The tests and documents behind what we know about Google, credited to the people who did the work, each graded by how much weight the evidence carries. Every claim on this site links back to an entry here.
How we grade evidence
Strong
Official documentation, sworn testimony, or a large controlled test that others have replicated.
Moderate
A credible test with a real sample and a stated method, but unreplicated or narrow in scope.
Weak
Small sample, correlation only, or a method that leaves obvious alternative explanations open.
Disputed
Credible people read this evidence differently, or later work contradicted it.
32 entries
Kevin Indig, Growth Memo
Reddit's SEO visibility fell 11.7% and its AI mentions fell 10.9% over the four weeks to 8 June 2026, the first time both dropped together since tracking began in February 2026. Over the same window X gained 40.9% in visibility.
Seer Interactive
CTR on AI Overview citations rose from 1.3% in December 2025 to 2.4% in February 2026, and traffic arriving via citations converted better than average.
Cyrus Shepard, Zyppy
URLs with 0 to 4 internal links averaged about 2 Google clicks; URLs with 40 to 44 internal links got roughly 4 times that, after which the effect reversed past about 45 to 50 links. Pages with at least one exact-match internal anchor had at least 5 times more traffic than pages without one. Anchor text variety correlated with traffic so consistently that Shepard re-ran the analysis three times.
SearchPilot (formerly Distilled ODN)
Published tests include a statistically significant +4.1% organic traffic uplift from removing embedded video carousels on product listing pages, and a mobile-only +14% where desktop showed no significant change.
Luke Harsel, Semrush
On ChatGPT, Reddit was cited in close to 60% of responses in early August, then fell to around 10% by mid-September; Wikipedia dropped from roughly 55% to under 20% over the same window. Google AI Mode was steadier, with LinkedIn in nearly 15% of responses and Wikipedia in about 2%. Perplexity leaned on Reddit, LinkedIn and NIH.
Louise Linehan and Xibeijia Guan, Ahrefs
On average only 11.9% of pages cited by AI assistants ranked in Google's top 10 for the same question, and about 80% of cited pages did not rank anywhere in Google for that query. Perplexity aligned most closely at 28.6%, while ChatGPT, Gemini and Copilot each sat near 8%.
Pew Research Center (Athena Chapekis and colleagues)
About 18% of searches produced an AI summary. When one appeared, users clicked a traditional search result on 8% of visits, versus 15% when none appeared. Just 1% of visits included a click on a source cited inside the summary. Browsing sessions ended on 26% of pages with a summary against 16% without.
Louise Linehan and Xibeijia Guan, Ahrefs
76.10% of cited pages ranked in Google's top 10 for the same query. 9.50% ranked between positions 11 and 100, and 14.40% ranked beyond position 100. The first cited URL sat at a median organic position of 2.
Brian Dean, Backlinko
The number 1 result had on average 3.8x more backlinks than the pages ranking in positions 2 to 10. Backlinko also reported that around 95% of all pages have zero backlinks.
Brian Dean, Backlinko
The average Google first-page result contained 1,447 words.
Brian Dean, Backlinko
Backlinko reported essentially zero correlation between using the exact keyword in the title tag and higher rankings. The same study also found zero correlation between site loading speed and rankings.
Ahrefs data team
Average organic CTR fell around 34.5% when an AI Overview appeared at the top of results; a later update measured a 58% drop for position 1.
Mark Williams-Cook, Candour / AlsoAsked
Google assigns site quality scores at subdomain level, sorts almost all queries into 8 semantic categories, and scores passages for agreement with consensus.
Louise Linehan, Ahrefs
99.2% of keywords that trigger an AI Overview have informational intent. AIO keywords have far lower search volume (median 500 versus 29,000) and are longer (median 4 words versus 2). 99.9% of AIO result pages also display at least one other SERP feature.
Rand Fishkin, SparkToro
The documentation references Chrome-derived clickstream data (e.g. a chrome_trans_clicks field tied to sitelink prominence) and host-level NavBoost evaluation. Fishkin argued this sits awkwardly with Google's repeated public denials that Chrome data influences rankings.
Rand Fishkin and Michael King (source: Erfan Azimi), SparkToro / iPullRank
Google stores far more than it publicly discusses, including click-based measures, site authority values, and per-document quality signals.
Mike King, iPullRank
The docs describe NavBoost storing click-based signals (goodClicks, badClicks, lastLongestClicks, unsquashedClicks) with click data normalised to limit manipulation, and a rolling click window. King read this as evidence Google captures and processes click behaviour as a ranking input.
Mike King, iPullRank
The documentation stores a compressed siteAuthority quality signal (used within a Q* ranking system) and a hostAge attribute whose stated purpose is to sandbox fresh spam at serving time. King set these against Google's public statements of 'we don't have website authority score' and 'there is no sandbox'.
Ahrefs
96.55% of the pages studied get zero traffic from Google, and 1.94% get between one and ten monthly visits. Ahrefs links most of this to a lack of backlinks and to search intent mismatch.
Ahrefs
Of roughly 20 million pages with no referring domains, only 2,997 get more than 1,000 search visits per month. Ahrefs reports a clear positive relationship between the number of websites linking to a page and its traffic.
Pandu Nayak (testimony), US Department of Justice (proceedings), Google / US DOJ
Navboost re-ranks results using aggregated click data over a rolling 13-month window, and internal documents describe it as one of Google's strongest signals.
Pandu Nayak, US DOJ v. Google
Nayak agreed RankBrain "looks at the top 20 or 30 documents and may adjust their initial score", that it is "certainly more expensive than some of our other ranking components", and that it is "too expensive to run on hundreds or thousands of results". On DeepRank he testified that "training transformer models is more expensive than training straightforward feedforward networks like Google Brain or RankBrain. Those are cheaper to train."
SearchPilot
A statistically significant 10% uplift in organic traffic, with an estimated 11,000 extra organic sessions per month if deployed across the section.
Google Search Central
The guide opens: "If your site doesn't have a large number of pages that change rapidly, or if your pages seem to be crawled the same day that they are published, you don't need to read this guide." It targets "Large sites (1 million+ unique pages) with content that changes moderately often (once a week)" and "Medium or larger sites (10,000+ unique pages) with very rapidly changing content (daily)", calling the numbers "a rough estimate" rather than exact thresholds.
Google Search Central
The documentation states: "While E-E-A-T itself isn't a specific ranking factor, using a mix of factors that can identify content with good E-E-A-T is useful."
Google Search Central
Google described "a new site-wide signal" and stated: "Any content -- not just unhelpful content -- on sites determined to have relatively high amounts of unhelpful content overall is less likely to perform well in Search, assuming there is other content elsewhere from the web that's better to display. For this reason, removing unhelpful content could help the rankings of your other content." It added: "The classifier process is entirely automated, using a machine-learning model."
SearchPilot
The shorter title tags gave an 11% uplift in organic traffic.
Cyrus Shepard, Zyppy
Google rewrote 61.6% of titles. Titles of 51 to 60 characters had the lowest rewrite rate (about 39% to 42%); titles over 70 characters were rewritten 99.9% of the time. Titles in [brackets] were rewritten 77.6% of the time versus 61.9% for (parentheses), and Google removed the bracketed text 32.9% of the time. Pipe separators were dropped 41.0% of the time versus 19.7% for dashes.
SearchPilot
Every page type lost traffic: an estimated 12% loss on product pages, 6% on subcategory pages and 7% on high-level category pages.
SearchPilot
The added internal links produced a 7% uplift in organic traffic to the affected pages.
SearchPilot
A conclusive positive result, with the dated variant titles outperforming the control by 5% in organic sessions.
Andrey Lipattsev, Google
Asked to name the top two ranking signals (RankBrain having been publicly confirmed as the third), Lipattsev said the top two are "content and links pointing to your site." This gave the running order as content, links, then RankBrain.