One Big Keyboard
Google search operators, tested on 14 November 2026
Lists of Google search operators are everywhere and they disagree with each other. One says twelve of forty-one still work, another says twenty-four, another about twenty-five. Several still list operators that have been dead for years.
The disagreement has a cause. Google documents almost none of them, so every list is someone's guess about undocumented behaviour that Google never promised to keep. This page does two things instead: it says exactly what Google documents, and it reports what happened when we ran the rest.
Google documents only four operators: filetype:, imagesize:, site: and src:. Of the rest we tested, site:, filetype:, intitle:, quoted phrases, minus and before:/after: worked. allinurl:, allintitle:, related:, link: and cache: did not, and intext: returned zero results where the same search without it returned five.
Google documents four operators. That is the whole list
Google Search Central has a search operators section. It covers four: filetype:, imagesize:, site: and src:.
That is worth sitting with. Every other operator in every list you have read, site-adjacent favourites like intitle: and inurl: included, is undocumented. Google has never committed to them and owes nobody notice before changing them.
One consequence is that imagesize: is officially supported, which most guides either omit or list as obscure. Another is that if you are building anything that depends on an operator, those four are the only ones with a published contract behind them.
How this was tested, and the limits of it
Queries were run against google.com on 14 November 2026, signed out, United States locale, desktop, in a browser. Each result set was read directly rather than inferred.
The test is honest but not exhaustive. Google personalises and regionalises results, so a query that behaves one way here may behave differently for you. A single query failing is weaker evidence than a controlled pair, so where it mattered we ran the same search with and without the operator and compared.
Operators we did not reach are listed as untested at the end rather than guessed at. A shorter true list is worth more than a long one with filler.
The ones that worked
Six operators did what they claim, and three of them came with something worth knowing. Google's own page warns that the list site: returns is not always exhaustive, and that site:https://www.example.com and site:https://example.com/ are not the same query, so treat it as a filter rather than an index dump.
The exact phrase test included the useful negative case: a phrase with no matches returned none, and Google offered the unquoted version instead rather than silently widening the search. The date bounds surprised us. A query bounded to the first half of October 2026 returned results dated the fifth to the fourteenth of that month, with nothing outside the window.
- site: narrowed to the domain, every time.
- filetype:pdf returned only PDFs.
- intitle: every returned title contained the term.
- "exact phrase" worked, including returning nothing when nothing matched.
- -exclusion worked: squirrel -rescue -rehab returned eight results with neither word.
- before: and after: bounded results to the exact fortnight asked for.
The ones that were quietly ignored
Six more were accepted and then quietly ignored. inurl: was the one to watch, because inconsistent is worse than broken: it held for one term and was ignored for another in the same session, on the same site, and a search for inurl:ipad-keyboards on apple.com returned a sitemap, a privacy page and an unrelated App Store listing.
- allinurl: ignored. Returned sites whose URLs contained neither word.
- allintitle: ignored. Returned titles containing two of three words.
- inurl: inconsistent, which is worse. Held for one term, ignored for another in the same session.
- related: returned the target's own pages, not related sites.
- link: returned the target's own pages, like a plain keyword search.
- cache: returned nothing, matching Google's statement that the feature is gone.
The one that failed in an interesting way
intext: is the clearest result of the whole exercise, because we tested it against a control.
A search for the exact phrase arctic ground squirrel restricted to bbc.co.uk returned five results, including a BBC Future article and two programme pages.
The identical search with intext: in front of the phrase returned zero results. Not fewer, not different, zero, with Google's no results message.
So intext: did not filter the set. It emptied it. An operator that silently returns nothing is more dangerous than one that is ignored, because an empty page reads as there is nothing out there rather than your query was rejected.
Returned results, effect unconfirmed
Two operators returned plausible results that we could not verify from the result page alone, and they are reported as such rather than counted either way.
AROUND(2) returned seven results, but confirming proximity means reading each page, which we did not do.
imagesize:1920x1080 returned a full page of image results. Confirming that each image is really that size means checking each file, which we did not do. This one is documented by Google, which is a reason to trust it more than the undocumented ones.
Not tested
OR, the wildcard asterisk, define:, inanchor:, source:, src: and the two-dot numeric range were not tested here, so this page says nothing about them.
If you want to check one yourself, the method is the only part worth copying: run the query with the operator, run it without, and compare the two result sets. One query on its own tells you very little, because you cannot tell a filtered set from a thin one.
Questions
Which search operators does Google actually document?
Four: filetype:, imagesize:, site: and src:, in the search operators section of Google Search Central. Everything else is undocumented behaviour, which is why published lists disagree with each other.
Does intext: still work?
Not in our test, and it failed in the worst way. The exact phrase restricted to one site returned five results. The same search with intext: in front returned zero. It emptied the result set rather than filtering it.
Does cache: still work?
No. It returned no results, which matches Google's own statement that the cached link feature no longer exists.
Do before: and after: work?
Yes. A query bounded to 1 to 15 October 2026 returned results dated between the fifth and the fourteenth, with nothing outside the window.
Why do the operator lists online disagree so much?
Because almost every operator is undocumented, so each list is somebody's guess about behaviour Google never promised to keep. Lists also age badly: operators get quietly dropped, and nothing announces it.
How should I check an operator myself?
Run the query with the operator and again without it, and compare the result sets. A single query tells you almost nothing, because a filtered set and a thin set look the same.
Tested on 14 November 2026 against google.com, signed out, United States locale, desktop. Google personalises and regionalises results, so your results may differ. Operators listed as not tested were not run and nothing is claimed about them. The documented list comes from Google Search Central's search operators section, and the site: caveats from Google's own site: reference page. Operator behaviour is undocumented and changes without notice, so treat the date as part of the finding.
One Big Keyboard puts these operators one tap away on iPhone and iPad