You were training AI while catching Pokemon
Credibility score: 84/100 — Highly Credible. This video is highly credible with well-supported claims.
BSmeter analyzed "You were training AI while catching Pokemon" and rated it 84/100 for credibility (a BS score of 16/100 — highly credible), on 2026-04-25. Its weakest claim — "Niantic spun off Niantic Spatial after selling Pokémon GO" — scored 45/100 and was flagged as dubious. 20 claims were checked against the video transcript. Scores are produced by BSmeter's AI analysis of the transcript, not independent human verification.
Of 20 claims analyzed: 0 scored under 40, 2 between 40 and 69, and 18 at 70 or above.
Claims analyzed
Pokémon GO had 500M installs in 60 days, 143M peak users — Solid (80/100)
At 0:00
Those Pokémon GO numbers are spot on — classic hype stats that hold up 📱🔥
Why this score: These are well-documented milestones from Pokémon GO's 2016 launch.
- Downloads: Niantic and Sensor Tower confirmed ~500M global downloads in first 2-3 months (often cited as 60 days).
- Peak users: ~143M monthly active users reported by Niantic around August 2016.
Slight variances in exact timing exist across sources, but the order of magnitude is accurate.
Original quote: “[0:00] 500 million people installed Pokémon GO [0:03] within 60 days of its launch. At its [0:05] [music] peak, 143 million people were [0:08] walking around parks, sidewalks, and [0:10] parking lots pointing their phones at [0:12] buildings and statues trying to catch [0:14] digital creatures…”
Niantic spun off Niantic Spatial after selling Pokémon GO — Dubious (45/100)
At 0:16
Niantic didn't sell Pokémon GO — they still own it. Spin-off maybe real, but this framing is off 🚩
Why this score: Core error: Niantic never sold Pokémon GO — they co-developed/own it with The Pokémon Company since 2016 and still do in 2026.
- Niantic Spatial: Real AI mapping spin-off announced ~2024-2025 for AR tech from Pokémon GO data.
- But the 'after selling' part is invented — no sale happened. This ties Pokémon GO success to a new venture misleadingly.
Original quote: “[0:16] real world. That was 2016. It's 2026 now [0:19] and Niantic just revealed what all of [0:21] that was actually worth. [0:24] Recently, Niantic Spatial, the AI [0:27] company Niantic spun off after selling [0:29] Pokémon GO, announced a partnership with”
Niantic spun off after selling Pokémon GO, partnered with Coco Robotics using VPS trained on 30B images — Verified (95/100)
At 0:30
Nailed it — Niantic did spin off into Niantic Spatial after the 2025 Pokémon GO sale, Coco partnership announced March 2026, and yes, 30B images power their VPS. Spot on. 📍✅
Why this score: All claims confirmed by recent developments.
- Niantic sold Pokémon GO to Scopely in March 2025, spun off geospatial AI into Niantic Spatial.
- Partnership with Coco Robotics (US/Europe delivery robots) uses VPS/LGM, announced March 2026.
- VPS trained on >30 billion images from user scans in games like Pokémon GO/Ingress.
Original quote: “company Niantic spun off after selling Pokémon GO, announced a partnership with Coco Robotics, a startup that runs delivery robots in cities across the US and Europe. Those robots are now navigating sidewalks using a visual positioning system trained [music] on 30 billion images.”
Pokémon GO players unknowingly built massive visual dataset from phone images while catching Pokémon — Solid (85/100)
At 0:47
True on the data collection via gameplay — players scanned landmarks for PokéStops/gyms, creating 30B+ images. 'Nobody told them' is more vibe than strict fact, but ethics angle is fair. 🎮🕵️
Why this score: Core claim holds up, with minor opinion tint.
- Niantic's VPS built on 30B images + 300M scans from Pokémon GO/Ingress players photographing landmarks for AR tasks.
- Players got in-game rewards, unaware of enterprise value (now powering robots).
- 'Largest dataset' is subjective but matches scale vs. Street View.
Original quote: “those images came from your phone while you were catching Pokémon. Players thought the only thing at stake was a shiny Charizard at the local gym, but these players were actually building one of the largest real-world visual data sets in history and nobody told them.”
Niantic VPS locates to centimeters using photos of buildings/landmarks, no GPS needed — Verified (95/100)
At 1:43
Dead accurate — VPS 2.0 (April 2026) hits cm-level precision by matching camera feeds to 3D maps from billions of images. Fixes GPS fails in cities. Genius tech. 🗺️🔍
Why this score: Precisely matches Niantic's documented capabilities.
- VPS compares live photos to vast 3D geospatial dataset for cm-accurate positioning without GPS.
- VPS 2.0 unifies global/mesh positioning, reliable in urban canyons or unmapped areas.
- First used in Coco robots, expanding to drones/AR.
Original quote: “Niantic's visual positioning system can pinpoint your location on a map to within a few centimeters. Not using GPS, but by looking at photos of nearby buildings and landmarks. You feed it a handful of pictures of your surroundings and it tells you exactly where you are on Earth down to the…”
Niantic's AR scan tasks collected 1M scans/week by 2024 for 3D models — OK (70/100)
At 1:57
AR scans are real (players built 3D models for rewards), and scale is huge — but exact '1M/week by 2024' not in public sources. Plausible given 300M total scans. 📈🤔
Why this score: Strong on concept, specific stat unverifiable.
- AR Scan tasks in Pokémon GO/Ingress rewarded players for location scans, fueling spatial model.
- Total: 300M+ scans confirmed; weekly rate aligns with growth to VPS/LGM but no exact 1M/week cited.
- Ties to web context on data power.
Original quote: “Niantic introduced augmented reality or AR scan tasks. >> [music] >> These are missions that rewarded players for scanning real-world locations to create 3D models. By 2024, they were collecting up to a million new scans per week.”
Niantic Spatial CTO quoted saying they know your position to cm accuracy — Verified (95/100)
At 2:30
Direct quote from Brian McClendon checks out perfectly — spot on, and yeah, it is kinda creepy 😬📍
Why this score: Quote is accurate and verifiable.
- Brian McClendon, ex-CTO of Niantic Spatial, publicly stated this capability of their Visual Positioning System (VPS).
- Matches web context confirming centimeter-level accuracy from player data.
- Speaker's sarcasm highlights privacy angle, but the fact stands solid.
Original quote: “Brian McClendon, the CTO of Niantic Spatial, said it plainly, quote, "We know where you're standing within several centimeters of accuracy and most importantly, where you're looking." That's not creepy to say at all.”
Niantic sold games division incl. Pokémon GO to Scopely for $3.5B in March 2025 — Solid (85/100)
At 3:09
Sale happened, price exact, games listed correct — closed a bit later in May but core claim nails it 💰✅
Why this score: Deal details match reports with minor timeline nuance.
- Announced/sold in March 2025 for $3.5B to Scopely, officially closed May 29, 2025.
- Included Pokémon GO, Pikmin Bloom, Monster Hunter Now as stated.
- Video's 'March 2025' refers to announcement; no material inaccuracy.
Original quote: “In March 2025, Niantic sold its entire games division, which included Pokémon GO, Pikmin Bloom, and Monster Hunter Now to Scopely for 3 and 1/2 billion dollars.”
Scopely owned by Saudi's Savvy Games Group, so Pokémon GO is Saudi-owned — Verified (92/100)
At 3:39
Ownership chain 100% accurate — PIF via Savvy owns Scopely, now with Pokémon GO 🕌🎮
Why this score: Corporate ownership trail confirmed.
- Scopely acquired by Savvy Games Group (PIF subsidiary) in 2023.
- Savvy is Saudi PIF's gaming arm, chaired by Crown Prince.
- Post-sale, Pokémon GO under Scopely = effectively Saudi-owned.
Original quote: “And Scopely is owned by the Savvy Games Group, which is the Saudi Royal Investment Fund's gaming arm. So, Pokémon GO... is now owned by Saudi Arabia.”
Niantic spun off spatial AI as Niantic Spatial, owns Pokémon GO training data — Solid (88/100)
At 4:09
Spin-off real, data from 8+ years of gameplay stays with them — players built it unwittingly 🤖🗺️
Why this score: Niantic Spatial formation and data retention accurate.
- Spun off in May 2025 with $250M capitalization, focuses on geospatial AI.
- Retains VPS model trained on billions of Pokémon GO images over 8 years.
- Games sold, but core spatial data/assets kept by Niantic Spatial.
Original quote: “Niantic kept the spatial AI business... rebranding the enterprise side as Niantic Spatial. So, the AI model trained on 8 years of historical Pokémon GO data belongs to Niantic Spatial.”
Pokémon Go provides centimeter-accurate, weather-tagged location data — Verified (95/100)
At 4:30
Spot on — Niantic's data is legit centimeter-level precise with all that metadata richness. Players built this goldmine without knowing 🔥📍
Why this score: Claim fully verified by Niantic's own tech specs and recent reports.
- Web context confirms centimeter-level accuracy from 30B+ AR images crowdsourced via Pokémon Go/Ingress over years.
- Data includes diverse angles, lighting, weather tags — exactly as described, powering VPS for robots.
- No exaggeration; this diversity makes models robust vs traditional single-angle maps.
Original quote: “The valuable asset isn't your identity. It is, but it's the location data. The centimeter-accurate weather-tagged angle-diverse, metadata-rich location data.”
Niantic Spatial's large geospatial model predicts location from photos like LLM predicts words — Verified (92/100)
At 4:52
Nailed the analogy — LGM is exactly the 'spatial LLM' Niantic announced. Show a pic, get your spot on Earth. Mind-blowing tech 🗺️🤖
Why this score: Direct match to Niantic Spatial's official description post-2025 spin-off.
- LGM uses 30B+ images to geolocate from single photos with high accuracy.
- Analogy to LLMs is precise: visual input → location output, enabling embodied AI.
- Ties to video overview; recent (2024-2026) developments confirmed in web context.
Original quote: “What Niantic Spatial is building is what they call a large geospatial model. Think of it as the spatial equivalent of a large language model. Instead of predicting the next word in a sequence, it predicts your location from visual input. You show it a photo of a building and it tells you where in…”
Pokémon Go data: thousands of images per spot from all angles, weather, over 8 years — Solid (85/100)
At 5:12
Thousands per location over 8+ years? Checks out with 30B total images. Way better than Google Maps' boring single shots 🌤️🌧️
Why this score: Strongly supported, though exact 'thousands' is illustrative not literal count.
- Web context: 30B+ images from Pokémon Go (2016+) = massive diversity vs traditional datasets (e.g., one fixed-angle image).
- 8 years aligns with game launch; diversity in angles/lighting/weather confirmed for robust ML.
- Minor caveat: 'thousands' per spot varies by popularity, but principle holds.
Original quote: “The reason this works is because of how the training data was collected. Traditional mapping data sets give you one image per location taken from a fixed height at one time. Pokémon GO's data set gives you thousands of images per location taken from every conceivable angle across every lighting and…”
LGM apps: AR glasses, autonomous vehicles, construction, urban planning, surveillance potential — Verified (88/100)
At 5:46
All real apps — from Coco delivery bots to AR overlays and yes, surveillance vibes flagged by PopSci. GPS killer in cities 🚀👀
Why this score: Applications confirmed across sources; surveillance is valid concern, not speculation.
- Robotics (Coco), AR glasses, AVs, construction/urban uses explicitly in web context.
- PopSci coverage notes law enforcement appeal; Niantic precedent for data access exists.
- No plans announced for gov use, but capability + precedent = fair point.
Original quote: “And the applications extend way beyond delivery robots. AR glasses need this kind of positioning to overlay digital content on the physical world. Autonomous vehicles need it for navigation in urban areas where GPS bounces off buildings. Construction companies need it for site mapping. Urban…”
reCAPTCHA trained Waymo AI with user image labeling — Verified (95/100)
At 6:30
Spot on — reCAPTCHA was straight-up free labor for Waymo's self-driving tech. Wild how we all pitched in unknowingly 😎✅
Why this score: Dead accurate historical fact. Google's reCAPTCHA indeed used user challenges to label images for computer vision, directly feeding Waymo datasets.
- Confirmed by multiple reports: users identified objects like bikes/traffic lights for over 15 years.
- Ties perfectly to the video's thesis on hidden data collection via games/apps.
Original quote: “Google's reCAPTCHA had you clicking on traffic lights and crosswalks to prove you were human. What it was actually doing was training computer vision models for self-driving cars. You were doing free labor for Waymo every time you identified a bicycle in a grid of photos.”
Niantic got 30B images from 143M Pokémon Go players over 8 years — Solid (85/100)
At 6:54
Nailed the insane scale — 30B images from players is legit Niantic Spatial gold. Numbers track with recent reports 📈✅
Why this score: Matches video overview and recent Niantic disclosures. Pokémon Go launched 2016 (8 years by 2024), peaked at 143M downloads/users.
- 30B+ images via AR scans/mapping confirmed for VPS training.
- Ties to March 2026 robotics announcements; minor uncertainty on exact peak users but directionally spot-on.
Original quote: “Niantic just ran it for 8 years at a scale of 30 billion images across 143 million people at its peak.”
AR scan consent hid true AI/robotics data use and Saudi ownership — Solid (80/100)
At 7:07
Harsh but fair — consent was buried in game bait, not upfront about AI/delivery bots or Saudi ties. Precedent-setter 🚩✅
Why this score: Core consent critique holds up. Players scanned for rewards without clear VPS/robotics disclosure.
- Niantic acquired by Scopely (Savvy Games/PIF Saudi fund) in 2025, spun off Spatial for licensing (e.g., Coco Robotics).
- No evidence of explicit robotics warnings in prompts; standard ToS covers broad data use.
Original quote: “When players opted into AR scans, they were presented with a game feature. Scan this statue, get some items. Nobody was presented with, "Your scan data will train a commercial AI product that will be licensed to robotics companies to navigate delivery robots on public sidewalks built on a platform…”
2018 Strava heatmap exposed secret military bases — Verified (100/100)
At 8:01
Classic example — Strava lit up secret bases like Christmas trees. Perfect analogy for fitness data risks 🎯✅
Why this score: Iconic, well-documented incident. Strava's 2018 Global Heatmap aggregated fitness tracker data, revealing activity patterns at classified sites worldwide.
- US military issued guidance; directly parallels Pokémon Go security concerns mentioned next.
- Strengthens video's broader data repurposing argument.
Original quote: “In 2018, Strava's heat map accidentally revealed the locations of secret military bases because soldiers were tracking their runs.”
Niantic got 30B images via gamified collection and $3.5B exit — Solid (80/100)
At 8:30
30B images spot-on from player scans — $3.5B sale to Scopely happened March 2025. Sneaky but real strategy. 📱💰
Why this score: The claim accurately reflects Niantic's data practices and exit. Key confirmations:
- Niantic collected vast geospatial data (AR scans) from Pokémon GO players for VPS/LGM, with web context noting 10M+ locations and ongoing weekly scans building toward massive datasets like the cited 30B images.
- $3.5B sale of Niantic's games unit (including Pokémon GO) to Scopely confirmed in March 2025.
- Privacy policies (updated April 2025/2026) do bury data usage details, supporting the 'gamify and bury' critique, though Niantic claims transparency.
Original quote: “Gamify the data collection, bury the terms in a privacy policy nobody reads, and by the time anyone notices, you've got 30 billion images and a 3 and 1/2 billion-dollar exit.”
Top AI datasets built by unaware people, not data centers — Opinion (50/100)
At 8:48
Bold take but fair — Pokémon players unknowingly fed Niantic's AI beast with real-world scans. Gamification genius or creepy? 🤔
Why this score: This is persuasive commentary on Niantic's model, not a strict fact. Why it resonates:
- Player data (AR scans for PokéStops/Gyms) directly built Niantic Spatial's VPS for robots, with users often unaware of downstream AI training uses.
- Broader context: High-quality human geospatial data is prized amid AI 'model collapse' concerns; Niantic's approach exemplifies 'crowdsourced' collection.
- Not exaggerated — privacy debates confirm low awareness despite ToS disclosures.
Original quote: “The most valuable AI training data sets in the world aren't being assembled in data centers. They're being built by people who have no idea they're building them.”
See the full analysis with timestamps →