New York City is the greatest survey-taking city in the U.S. From Albuquerque to Zephyrhills, most cities take more surveys, with better quality, during the day, and far fewer, far worse ones at night.

NYC barely slows down. At 4 AM, its slowest hour, it fields about as many survey starts as Dallas and Philadelphia at their afternoon peaks. The quality at 4 AM? Not so good. Data center towns like Ashburn, Virginia skip the cycle entirely: steady volume, sky-high fraud, all day.

Should you block every VPN user? Nope! Across 270M survey starts, proxies sit at one extreme: cheap, disposable tools for spoofing location. Corporate VPNs and private relays sit at the other, and their users are higher quality on average. Privacy without spoofing usually means a legitimate respondent. Spoofing across continents is the red flag.
‍

Are all proxies bad?

Usually, but don’t block them all. A blanket proxy block will introduce some false positives. (For example, university and corporate proxies have very legit reasons to exist.) This question was a hot topic in the comments.

‍

‍

‍

You want bots in your surveys. (No, not synthetic sample ;) )

Corporate bots, the kind that scan links for phishing before you click, are associated with 51% higher quality traffic than the market research average. Automation tied to data centers like Microsoft and AWS, or ISP router hubs, is far riskier.

From the comments:

• Show the research. I pulled fake and suspicious rates for respondents touched by Cloudflare and other corporate bot signals. They came in well below the market research mean.
• Why it makes sense. If your company scans links before you open them, someone there cares about security.

When the same respondent reaches your project through multiple panels, you get a duplicate. Nobody's fault, but you still remove it. This "friendly fraud" is 17% of all fraud and about half of all multi-accounting we see. So go easy on panels over dupes; there's a decent chance a real person is on the other end.

From the comments (the best debate of the quarter):

• Split your dupes. Few teams separate duplicates within one supplier from duplicates across suppliers, and some platforms can't route dupes to their own exit URL.
• Duplicate, fraud, or both? One camp doesn't distinguish: dupes, bots and disengaged respondents all get tossed and replaced at no cost.
• Is it really friendly? B2B respondents sometimes complete the same survey twice through different panels, with different answers. That raises questions about how authentic they were to begin with.
• Counterpoint. Panelists usually aren't told not to repeat, and may not notice until they're minutes in.

‍

Hyperscale data centers → hyperscale survey fraud

How does a city of 24,000 people end up with 45.8% of its survey respondents flagged as fraud? It sits next to 50 million square feet of server farms. Fraudsters spoof their location straight into “Data Center Alley” in Northern Virginia. ISP routing hubs carry similar risk: Farmingdale, on Long Island, isn't full of fraudsters. A huge pool of New York IPs resolves there, with fraudulent traffic mixed in.

From the comments:

• Some ISPs are a giveaway. With survey status visible, certain ISPs show 90%+ completes.
• GeoIP is shaky ground. Weak validation standards across the GeoIP industry mean IP location is easy to fake.
• Clickfarms are real, too. One researcher found a team in a building in China posing as panelists from western cities.
• Residential proxies are common, especially for browser automation services.
• People want the data. Nothing public yet, but we're considering a researcher-accessible data lake.
• Or it's golf. One theory: Farmingdale's spoofers are just after tee times at Bethpage Black ;)

87% of US traffic from Chinese phones run through a proxy or VPN

A Huawei phone taking a US survey is a Chinese clickfarm, right? That was my take when I first played with the data. Wrong. Most of these location-spoofed "US" responses come from outside China. Nigeria alone has 4x as many as China itself. 🤯

These phones are cheap and global, which makes them a favorite of fraud hubs everywhere. So a Chinese phone in a US survey is a strong sign of location spoofing, just not a sign the respondent is in China.

That's a wrap!

Thanks to everyone who commented, asked for receipts, and pushed back. You made these posts better. See you in Q4.

-- Sebastian