Why is my internet slow? A step-by-step diagnosis

The short version

• Test in order, from the provider inward: check for an outage and restart, run a speed test on a cable plugged into the router, then test Wi-Fi beside the router and in the slow room. The first test that comes out slow points to the link at fault.
• Wired slow too? The problem is the provider, the modem or gateway, or your router. Compare the wired result with your plan's typical speed on its broadband label, not the headline speed, and run the provider's own equipment test if it has one.
• Wired fast, Wi-Fi slow? Then the connection into your home is fine. Xfinity says differences between its gateway test and device tests are "most often caused by WiFi connections, rather than by the network itself". Fix coverage before you call the provider.
• Speed fine but calls and games stutter? That's latency under load (bufferbloat), not a lack of speed. A normal speed test won't show it; a loaded-latency test will.
• The caveat: one test proves little. Results vary with the server, the device, the app and the time of day, so repeat the test that matters, wired, at the times the problem happens.

Start with the symptom

How it's slow narrows the cause before you run anything. Find the row closest to what you see, then run its first test.

Step 1: rule out an outage, then restart

Check your provider's outage page or app before you change anything. Xfinity's app and Outage Map show "the estimated time to fix the outage and status of repairs" for your address, and AT&T has an outage check and outage alerts in its support pages. If there's an outage, nothing at home will fix it; wait for the repair.

If there's no outage, restart the modem or gateway and then the router. Ookla, which runs Speedtest, calls this "the stock advice, but it solves an astonishing number of problems". Xfinity's in-app gateway test does the same when a gateway that matches the plan tests slow: it prompts a restart and tests again.

Step 2: run a clean wired test

This is the test that splits the problem in half. Plug a computer into one of the router's LAN ports with an Ethernet cable and run a speed test. Ookla's advice for a clean result: stop downloads and close programs using the connection, and use a wired connection "to eliminate the possibility of wireless interference". Use the Speedtest app rather than the website, since Ookla says browsers "may provide different results, particularly on high-speed connections". Turn off any VPN for the test, and if results stay oddly low, briefly disable antivirus or firewall software, which Ookla says can intercept test traffic.

If your provider runs its own equipment test, use it too. It measures the connection into your home without your Wi-Fi or devices in the way:

• Xfinity (rented gateways, in the Xfinity app: WiFi, Troubleshoot, Test your whole home network): "If your speed test results are higher than 90% of the speeds included with your Xfinity Internet plan, your network connection is working properly." Below that, it offers a gateway replacement or a restart, and then a technician visit if speeds stay low.
• AT&T (its equipment speed test): "checks the speed between your AT&T All-Fi Hub or Wi-Fi Gateway and our network. It reflects the speed coming into your home."

Verizon and GFiber have similar tests; the multi-gig guide lists them.

Compare it with the right number

• Your plan's broadband label, not the advertised tier. FCC rules (47 CFR 8.1) require a broadband label for each plan a provider sells, and the FCC summarizes the label as having to "disclose typical upload and download speed and typical latency". Providers that offer online account portals must also make each customer's label "easily accessible" there. The FCC's July 2026 order (FCC 26-48) keeps that portal requirement and would let a link or icon satisfy it, but that part of the order has been delayed indefinitely. Look for the label in your account.
• The link's own ceiling. A 1 Gbps Ethernet port carries about 940 Mbps of data after protocol overhead, so a gigabit plan that tests at 940 Mbps is working. The arithmetic is here.
• Units. Speed tests report megabits per second (Mbps); download managers and game stores often show megabytes per second (MB/s). Ookla's example: 100 kilobytes per second is 0.8 megabits per second. Arithmetic: a game store showing 50 MB/s is running at 400 Mbps.

For context, the last fixed-broadband report from the FCC's Measuring Broadband America program (the Thirteenth, released in August 2024 and based on September and October 2022 measurements from test routers in panelists' homes; the program is no longer collecting data) found that eight of the 12 provider and technology combinations it measured delivered 100% or more of advertised download speed at peak hours, and the other four 86% to 90%. On its "80/80" consistency measure (the speed at least 80% of subscribers got at least 80% of the peak time), DSL providers delivered 63% to 72% of advertised speed. A wired result well below your plan on a cable or fiber line is worth a call; on DSL it may be what the line delivers.

What the result tells you: if the wired test or the provider's equipment test is well below the plan after a restart, the problem is outside your Wi-Fi: go to step 5 to see whether it's the time of day, check your own router if you have one, then step 7. If the wired test is close to the plan, your connection is fine and the problem is inside your home: carry on to step 3.

Step 3: test Wi-Fi beside the router, then in the slow room

Run the same test on a laptop or phone a few feet from the router, then where you actually use it. The FCC's report lists the home's Wi-Fi among the bottlenecks its measurements leave out, noting that "some Wi-Fi connections may be limited to tens of Mbps" even on a plan of hundreds. Ookla adds that devices "can have very different Wi-Fi and cellular radio capabilities", so a phone may not be able to show your full speed at all.

If you already have mesh and some rooms are still slow, the usual cause is a weak link between nodes: see why mesh doesn't fix slow rooms. If you're choosing what to add, extender, mesh or access point compares the options.

Step 4: test latency under load

If pages, calls or games feel slow while speed tests look fine, measure latency (the round-trip delay) while the connection is busy. A plain ping measures a quiet line and misses the problem. The FCC began reporting this "working latency" in its Thirteenth report and found that latency under download load "generally is significantly higher than idle latency", with a bigger gap on DSL. Its idle figures, for comparison: fiber 7 to 14 ms, cable 12 to 24 ms, DSL 23 to 34 ms. The FCC also notes that some providers have since added active queue management to their networks and improved.

Two easy tests:

• Speedtest apps for Android and iOS show ping when idle, during download and during upload. Ookla says a high idle ping is "an overall latency problem": test another device, and if both are affected, restart the router.
• On a Mac, type networkQuality in Terminal (macOS Monterey or later). Apple rates the result Low, Medium or High; Low means connections "might be unreliable" when other devices are using the network heavily.

If latency jumps under load, the fix is usually smart queue management on your router. The bufferbloat guide has more tests, the grades, and the settings for each router brand. For calls specifically, see the work-from-home troubleshooting playbook; for games, check your Wi-Fi hop.

Step 5: check the time of day

The FCC measures provider performance at peak hours, "weeknights between 7:00 p.m. to 11:00 p.m. local time", because that's when local networks see the highest demand. GFiber's support says the same: "internet traffic across the web peaks between 7:00 PM and 11:00 PM." Across the providers it measured, the FCC found "only a slight degradation from off-peak to peak hour performance" for most subscribers.

So run your wired test twice a day for a few days: once between 7 and 11 p.m., and once at a quiet time such as late morning. M-Lab, which runs another widely used test, makes the same point: "testing regularly and looking at the data over time" is much more reliable than one result.

• Wired drops sharply only at peak: likely congestion in the provider's network. Write down the times and results for the call in step 7.
• Wired steady, Wi-Fi drops in the evening: your home is busier then. Streaming takes "a large chunk of your available bandwidth", as GFiber puts it; see how bandwidth splits between devices and how much speed streaming needs.

Step 6: rule out the device

If one device is slow wherever it is while others are fine in the same spot, look at the device. GFiber's list of speed factors starts here: "Older devices, aging operating systems, outdated web browsers, or slow internal network cards can bottleneck your speed", and with a VPN "your speed is capped by how fast the VPN server can process your data". Check, in this order:

1. VPN and security software: disconnect the VPN and test again; briefly disable antivirus or firewall software for one test (and turn it back on).
2. App vs browser: run the Speedtest app as well as the website.
3. The connection itself: a wired port that negotiated 100 Mbps instead of 1 Gbps, or a weak Wi-Fi link, caps everything on that device. How to read the negotiated speed on Windows and Mac.
4. Updates and age: update the operating system and Wi-Fi driver. If the device's Wi-Fi is several generations old, it may simply be the slowest thing on the network.

Is it your router?

If the provider's equipment test is fast but your wired test behind your own router is slow, test once with the computer plugged straight into the provider's gateway or modem (where it has a spare port). Fast there and slow behind your router means the router's ports or routing are the limit. To check whether a router feature is the cause, turn off intrusion prevention, VPN and traffic shaping for one test and see whether the result jumps. The multi-gig guide covers internet ports and routing, and when to replace your router covers the age question.

Two things that rarely cause slowness, though they're often blamed: double NAT and carrier-grade NAT. Google's help says that for most people double NAT "doesn't create a noticeable effect", and CGNAT tests in RFC 7021 found browsing and email worked normally. Both break reachability (port forwarding, hosting, remote access), not speed: see double NAT and CGNAT explained.

Why different speed tests disagree

Two tests on the same line can disagree without either being wrong:

• Connections: M-Lab notes that "other popular tests" transfer data "across multiple connections" at once, while its own NDT test uses a single connection, which it says "is more likely to help diagnose problems in the network".
• Server location: M-Lab's tests "are always conducted off-net", outside your provider's network. It says results from servers inside your provider's network "are often higher", because the path is shorter and one provider controls all of it.
• Timing: M-Lab notes that "A test conducted five minutes ago may show very different results from a test conducted twenty minutes ago."

Use one test, on one wired computer, to the same server, when you compare times of day or before and after a change.

Step 7: what to tell your provider

Ookla's advice: once you've tested wired and restarted the modem and router, "you should now have enough evidence to contact your ISP if there's a mismatch between your plan and the speeds you're getting." Make the call short and specific:

• The plan and its label: the plan name and the typical download, upload and latency figures on its broadband label.
• The provider's own equipment test, with the date and time, if it offers one.
• Your wired results: computer plugged into the gateway or router, app used, server, and times, including evening and daytime results from step 5.
• What you've ruled out: outage checked, equipment restarted, VPN off, a second computer tested, Wi-Fi not involved.
• Latency under load, if calls or games are the problem: idle and loaded figures.
• What you're asking for: a line or signal check from their side, a replacement gateway or modem if the equipment test fails, or a technician visit. Xfinity's own flow ends in a technician visit when its gateway test stays under 90% after a restart.

If your provider's equipment is the problem and you'd rather own it, what each provider allows is in our guide to owning your network. What the rental costs, and how many months of fees a modem of your own takes to pay for itself, are in ISP equipment rental fees; which modem to buy is in buying your own cable modem.