Blue links seem to fall, lower and lower on the page. Paid spots probably crowd the summit. AI Overviews now fill the next space in line. Old search furniture like FAQ blocks and People Also Ask panels appear even further below. Someone who types a search in 2026 might feel lost if they traveled here from 2015.
Clicks may scatter differently now. The way Google chooses winners has likely stayed much the same. Where your page falls in organic order probably still decides if Google picks you for the new AI summary at the peak. Organic place also seems to choose who gets the few clicks left below. Google has called this change AI helping people find better results. The phrase might fit. The engine that sorts what shows up comes from decades of Google’s fine-tuning.
So you may want to focus even harder on how Google ranks pages. Each visible slot feels tighter than ever. Every place feels more precious. Mistakes from wild guessing may now hurt more. Publishers who know the moving parts can probably play the game better. Those who stick with old slogans may fall behind.
One name caused most of the confusion
RankBrain went live at Google in spring, 2015. For months, nobody outside the team seemed to notice. News broke late in October, through a Bloomberg scoop. A Google engineer, Greg Corrado, spoke to reporters.
Corrado called RankBrain the third most important signal. Content and links stood above it. These brief words set off years of confusion. Suddenly, marketers pictured RankBrain as something anyone could flip on or off. Trade writers picked up the term. The word RankBrain soon became another way to say all of Google search.
Actual practice was smaller. On day one, RankBrain answered only search puzzles Google never solved before. This made up maybe a slice of daily searches, based on reports from that time. Later, Google allowed RankBrain to look at every search. That part never changed — machine learning tools always stayed bolted onto a core shaped by real people in Google’s labs.
Best to remember that. RankBrain was never the whole recipe. The label stuck just because Google named that tool loudly, first and boldest.
Afterwards, Google grew quiet. Neural matching slid in around 2018, with barely any noise. The next year, BERT arrived to help Google read language better. MUM came in 2021. Google says clearly: MUM does not shape basic ranking.
What the court papers actually named
Two big events pulled Google’s curtains back. One came with the antitrust trial from the United States Department of Justice, which gave the world sworn comments and shared trial papers between 2023 and 2025. The other came from a leaked dump of Content Warehouse API files during May, 2024.
These two doors into Google are not equal. Testimony given under oath stands as real evidence. A batch of leaked files can help, but feels weaker. Many names for Google tools you see in SEO blogs likely come from leaks, not the courtroom. Writers rarely tell you which is which. Below, each part lists its source — if known.
Navboost
Navboost changes the order of results by looking at where real people clicked before. The search experts say this tool has probably worked inside Google for many years. Pandu Nayak, who leads the search team at Google, said clearly during a trial that Navboost stands among the most powerful tools for deciding Google results. Nobody seems sure how far into the past the click records go. Some reports mention a time frame over a year, but the details are still debated. confirmed under oath
Web creators face a serious truth. Google probably pays close attention to the moves that searchers make on your page.
Glue, and its relatives Super Glue and Instant Glue
Glue manages everything besides the classic blue links. The system gathers information from the ways people touch search tools, then helps pick which tools should even show up. All those carousels, info panels, and special boxes on the results page fall under Glue’s control.
QBST, or Query-Based Salient Terms
QBST acts like a memory bank, built from mountains of actions by searchers over the years. The focus stays tight. QBST learns the exact terms that must show up before Google thinks your page really handles the topic. Miss those important pieces, and your hard work might never seem right to the search engine.
Term Weighting
Term Weighting looks for the most important words in any query. Not every word helps. Some words matter far more. Term Weighting works to spot the words that truly guide the result.
RankEmbed and RankEmbedBERT
Both of these systems turn sentences into meaning-rich numbers, then use those to find and sort pages. Papers that came out in the court battle said the training mixed in both click numbers and search questions. Years ago, the matching rules stopped needing an exact word-for-word match. describe training that drew partly on click and query data
DeepRank
DeepRank grew from BERT technology. Language meaning stands at the center of DeepRank. The system tries to pull out the real message of a search question, finds the true message in a web page, then puts the answers in a strong order using both sides of the story.
Twiddlers
Twiddlers probably work by changing the results after the main order sets in. A twiddler might move one answer up, pull another down, right after the core system judges. Some mystery hangs over this word. The experts who studied this story cannot agree if the name “twiddler” first came from court papers, from a software leak, or from both, so readers should stay careful with this idea.
The hand-built core
Here sits a surprise for everyone planning online work. A Google engineer named HJ Kim said under oath that most signals come from people, not machines. The point asks for attention. Machines help Google sort results, but human-made rules still control most of the process. Under the language of AI, the old expert logic keeps driving results behind the curtain. the vast majority of signals are hand-crafted
Many believe AI alone now chooses what ranks high. Words from the trial suggest another story together.
Chrome as a sensor
Case records showed behavior inside the Chrome internet tool can shape Google’s understanding of what people like. Chrome serves Google in more ways than simply showing websites. This browser probably acts as a measuring device.
A stack, not an algorithm
Stack the pieces together and a pattern comes alive. No single engine chases just one purpose. Different groups, trained with their own methods and launching on their own clocks, send tasks along a long road. Each section of that road belongs to a different squad. Together, they build something much bigger than one mind can dream up.
The movement usually shakes out in a familiar way. First, Google stares at the question. Next, possible texts pour out from a big set of saved material. Human hands built most of the signals that get checked when Google scores each choice. Then, behavior waves in, and something called Navboost probably changes lots of the order. The last moments bring together collected pieces, while a tool named Glue may pick what you actually see.
Outcomes appear, and both create direct challenges for content creators.
- No single ranking factor exists to optimize. A change inside one component may move your traffic for reasons that make no sense at the output. Cause and effect stay muddy by design.
- Crawling and indexing run as a separate pipeline from ranking, with separate storage and separate ways to fail.
Ranking sorts pages that indexing already accepted
This is really key. SERP has two very different components. First – your link needs to be accepted. Only when it is accepted will it be ranked. And the engines that handle these two processes are completely different.
Great writing, hidden behind closed gates, may get no reward at all.
What the stack means for your SEO work
Real guidance can spring from the whole setup—though the ideas will probably not match the usual checklist style.
- Cover the words your topic genuinely requires. QBST rewards documents that read like real coverage of a subject, not documents that dodge the obvious vocabulary.
- Write titles and descriptions that earn a click honestly. Navboost watches behavior, so a listing nobody chooses probably fades.
- Stop optimizing for named systems. Nobody outside Google can tune for DeepRank, and the components shift on schedules Google never publishes.
- Check whether Google indexed the page before you rewrite the page. A great deal of wasted effort goes into improving pages that never entered the index at all.
Rules about fair play might change some of the pieces again, so any single look at Google search might feel out of date soon. The wider message stands tall even when details move. Google acts as a machine built from many sides. Human-made rules still control almost everything. Earning a spot and earning a good place count as two separate struggles.
A later story will pull apart the way pages join the main set, and may show why good writing sometimes never makes it to the Google crowd at all.