Kagi Proxy: Account Preferences, Exit Country and Repeatable Measurement

What makes Kagi interesting from a proxy standpoint is that it cannot be used without a session. The subscription account determines who the query belongs to, and that identity is independent of the exit address. When you enable a proxy, which network the request leaves from changes; who your account is does not. If you do not establish this distinction from the start, you will attribute every difference to the wrong cause.

The second point is that the engine does not produce its result surface from a single source. Kagi states in its own documentation that it combines its own crawled indexes with external search sources; on top of that come your account's domain preferences and the filters you have selected. The exit address is only one of these inputs.

The third is a practical caveat: what is described here is for verifying how the engine's public result surface looks from different countries. Compliance with the subscription conditions and terms of service is the user's responsibility.

Identity sits with the account, location with the network

When you send a Kagi query there are two separate pieces of information on the server side. The first is your session: the cookie or token your browser carries tells which subscription the request belongs to. The second is the exit IP address the connection comes from. A proxy touches only the second. When you switch to a Frankfurt exit your account does not become a Frankfurt account; your request leaves from there.

This distinction is actually an advantage for someone doing measurement. When you send queries from two different countries with the same account, everything bound to the account — preferences, language setting, plan — stays fixed; the only input that changes is the network side. That is exactly what you want for a controlled comparison. Verify that your exit has really changed with the my IP address tool; if the country the tool shows and the country you selected in your panel are not the same, do not start measuring.

There is a reverse direction too. If you change something in the account settings and then change the exit as well, you will attribute the resulting difference to the network although it may come from the preference. The rule is simple: move only one variable per run and leave everything else as it is.

The autonomous system the exit belongs to is also an input. Datacenter blocks and home subscriber blocks sit in different ASNs, and that classification can affect how the engine handles the request (ASN and IP reputation). If you want to compare the behavior of other search engines, the search engine guides show the same method on other surfaces.

What do you need to hold fixed in a monitoring series?

SERP monitoring is not a one-off look but running the same query set at regular intervals under the same conditions. The value of a series comes from its comparability; if the conditions have shifted in between, what you are left with is a mixture of two different experiments. So the first job is to write down which variables count as frozen.

In practice a run erodes in three stages. Some of the queries you planned never complete because of timeouts, quota exhaustion or dropped connections. Some of those that do complete are not admitted to the series because their conditions broke: the exit changed midway, the language setting differs, the page loaded only partially. What remains is your real measurement. Ignoring this attrition rate is the most common cause of unexplained jumps in a series.

Filling this table in once and setting it aside is not enough; attach the same fields to the output of every run. When you see a jump in a chart six weeks later, the first place you look is not the engine but that day's exit label. Teams that keep no records are left guessing at this point.

How do an account's ranking preferences contaminate measurement?

One of Kagi's distinguishing features is that the user can intervene in the results: you can boost a domain, demote it, pin it or remove it from the list entirely. With lens and filter setups you can also confine the result set to a narrow area. In daily use this is a powerful feature; in measurement it is a permanent source of bias.

The problem is that preferences live on the account. Even if you change your exit, reset the browser, or move to another machine, the preferences come back the moment you sign in with the same account. A domain you demoted six months ago out of irritation is still down in today's measurement, and you mistake it for a regional difference.

The solution is to keep a separate measurement profile. Do not bring the account you use daily into measurement; leave preferences neutral on the account you use for measurement and keep that decision in writing. If you have had to change a preference, record the change with its date, because measurements before and after that date no longer count as part of the same series.

Separating the inputs that feed the result surface

Thinking of a result page as a single box also sets the wrong expectation of a proxy. On the Kagi side at least five inputs work at once: the engine's own crawled index, external search sources, your account's domain preferences, the active lens or filter, and the network the request leaves from. A proxy changes only the last of these.

The fastest way to understand this is to run a single-variable test. At the same time, with the same account, run the same query from two different exits and put the two outputs side by side. The part that stays common most likely comes from the index and preferences; the part that changes is the part sensitive to the network side. Once you have made this distinction, you also learn which queries are region-sensitive.

Some results do not change on any exit. For a technical definition search, a library document or a historical record, the weight carried by the country signal is low. The region-sensitive ones are generally queries about local services, current news, regulations and commercial intent. It is more efficient to spend your measurement budget on this second set.

When deciding which country the exit should come from, base it on your target audience; wherever the user you want to measure stands, that is where the exit should be. You can see the options location list . The general framework on the search visibility side is gathered on the proxy for SEO tools page.