Google uses click data in Search. That does not make CTR a shortcut

The Google document in the video is real. The shortcut built on top of it is not. Google uses click and query data in ranking systems. That does not turn raw click-through rate into a button an SEO can press. We checked the court record, Rand Fishkin's original experiment, Google's current title guidance, and 39 titles on this site before changing anything.

The document is real, with two missing details

The slide deck is called "Ranking for Research." The US Department of Justice published it as trial exhibit UPX0204 in the Google search antitrust case.

The first slide says access was restricted to specific people in Research who were actively improving Search. It was not simply a document for every Google employee. The public copy is also abridged and redacted.

The lines shown in the video are genuine. Google warned that attempts to manipulate results were continuous and well funded. It told staff to keep details about Search on a need-to-know basis. A later slide named SEOs, patent trolls, and competitors as likely users of leaked information.

One slide is more awkward for Google. It told staff not to discuss clicks outside a need-to-know group, then said Google's public position was debatable. That is evidence of a gap between the public message and the internal discussion. It is not a ranking recipe.

What the court confirmed about clicks

Judge Amit Mehta's 2024 opinion describes Navboost as a signal that pairs queries with documents by remembering user click data. Google trained it on about 13 months of user data.

The opinion also says Google uses click and query data to re-rank results, improve freshness, and train models that improve relevance. That settles the broad question. Click behavior has informed ranking.

It does not settle the narrow claim in the video: give one page more clicks and its rank will rise. Navboost works across old, aggregated query and interaction data. A Search result has a position, a query, a location, a device, and competing results. A raw CTR number drops most of that context.

The famous 228-click test had a warning label

Rand Fishkin ran the test in April 2014. He asked his Twitter followers to search for "IMEC Lab" and click his result. The page received 228 Google visits that day. It moved from position seven to position one in US results within about three hours.

Fishkin's own write-up says the test did not prove that click volume caused the move. The tweet could have produced links or other signals. The boost could have been temporary. Results outside the US stayed lower. The query was rare enough that a sudden burst of interest could also look like freshness.

The experiment justified more testing. It was not a controlled test of modern Google Search, and it never claimed to be one.

Better titles still matter

A title has a simpler job. It tells a searcher what the page contains and why it may answer the query.

Google's current title guidance asks for descriptive, concise, distinct titles. Google can replace the title link using the page heading, visible text, and links from other pages. A clever formula cannot force the final wording.

This site already keeps two title fields:

title: "How we made AI agents write in simplified technical English",
seoTitle: "ASD-STE100 for Claude: Clearer AI Agent Writing",

The first one is the article heading. The second one is the search and social title. One can sound natural while the other names the query clearly. Both must describe the same page.

We checked the data before rewriting 39 titles

Our title check confirmed that all 39 blog posts have a separate search title and that every search title stays below 70 characters. The words alone did not show where the real problems were, so we pulled the latest 28-day Google Search Console report, from August 6 to September 2, 2026.

The simplified-English post had 1,920 impressions, 32 clicks, and an average position of 7.6. Its most visible named query was "asd-ste100 claude." We changed the search title to put Claude beside the standard.

The Common Crawl post had 1,416 impressions and no clicks at position 9.0. Almost all visible impressions came from searches for the exact CC-MAIN-2026-30 index endpoint. Those searchers appear to want the endpoint, not an article about the dataset. A more dramatic title would not fix that intent mismatch.

The Notion block-limit post had 731 impressions and a 0.7 percent overall CTR. The only visible high-volume query in our page-level sample had a 20 percent CTR. Search Console hides some low-volume queries for privacy, so the aggregate cannot tell us which hidden wording to target.

The LocalSend review had 401 impressions, no clicks, and an average position of 13.3. Its search title already says "LocalSend Review" and "Open-Source AirDrop Alternative." Position is the clearer constraint there.

We changed four titles. Two article headings did not name their subjects early enough. Two search titles had clear evidence for tighter wording. We left the other 35 alone.

What we will measure next

We will compare the changed pages after another complete 28-day window. The comparison will stay at page and query level. We will also check position, because CTR normally changes when position changes.

We will not buy clicks, coordinate searches, or treat a short spike as proof. The useful lesson from Google's internal slides is that user behavior matters at scale. The useful lesson for a site owner is smaller: make the result clear, keep the promise on the page, and measure the right query.