X-Ray Search in 2026: What Google Killed, What Still Works
Google killed num=100, AI Overviews ate the SERP, and LinkedIn de-indexed snippets. Here's what x-ray search actually looks like in 2026.
If you learned to source on site:linkedin.com/in "software engineer" "Python", the last 24 months have not been kind. LinkedIn stripped its public snippets, Google quietly killed num=100, and AI Overviews now sit on top of every SERP eating the results you actually want. The good news: x-ray search isn't dead. The honest news: most of what recruiters call "x-ray" in 2026 is a different toolkit aimed at different platforms.
Here is what still works, what is permanently broken, and where Boolean is now ceremonial rather than productive.
The three changes that broke 2019-era x-ray
Three discrete events, between January 2024 and September 2025, gutted the classic recruiter playbook. If you haven't internalized them, your search strings are silently underperforming.
LinkedIn de-indexed itself in January 2024
In January 2024, LinkedIn removed Headline, About, Experience, Education, and Location data from Google's public index. A single change broke the most popular site:linkedin.com/in use case worldwide. Before that, a query like site:linkedin.com/in "staff engineer" "Rust" "Berlin" returned snippets that exposed titles, skills, and locations directly in the SERP. You could skim 30 results, copy the names, and move on.
After January 2024, those snippets are mostly empty. You get a name and a URL. Sourcing trainer Irina Shamaeva flagged it on the Boolean Strings blog within weeks; Jan Tegze's Full Stack Recruiter newsletter documented the stripping in detail. The community noticed. A lot of tooling did not, and silently degraded.
Google killed num=100 in September 2025
Between September 12 and 14, 2025, Google silently disabled the num URL parameter. Append &num=100 to a search and you still get 10 results. No announcement, no migration path. JavaScript rendering became mandatory at the same time, which broke the simpler scraper stacks that recruiters were quietly running behind Chrome extensions.
This matters more than it sounds. Half the "scan 100 candidates a click" tools on the market in 2025 were thin wrappers over num=100. If you bought one of those before September 2025 and never noticed your hit rate dropped, you were paying for a tool that was returning a tenth of its previous yield.
AI Overviews now eat the top of the SERP
AI Overviews rolled out to 200 countries during 2025. They sit above organic results, summarizing instead of linking. For sourcing, that's pure friction: the answer you want is "show me the 10 GitHub profiles", not "here's a paragraph explaining what a Rust developer is".
There is exactly one trick worth memorizing here: append &udm=14 to your Google search URL. It forces classic Web mode, strips AI Overviews, and gives you back the SERP you used to have. If you teach one new keyboard shortcut to your sourcing team in 2026, make it this one.
What's deprecated, what's left
Google has been quietly retiring operators for years. The current 2026 inventory is much shorter than the cheat sheets you'll find on most recruiting blogs.
Deprecated or dead:
cache:(removed September 2024)related:(removed July 2023)link:,info:,phonebook:,allinanchor:,id:&num=URL parameter (September 2025)
Still works and worth using:
site:intitle:andinurl:filetype:- Quoted phrases
"..." - Negation with
- OR(uppercase) and grouping with()AROUND(n)for proximity
That's the whole list. Six operators, one URL parameter trick, and your quotes. Anything more elaborate is either deprecated or wasn't doing what you thought it was doing.
The new x-ray map: rotate off LinkedIn
The single biggest mental shift in 2026 is treating LinkedIn as the worst site: target rather than the default. Public LinkedIn pages are now name-and-URL stubs. The platforms where x-ray still produces dense, keyword-rich snippets are the ones LinkedIn doesn't control:
- GitHub profiles, READMEs, gists, and pinned repos
- Stack Overflow profile pages, answer pages, and tagged questions
- Behance and Dribbble for design work
- Personal portfolio sites and university faculty directories
- Conference speaker pages (KubeCon, RustConf, PyCon, etc.)
- arXiv author pages and lab pages for ML research
A query like site:github.com "Rust" "tokio" "Berlin" -intitle:"sign in" still returns rich, scannable snippets. Same with site:stackoverflow.com/users "TypeScript" "Toronto". These platforms haven't de-indexed because their growth model depends on Google sending traffic in.
LinkedIn x-ray gives you a name and a URL. GitHub x-ray still gives you a person.
filetype:pdf is the most underused operator in recruiting
The honest sleeper play in 2026 is filetype:pdf (and filetype:doc, filetype:docx). It surfaces actual resume and CV files candidates have uploaded to personal sites, university directories, conference proceedings, and portfolio platforms. ERE Media flagged this years ago and the technique has only gotten more valuable as LinkedIn snippets dried up.
Try this:
filetype:pdf "resume" OR "curriculum vitae" "machine learning" "PyTorch" "San Francisco" -job -hiring -template
You get artifacts no platform's internal search will surface: actual CVs, talk decks, conference papers with author affiliations, even thesis PDFs from candidates who are about to graduate. These are immune to the LinkedIn de-indexing wave because they live on the open web.
The Boolean ceremony problem
Here is the part most sourcing blogs won't say out loud: Boolean is now a ceremonial skill. RecruitEm (recruitin.net) is still a perfectly fine free Boolean-string builder, and it's where most sourcers start. But the productivity ceiling on hand-crafted strings is low.
A real example. The string:
site:github.com ("software engineer" OR "senior software engineer" OR "staff software engineer") "Python" (San Francisco OR "SF Bay Area" OR "South Bay")
gets you a slice. But you're manually expanding titles, manually expanding location synonyms, and manually filtering out company recruiter pages. You'll miss "ML engineer" who codes Python all day. You'll miss "platform engineer" who happens to live in Oakland. You'll spend 40 minutes building a string that returns 60 hits, half of which are noise.
Run the same intent against an indexed people-search layer and the picture is different. A query for "Software/Senior/Staff Software Engineer with Python in the United States" returns 74,268 matching profiles in Refolk's index, with top employers including Databricks, LinkedIn, GitHub, and Glean, and the heaviest concentrations in San Francisco and New York. That's not a Boolean string winning. That's an index doing synonym expansion, title normalization, and location handling for you.
This is why we built Refolk the way we did: you describe the person in plain English ("senior Rust engineers in Berlin who've contributed to tokio or actix") and get a ranked shortlist across GitHub, LinkedIn, and the open web. The Boolean is implicit. The synonym expansion is implicit. The result is a list of people, not a SERP you have to clean up.
A practical 2026 x-ray workflow
Here's the stack that actually works right now, in order of when to reach for it.
1. Start with intent, not syntax
Before you open a Google tab, write the description. "Founding backend engineer, comfortable owning infra, has shipped a system at 100k+ QPS, probably in their second or third startup." The string comes from the description. Sourcers who skip this step end up writing strings that match titles instead of people.
2. Use Refolk for LinkedIn-shaped queries
LinkedIn x-ray is broken. Don't fight it. For role-and-location-and-skill queries (the classic "Senior Python engineers in Austin" shape), natural-language people search beats site:linkedin.com/in because LinkedIn snippets are gutted and the index has the data Google can't see anyway. Use Refolk here, save Google for the open-web tail.
3. Use Google x-ray for GitHub, Stack Overflow, and artifacts
This is where Google still earns its keep. site:github.com and site:stackoverflow.com queries with intitle:, quotes, and AROUND(n) produce real signal. Append &udm=14 to kill AI Overviews. Don't bother with &num=100, it does nothing.
A useful pattern for engineering sourcing:
site:github.com inurl:readme "Rust" AROUND(5) "production" -tutorial -example
That finds READMEs claiming production Rust use, which is a much stronger signal than "Rust" appearing on a profile.
4. Run a filetype: sweep last
After your platform x-rays, run a filetype:pdf pass for resumes and a filetype:pdf "talk" OR "slides" pass for conference decks. You'll find candidates whose LinkedIn says "Software Engineer at Stealth" but whose RustConf 2025 deck reveals exactly what they built.
5. Skip the Chrome extensions that promised num=100
If a sourcing tool's pitch deck from 2024 mentioned scanning 100 results per click, check whether it still works. Most don't, post September 2025. Google's broader social indexing shift, where LinkedIn, Reddit, TikTok, and Instagram posts now appear beside websites, has also changed what scrapers can reliably target. Treat anything Chrome-extension-based as suspect until you've watched it return real results in a clean profile.
What this means for boolean recruiting hires
If you're hiring sourcers in 2026, "expert in Boolean" is no longer a strong signal. Everyone competent has the operators memorized; the operators that matter fit on an index card. The actual signal is: do they understand intent translation? Can they read a hiring manager's vague description and produce a list of 30 real, reachable people in an afternoon?
The fastest path to that outcome combines a small amount of x-ray (GitHub, Stack Overflow, filetype:) with a natural-language people search layer that does the LinkedIn-shaped work. Refolk's pitch ("Find anyone. Just ask.") is that combination, but the broader point holds even if you build it yourself: stop optimizing the Boolean string, start optimizing the description of the person.
X-ray search isn't dead. It's just sharper, narrower, and pointed at different platforms than it was three years ago. Adapt the toolkit and it still earns its keep.
FAQ
Is x-ray search on LinkedIn completely dead in 2026?
Not completely, but close enough that you should reset expectations. Since January 2024, LinkedIn has stripped Headline, About, Experience, Education, and Location from public snippets. You can still find profile URLs with site:linkedin.com/in "name fragment", but the keyword-rich snippets that made LinkedIn x-ray productive are gone. For role-and-skill queries, an indexed people-search tool will outperform Google x-ray on LinkedIn by a wide margin.
What is the udm=14 trick and why does it matter?
Appending &udm=14 to a Google search URL forces classic Web mode and strips AI Overviews from the results page. Since AI Overviews rolled out to 200 countries in 2025, they sit at the top of most SERPs and push organic results below the fold. For sourcing, where you want a list of profiles rather than a summary, udm=14 is the single most valuable URL trick to learn in 2026.
Which Google operators still work for recruiting?
The short list: site:, intitle:, inurl:, filetype:, quoted phrases, negation with -, OR (uppercase), grouping with parentheses, and AROUND(n) for proximity. Deprecated operators include cache:, related:, link:, info:, phonebook:, allinanchor:, and id:. The &num= URL parameter was silently disabled in September 2025 and now does nothing.
Why is filetype:pdf underused for sourcing?
Because most recruiters were trained to point site: at LinkedIn and stop there. filetype:pdf surfaces actual artifacts (resumes, CVs, conference decks, theses, faculty pages) that no platform's internal search indexes. These files are immune to the LinkedIn de-indexing wave and often reveal what a candidate actually built, not just their job title. Combine filetype:pdf with skill keywords and a location, exclude generic terms like "template" and "sample", and you'll find candidates competitors haven't seen.