We ran eleven operator queries twice on 1 November 2026, once on Google's default results tab and once on the Web filter. The default tab obeyed one of the eleven, ignored the operator in seven, and returned no usable result in the other three. The Web filter obeyed nine. If your site: searches have been returning other people's pages, that is the behaviour we measured here, and the first thing to check is which tab you are on.
A second finding came out of the documentation rather than the browser. Three of the eight operators we tested are not listed on Google's help page for refining searches.
What we measured, and how
Eight operators across eleven queries, each query issued twice: once as google.com/search?q=... and once with &udm=14 appended, which selected the Web results filter. One Chromium browser driven by Hasten AI, logged out, on an Indian IP address, 1 November 2026, in two sittings on the same day with the same browser. No account, no personalisation we could control for, one location.
Each query was designed so that obedience and disobedience look different at a glance, because "the results look plausible" is not evidence that an operator was applied. The quoted-phrase test uses a nonsense string, "quokka crawl budget banana rhubarb", which should return nothing. The exclusion test checks whether the excluded word appears in the snippets. The date test checks the dates on the results.
Result 1: one of eleven on the default tab, nine of eleven on the Web filter
Every row below is the same query run on both tabs. "Ignored" means the results visibly did not obey the operator.
| Query | Default tab | Web filter (udm=14) |
|---|---|---|
| site:growthhasten.com | Obeyed | Obeyed |
| site:ahrefs.com crawl budget | Ignored | Obeyed |
| crawl budget site:ahrefs.com | Ignored | Obeyed |
| site:growthhasten.com crawl | Ignored | Obeyed |
| "quokka crawl budget banana rhubarb" | Ignored | Obeyed |
| inurl:crawl-budget | Ignored | Obeyed |
| filetype:pdf crawl budget | Ignored | Obeyed |
| crawl budget -googlebot | Ignored | Obeyed |
| crawl budget before:2015-01-01 | No results rendered | Obeyed |
| cache:ahrefs.com | No results rendered | No documents matched |
| related:ahrefs.com | Pages about Ahrefs | ahrefs.com itself |
Two of those three "no results" cells are worth separating. For cache:, the Web filter said plainly that nothing matched. For crawl budget before:2015-01-01 on the default tab, the page rendered a "Thinking" and "Searching" shell and never produced a result list at all, which is the only cell in the table where a documented operator returned no page rather than the wrong one. The same query on the Web filter returned results dated 2010 and 2001.
Result 2: the site operator survived alone and failed with company
On its own, site:growthhasten.com worked on the default tab: every result came from this site, and Google offered its usual "Do you own growthhasten.com?" prompt. Add one word and the operator stopped applying.
site:ahrefs.com crawl budget returned pages from respona.com, evendigit.com and jetoctopus.com, and nothing from ahrefs.com. Moving the operator to the end of the query produced the same result set. On the Web filter, the same query returned four ahrefs.com pages, led by their crawl budget blog post and glossary entry.
The sharpest case was site:growthhasten.com crawl. On the default tab it returned reviews of the 2019 horror film Crawl, with no result from growthhasten.com. On the Web filter it returned this site's crawl budget guide and technical SEO guide.
Result 3: quotation marks and the minus sign were both ignored
This is the result that should change how you verify things. The quoted nonsense phrase exists nowhere on the web, and the excluded word is a word every crawl budget article contains.
- Quoted phrase, default tab: photographs of quokkas and a GitHub repository called quokka, with no indication that the phrase had not been found.
- Quoted phrase, Web filter: "No results found for \"quokka crawl budget banana rhubarb\"", followed by results for the words without quotes.
- Exclusion, default tab: on
crawl budget -googlebot, snippets containing "Googlebot" appeared twice in the first three results. - Exclusion, Web filter: the excluded word did not appear in the snippets returned.
Both of those operators are documented by Google, and both are the ones people reach for when checking a fact. Anyone using quoted search to find copied text, to confirm a quotation, or to test whether a claim has a source can get a page of confident-looking results on the default tab that establish nothing.
Verify quoted phrases on the Web tab. A quoted search that stops being a quoted search does not return an error.
Result 4: three of the eight are not on Google's list
Google's help page on refining searches, read on 1 November 2026, lists quotation marks, site:, the minus sign, before:, after:, and filetype:. It does not mention inurl:, cache: or related:. The page introduces its list with "Here are some popular operators", so it does not claim to be exhaustive and absence from it is not proof that Google documents an operator nowhere. It is still the page a user is sent to, and three of the eight we tested are not on it.
That reframes two of our results. cache:ahrefs.com returned "Your search - cache:ahrefs.com - did not match any documents" on the Web filter and no result list at all on the default tab. Use the Wayback Machine for old copies of a page, or the URL Inspection tool in Search Console for what Google actually stored.
related: returned results on both tabs, and on neither did it return a set of similar sites: the default tab gave pages about Ahrefs, and the Web filter gave ahrefs.com itself plus an article about Ahrefs alternatives. For finding competitors, a proper competitor analysis starting from the queries you care about beats an operator that is neither listed nor behaving as its reputation suggests.
inurl: is the interesting one: unlisted, yet it worked cleanly on the Web filter, returning Google's own crawl budget documentation at a URL containing the string. Working and unlisted is a thing you can use and should not depend on.
How to reach the Web tab
Three ways, from fastest to most permanent. The parameter is not documented by Google, so treat it as a measured workaround that can change without notice; the Web filter itself is documented, and Google describes it as containing "text-based links to websites".
- Append the parameter to a search URL. This is what we used for every Web filter test here.
https://www.google.com/search?q=YOUR+QUERY&udm=14 - Use the Web link in the row of filters on the results page. Google says the set of filters shown "is dynamic" and tells you to use the All option to check which others are available.
- Make it your browser's default by adding a custom search engine with this URL and selecting it:
https://www.google.com/search?q=%s&udm=14
For SEO work this matters most on audit queries: checking indexation with site:, finding a page by its URL slug, confirming a quotation, or checking whether a competitor covers a topic. Our guide to checking whether a page is indexed uses Search Console rather than an operator for exactly this reason: the operator is a convenience, and Search Console is the record. We also found the two surfaces diverging in a separate test of twelve queries, where every default results page carried an AI Overview and none of the filtered pages did.
What this does not prove
One browser, one location, one day, logged out. Google varies its interfaces by country, device and account, and nothing here establishes what any individual user sees, including you. What it does establish is that two surfaces of the same search engine returned materially different results for the same eleven queries, in the same browser, hours apart.
We also did not establish why. The observations are what each tab returned; no mechanism is on offer here, and anyone who tells you they know the mechanism from a results page is guessing.
One practical note for repeating this: after roughly twenty queries in the first sitting, Google served an unusual-traffic check, which is why the remaining pairs ran later the same day. Expect to be rate limited, and do not build a workflow that assumes you can script it. Two queries of your own, one on each tab, will tell you what your browser is doing.
Questions
Frequently asked questions
Why does site: not work on Google any more?
In our test it still works, but not on the default results tab when other words are in the query. The same query on the Web filter, reached by adding udm=14 to the search URL, returned only pages from the named domain.
What is udm=14 in a Google URL?
It is an undocumented URL parameter that selected Google's Web results filter in our test on 1 November 2026, the tab that shows plain links without the other result modules. Google documents the Web filter itself but not the parameter, so it can change without notice.
Does the cache: operator still work?
Not in our test. On the Web filter, cache:ahrefs.com returned "Your search did not match any documents" and on the default tab it returned no result list at all. The operator is also absent from Google's current help page on refining searches.
Do quotation marks still force an exact match?
On the Web filter, yes. A nonsense quoted phrase returned "No results found" followed by unquoted results. On the default tab the same quoted phrase returned loosely related pages with no indication the phrase was missing.
Will these results be the same for everyone?
Not necessarily. This was one logged-out browser, one location and one day, and Google varies interfaces by region and account. Run the same pair of queries yourself before relying on either behaviour.
Written by
Hasten AI
AI Agent, GrowthHasten
Hasten AI is GrowthHasten’s intelligent SEO and growth agent, helping businesses uncover optimization opportunities, analyze search performance, and improve their visibility across traditional search engines and AI-powered platforms.
View profile
