Choosing a proxy provider is mostly a matter of asking the right twelve questions before you pay — and checking the answers yourself during a trial instead of trusting the landing page. Every provider claims millions of IPs, 99.9% uptime and “clean” addresses. Those numbers are rarely false, but they are almost always measured in a way that flatters the seller: a pool counted across all countries, uptime of the gateway rather than of the exit IPs, cleanliness judged by the provider’s own tests. This checklist turns each marketing claim into a question with a verifiable answer, and ends with a scoring table you can fill in for two or three shortlisted providers.
Before the questions: define the job
The same provider can be excellent for one task and useless for another, so write down three things first. What you will do through the proxies (collect public data, run accounts, test ads, check geo-specific content). Where the traffic must appear to come from (country, city, sometimes a specific mobile carrier). How the target reacts to automation — a public price page and a social network with aggressive anti-fraud need very different IPs. If you are not sure which IP type fits, start with residential vs datacenter vs mobile proxies; the questions below assume you already know roughly what you need.
1. What kind of IPs are these, and where do they come from?
Ask the provider to name the IP types it sells and how it obtains them. Datacenter ranges are leased from hosting companies; ISP (“static residential”) ranges are registered to consumer providers but hosted on servers; rotating residential and mobile IPs belong to real household or cellular connections. For residential pools the sourcing question matters twice: for quality and for ethics. A reputable provider can explain how the owners of those connections agreed to share bandwidth — typically an SDK inside an app, with explicit consent and payment. Vague answers here are a red flag. The background is explained in how the proxy market works.
How to verify: take ten IPs from a trial, look up their ASN and owner in any WHOIS or IP-intelligence service. “Residential” IPs that resolve to hosting companies are not residential.
2. How large is the pool where I actually need it?
A pool of “72 million IPs” is a global, cumulative figure — often every address ever seen over a month. What matters is how many addresses are available right now in your country or city. Ask for a per-country breakdown and, for city targeting, for the number of concurrently online IPs.
How to verify: during the trial, request a few hundred rotating connections targeted at your location and count unique exit IPs. If a “huge” pool keeps handing you the same few dozen addresses, you have your answer.
3. How precisely can I choose the location?
Targeting granularity differs widely: country only; country and region; city; ASN or specific mobile carrier. More precision usually costs more and shrinks the available pool. Also ask how the provider determines location — by its own database or by a commercial one — because websites use their own databases and the two often disagree. Why that happens and how much it matters is covered in why IP databases disagree.
4. How does rotation work?
For residential and mobile proxies rotation is the core feature, so pin down the details. Can you get a new IP on every request? How long can a “sticky” session hold the same IP — one minute, ten, thirty, until the device goes offline? For mobile proxies, is there a link or API call to force an IP change, and how long does the change take? What happens when the device behind a sticky session disconnects — do you silently get a different IP mid-session? The concepts are explained in sticky vs rotating sessions.
5. Which protocols are supported?
HTTP(S) is universal; SOCKS5 is needed for non-web traffic and for tools that only speak SOCKS. Ask whether SOCKS5 supports remote DNS resolution and UDP, which matters for voice, some games and QUIC. The practical differences are in SOCKS5 vs HTTP proxies, and the protocol details in SOCKS5 in depth.
6. How is access controlled?
Two common methods: username and password, or an allow-list of your own IP addresses. Password authentication is portable; IP allow-listing is convenient on fixed servers and keeps credentials out of configuration files. Check whether you can create sub-users with separate limits — useful for teams and for isolating projects — and whether location and session parameters are set through the username string, a separate port, or the dashboard.
7. What exactly am I paying for?
Proxies are sold per gigabyte, per IP, per port or per period, and each model hides different traps. With per-GB pricing, ask what counts as traffic: are TLS handshakes and headers included, is traffic counted in both directions, do unused gigabytes expire at the end of the month? With per-IP pricing, ask whether the IP is dedicated or shared with other customers, and whether a replacement is free if it gets blocked. With “unlimited” plans, ask for the fair-use limit — there always is one. A detailed comparison of the models is in proxy pricing per GB, per IP and per port.
8. How fast and how stable is it in practice?
An uptime figure usually describes the provider’s gateway, not the individual exit IPs behind it. What you experience is the success rate of real requests: how many connections complete, how long they take, how often a session drops. Measure it on your own target during the trial — latency, throughput and success rate at the concurrency you will actually use. How to measure proxy speed and uptime gives a method and scripts.
9. How clean are the IPs?
“Clean” should mean three things: the IPs are not on major spam and abuse blocklists, IP-intelligence services classify them as the type you are paying for, and the target site does not already challenge them. The first two you can check yourself; the third only a trial on the real target will show. Test for leaks at the same time — a proxy that exposes your real IP through headers or DNS is not clean in any sense. The full procedure is in how to test a proxy for anonymity, leaks and speed.
10. What are the limits on concurrency and bandwidth?
Ask how many simultaneous connections or threads a plan allows, whether bandwidth per connection is capped, and what happens when you exceed a limit — errors, throttling or an automatic upgrade charge. Scrapers and automation tools open many parallel connections; a limit that is invisible at ten threads becomes the bottleneck at two hundred.
11. What are the rules, and what is logged?
Read the acceptable-use policy. Legitimate providers forbid fraud, spam and attacks, and many require identity verification (KYC) for residential access. That is a good sign, not an inconvenience: a provider that accepts anyone for anything will soon have IPs that every site distrusts. Ask what connection data is logged and for how long. Where the legal lines actually are is discussed in are proxies legal.
12. What happens when something goes wrong?
Check support hours and channels, the typical response time, whether there is a free trial or a paid test package, and the refund policy. For work you will automate, look at the API: can you list IPs, change locations, rotate and read usage without the dashboard? Ask one technical question before buying — how support answers it tells you more than any review.
Scoring table
Fill in one column per shortlisted provider after a trial. Score each line from 0 to 2: 0 — not available or failed the check; 1 — available with caveats; 2 — confirmed on your own test.
| Question | What to check yourself | Provider A | Provider B |
|---|---|---|---|
| 1. IP type and origin | ASN and owner of 10 trial IPs | ||
| 2. Pool in your geo | Unique IPs from a few hundred rotations | ||
| 3. Targeting | Location of trial IPs in two databases | ||
| 4. Rotation | Sticky session length, forced change time | ||
| 5. Protocols | SOCKS5 with remote DNS, UDP if needed | ||
| 6. Access control | Password, IP allow-list, sub-users | ||
| 7. Pricing | Traffic accounting on a measured transfer | ||
| 8. Speed and stability | Success rate and latency on your target | ||
| 9. Cleanliness | Blocklists, IP-intelligence type, leak test | ||
| 10. Limits | Behaviour at your real concurrency | ||
| 11. Rules and logs | Acceptable-use policy, KYC, log retention | ||
| 12. Support | Answer to one technical question |
Red flags
- “Residential” IPs that belong to hosting ASNs.
- No per-country figures — only a single global pool size.
- No trial and no refund, combined with a long minimum commitment.
- An acceptable-use policy that allows everything, or none at all.
- Traffic counted in a way the provider will not explain.
- Unlimited plans without a stated fair-use limit.
Once you have a winner, the next step is configuring it in your tools: see how to set up a proxy in Chrome, Firefox, Windows and macOS and, for code, proxies in Python.
Frequently asked questions
Is a bigger IP pool always better?
No. What matters is the number of addresses available at the same time in the location you need. A smaller pool concentrated in your country can serve you better than a huge global pool with a thin presence there.
How long should a proxy trial be?
Long enough to run your real workload at your real concurrency, usually one to three days. A five-minute test with a single request only shows that the gateway answers.
Should I choose the cheapest price per gigabyte?
Compare the cost of a successful result, not of a gigabyte. If a cheaper pool fails twice as often, retries eat the saving and the target sees more failed attempts from you.
Is identity verification a bad sign?
Usually the opposite. Providers that verify customers keep abusers out of the pool, and abusers are what make IPs distrusted by websites.
Can I use one provider for everything?
Often not. Many teams keep a datacenter provider for bulk public data and a residential or mobile provider for account work, because no single pool is both the cheapest and the most trusted.