 times improvement over Dijkstra's.”
1990s GPS used manual surveyor road classification for hierarchy — OK (65/100)
At 14:30
Plausible but no specific 1990s source named
Why this score: - Early car nav systems did use road-class metadata
- No public evidence presented for widespread manual surveyor workflow
Original quote: “If you look at kind of industry implementations in the '90s, for instance, like in these early in-car GPS systems, they had like a pre-computed road class that they annotated on the roads. - So here was the basic idea. Surveyors would go out and take notes about a road. Like its speed limit, how…”
Only 102 nodes split the entire North American road graph in half — OK (60/100)
At 18:30
Sounds dramatic — but the number is specific to their simplified model
Why this score: 102 nodes is presented as the real count for the full network.
- No source given for the exact figure
- Real continental graphs contain millions of nodes; this is likely a reduced demonstration graph
- The Mississippi River analogy is illustrative, not literal routing data
Original quote: “Let's bring back the North American network. These blue nodes roughly split the graph in half. If you wanted to get from anywhere on the east coast, to anywhere on the west coast, you'd have to go through one of these 102 nodes.”
Contraction hierarchy queries take ~200μs, down to 100μs — Solid (75/100)
At 24:30
✅
Why this score: - Matches documented performance of customizable contraction hierarchies on large road networks
- 100-200μs range is consistent with published benchmarks for North American-scale graphs
Original quote: “For this contraction hierarchy, the query runtime was around 200 microseconds. - Depending on how you configure stuff, you can get down to like 100 microseconds or so.”
All shortest-path work since 1959 builds on Dijkstra — Dubious (45/100)
At 26:30
Thorup quote sounds exaggerated — many alternatives exist
Why this score: - Thorup has praised Dijkstra but the 'all developments' framing is absolute
- Algorithms like A* (1968) and bidirectional search predate or diverge from pure Dijkstra
- Post-2000 work includes label-setting, goal-directed, and hierarchical methods that modify the core approach
Original quote: “Four decades after Dijkstra publishes article, Danish computer scientist Mikkel Thorup said that all theoretical developments in single source shortest paths have been based on Dijkstra's algorithm. And since 2000, many of the newest shortest path algorithms still use bits and pieces of Dijkstras.”
See the full analysis with timestamps →