How to Detect ISP Throttling (and Prove It)
TL;DR. Modern ISP throttling almost never looks like a blunt speed cap. It looks like “Netflix is slow but everything else is fine”, or “Zoom drops quality between 7 and 11 PM”, or “YouTube tops out at 480p on this carrier”. Generic speed-test sites won't catch it because they hit dedicated test servers that aren't being shaped. The fix: compare your speed to several different destinations and look for the odd one out. The widget further down does exactly that.
What “throttling” usually means in 2025
In the popular sense, “throttling” covers any case where an ISP intentionally degrades a class of traffic. In practice there are three distinct shapes that show up in the wild:
- Plan-tier caps. US mobile carriers publish them: a basic plan tops video out at 480p, premium tier unlocks 1080p or 4K. Same data, different speed depending on what port and host the traffic is going to.
- Peering-induced congestion. Less a deliberate cap and more a refusal to upgrade an interconnect. Traffic to Netflix or another high-volume CDN saturates the link; traffic to other destinations doesn't. The user perceives it as throttling even when no config rule was changed.
- Active protocol shaping. Deep-packet-inspection rules at the ISP's border that recognise specific protocols (BitTorrent, video streaming patterns) and apply a QoS class with lower priority or a hard rate limit. This is what Comcast got caught doing to BitTorrent in 2007 and what the Wehe project documented across multiple US carriers in 2022.
All three are detectable from a browser. The first looks like “video sites are slow, image CDNs aren't” even though both should saturate your connection. The second looks like “a specific CDN is slow regardless of region”. The third looks like “Netflix is slow only when the traffic pattern looks like video, not when it's the same bytes randomised”.
Probe several hosts from this page
The widget below times your browser's connection to a handful of representative hosts — YouTube's CDN, Netflix's CDN, Steam, Cloudflare, and our own origin as a control. It runs a small number of samples per host and compares the medians. If your ISP is throttling a specific service, you'll see that one host sit consistently above the others; the verdict line will tell you when the gap is large enough to be suspicious.
We measure connection time to 5 hosts (5 samples each) and compare the median round-trip. An ISP that throttles a specific video service typically also slows the path to it — if YouTube or Netflix is consistently several times slower than the others, that is what throttling looks like at the browser level.
Caveats: ISPs throttle in different ways (per-port, per-protocol, per-host, time-of-day). Latency to a brand domain isn't the same as bulk-download throttling. The probe is a hint, not a court exhibit; for a real assessment run it from several locations and combine with a sustained-download speed test from each target.
The probe measures connection latency, not bulk-download throughput — you can't fetch a multi-megabyte file from a cross-origin host without CORS letting you. Latency catches QoS-class differences and saturated-peering issues, which between them are how most modern throttling shows up. To catch a deliberate bandwidth cap on video specifically, you'd need a sustained-download test from each host; see the section below.
Doing it properly with sustained downloads
For a real assessment, you want to measure actual throughput to each candidate target. The methodology that survived peer review is the one from Northeastern's Wehe project:
- Record a real session to a target service (Netflix HLS stream, YouTube TS segments, etc.).
- Replay the exact same bytes from a control server twice: once as the original protocol, once with the bytes random-shuffled so DPI can't recognise it as video.
- Compare the resulting throughput. If the recognised version is consistently slower than the shuffled version, the ISP is differentiating based on traffic recognition — which is what throttling is.
The Wehe project ships this as a mobile app (wehe.meddle.mobi) and the team has run their data through ACM IMC 2022 with a sample of about 1.6 million tests from real users.
Lower-effort home alternatives:
- Test via VPN. If your speed to Netflix is poor but dramatically improves when you tunnel through a VPN to the same general region, your ISP is recognising and slowing Netflix-shaped traffic. The VPN obscures the destination, so DPI can't classify it. This is the single most reliable home test there is.
- Run fast.com. Netflix's own speed test runs over Netflix's Open Connect infrastructure, so it sees exactly the conditions Netflix sees. Compare it to a generic Cloudflare or Ookla test; a sustained gap is meaningful.
- Mobile carrier app self-tests. T-Mobile, Verizon and AT&T all publish their video-resolution caps in their plan terms. If your plan is documented as “480p max video”, that's throttling by definition; nothing to prove.
Where this actually happens
The pattern repeats: throttling is technically easy, hard to prove without specialised tooling, and usually only acknowledged after a journalist or academic exposes it. None of these were inferred from a single user's speed test; all are documented in court filings, FCC complaints, or peer-reviewed studies.
- Aug 2007Comcast vs BitTorrentComcast → BitTorrent / Gnutella seedersHow: TCP RST injection — Comcast forged reset packets to terminate seeding connections.Outcome: AP exposed it October 2007. The FCC ruled it a violation of the 2005 Internet Policy Statement in 2008; Comcast appealed and the DC Circuit overturned on the FCC’s authority in 2010 (Comcast v. FCC).
- Feb 2014Netflix vs Comcast / Verizon (peering wars)Comcast, Verizon, Time Warner Cable → Netflix CDN (Open Connect via Cogent / Level 3 transit)How: Refusing to upgrade peering links until Netflix paid for direct interconnect.Outcome: Netflix’s reported speeds on Comcast dropped from 2.5 Mbps to 1.4 Mbps over six months, then jumped back as soon as Netflix signed a paid-peering deal in February 2014.
- Apr 2017AT&T zero-rating DirecTV NowAT&T Mobility → Competing streaming servicesHow: Counted DirecTV Now traffic against $0 of the user’s data cap while charging full quota for Netflix / YouTube.Outcome: FCC investigation concluded it violated Open Internet rules in January 2017; rescinded by the new administration two weeks later.
- Dec 2017FCC repeals 2015 net-neutrality rulesn/a → n/aHow: Restoring Internet Freedom Order reclassifies broadband as Title I, removing the FCC’s authority to enforce no-throttling rules.Outcome: Took effect June 2018. State-level rules (notably California SB 822, 2018) re-imposed similar rules in their jurisdictions.
- Sep 2018Verizon throttles Santa Clara fire departmentVerizon → Mendocino Complex fire response coordinationHow: Plan-cap throttling triggered during an active wildfire, reducing emergency-services connection from 50+ Mbps to <0.5 Mbps until the department upgraded their plan.Outcome: Documented in court filings supporting California’s SB 822. Verizon called it a customer-service failure rather than throttling; either framing was bad PR.
- 2019 onwardsMobile video resolution capsT-Mobile, AT&T, Verizon (US) → All over-the-top videoHow: Plan tier dictates max video resolution: 480p on basic, 720p on mid, 1080p on premium. Implemented via deep-packet inspection of known streaming hosts.Outcome: Largely legal under current US rules; presented as a “feature”. The 2022 academic Wehe project showed plan-tier video caps active on most major mobile carriers worldwide.
- Feb 2022Wehe study: 24 % of US connections throttled videoMultiple US ISPs → YouTube, Netflix, Amazon, Hulu, Disney+How: Northeastern University's Wehe app tested replay of recorded streaming traffic vs randomised bytes; an ISP that throttles the recognisable streaming pattern but not the random one is throttling.Outcome: Published in the “A Large-Scale Study of ISP Traffic Differentiation” paper at ACM IMC 2022. T-Mobile, Verizon, AT&T, Sprint, and others identified as differentiating; speeds for video typically capped at 1.5–4 Mbps.
The trend across these incidents is that throttling is technically easy, gets caught only when a researcher or journalist deliberately looks for it, and is usually defended by the ISP as either a feature or an emergent property of capacity planning. Net-neutrality rules push back on the practice but enforcement is uneven across jurisdictions and changes with each administration.
If you find it, what then?
- Document. Run the test several times over several days, at different times. Screenshot the results. Save the raw timings.
- File with the regulator. In the US the FCC's consumer complaint form is at consumercomplaints.fcc.gov; in the EU your national regulator (BNetzA, Ofcom, ARCEP, etc.) usually has a similar form. State-level filings (California PUC, NY PSC) tend to get faster responses than federal in 2025.
- Press the ISP's social channels. Visible documented complaints on Twitter / Reddit have moved more throttling cases than back-channel tickets — the carrier decides quietly that the PR cost is higher than the QoS savings.
- Use a VPN as a workaround. Not a fix, but the right tool to bypass per-host throttling. See our VPN vs Tor vs Proxy article for picking the right one.
Speed-test the easy way too
The probe on this page is a comparison tool. To establish your actual baseline throughput across your connection — the number you compare individual destinations against — run our Speed Test for download / upload to a single high-capacity target.