How Google's NavBoost Re-Ranking Actually Works

Google ranks pages in two stages, not one. Most SEOs only optimise for the first stage. The second stage is where competitors leapfrog you, and almost nobody is talking about it.

The system is called NavBoost. It has been running in production since around 2005. Google denied its existence for nearly two decades until DOJ trial testimony in 2023 and the May 2024 API leak made denial impossible.

If you understand how NavBoost re-ranking works, you understand why some sites with weaker links and thinner content somehow outrank yours. Here is exactly what it does and how to optimise for it.

The two-stage ranking architecture

Every Google search runs through two distinct ranking systems before results hit your screen.

Stage 1: Mustang generates the initial ranking

Google's primary algorithmic system, internally called Mustang, does the work most SEO guides describe.

• Matches keywords to indexed pages
• Calculates PageRank from your backlink profile
• Evaluates content quality, freshness, and topical relevance
• Scores technical signals: load speed, mobile-friendliness, schema

Mustang produces an initial ranked list, often the top 1,000 results for a query. This is the SEO most practitioners optimise for. Title tags, meta descriptions, header structure, internal links, technical fixes.

You never see this list. It does not get served to users yet.

Stage 2: NavBoost re-ranks before results are served

Before Google displays anything, NavBoost intercepts Mustang's list and asks one question.

Historically, when other users searched this exact query, which of these pages did they actually find satisfying?

NavBoost consults its 13-month rolling database of click behaviour, then re-orders the list based on what real users have been clicking, staying on, and not bouncing from.

The result you see is the re-ranked list. Mustang's original order is invisible to everyone outside Google.

This is the part Google denied for 20 years. Pandu Nayak, Google's VP of Search, confirmed under oath at the 2023 DOJ trial that NavBoost dates back to roughly 2005 and is one of Google's strongest ranking signals. The trial documents called it "perhaps the central way that web ranking has improved for 15 years."

Not a tiebreaker. Not a supplementary adjustment. Central.

The five click signals NavBoost uses

The May 2024 API leak revealed the specific metrics NavBoost tracks for every query-page combination.

The internal module that processes these signals has a name. Google calls it Craps.

In 2019, Gary Illyes (Google Search Analyst) told the SEO community that click theories were "generally made up crap." The internal module processing those exact signals was, at that moment, named Craps. Whether deliberate joke or coincidence, it is now in the public record.

A worked example: emergency plumber Melbourne

Let me show you how this plays out in practice.

Mustang's initial ranking for "emergency plumber Melbourne" might look like this:

Now NavBoost checks 13 months of clicks for this exact query in Melbourne, on mobile, in English.

What is happening: Bigplumbingco gets the click because their title is great. But users land on the page, see a popup, can't find the phone number, and bounce back to Google to try another result. Every one of those is a badClick going into NavBoost's 13-month memory.

Acme delivers. Users land, see the phone number above the fold, call it, and the search session ends. That is a textbook lastLongestClick.

NavBoost re-ranks the list before serving it:

That is the list Melbourne mobile users actually see. Bigplumbingco might still be "objectively" the best-optimised site by traditional SEO standards. They lost anyway.

The COEC model: clicks over expected clicks

NavBoost does not just count raw clicks. It compares your actual click count to the expected click count for your ranking position.

Position 1 normally captures around 30% of clicks for a query. Position 4 captures around 7%. Position 10 captures around 2%.

Now imagine your page ranks at position 4 but is getting 15% of clicks. That is more than double the expected rate. NavBoost reads this as users actively choosing your result over the three results above it. Strong positive signal. Boost incoming.

Conversely, if your page ranks at position 1 but is only getting 18% of clicks (below the 30% expected), NavBoost reads this as users scrolling past you to find something better. Negative signal. Demotion incoming.

This is why title tag optimisation and meta descriptions have real ranking consequences, not just CTR consequences. And it is exactly why clickbait backfires. You might win the click-through rate, but if those clicks are badClicks, you are feeding NavBoost a negative signal that compounds for 13 months.

Glue: NavBoost's companion for non-blue-link results

NavBoost is not just for the standard blue link results. It has a companion system called Glue that processes click signals for everything else on the SERP.

• People Also Ask boxes
• Image carousels
• Video results
• Knowledge Panels
• Featured snippets

Per Pandu Nayak's trial testimony: "Glue is just another name for NavBoost that includes all of the other features on the page."

The implication for you: clicks on your PAA appearance, your YouTube video carousel slot, or your image search result feed the same ranking machinery as your standard organic listing. Optimise for cross-format visibility, not just blue links.

How NavBoost slices the data

NavBoost does not use one global click database. It maintains separate slices for different contexts.

• Country: clicks from Australia don't help your US rankings
• Device: mobile clicks and desktop clicks are scored separately
• Language: English-language clicks don't carry over to French results
• Query type: clicks for "how to" queries are weighted differently from transactional queries
• User attributes: signals are calibrated by user type

This is why ranking #1 in one market does not automatically translate to ranking #1 globally. NavBoost is contextually calibrated. A UK mobile user gets re-ranked results based on other UK mobile users. Not your AU desktop traffic.

It is also why testing with VPNs gives you a false read. The signals NavBoost shows you depend on the slice.

Why Google denied this for 20 years

The patent (US8595225B1) was filed in 2004 by Amit Singhal and Urs Hoelzle, two of the most senior engineers in Google's history. The system went live around 2005. Then for nearly two decades, Google's public spokespeople said the opposite.

Two plausible reasons for the denials.

The charitable explanation: click signals are genuinely manipulable, and Google did not want to incentivise large-scale click fraud. The 13-month rolling window provides stability, but if SEOs knew the system existed and knew exactly how it worked, the manipulation arms race would get expensive fast.

The cynical explanation: the DOJ antitrust case was partly built on the argument that Google's accumulated user click data is its insurmountable competitive moat. Admitting that click data is "perhaps the central way web ranking has improved for 15 years" would have strengthened every antitrust argument made against Google for two decades.

Both explanations can be true at the same time.

What to actually do about NavBoost

Theory is useless without action. Here are the four plays that move click signals in your favour.

1. Above the fold determines whether you get badClicks or goodClicks

The badClicks signal is generated in the first few seconds. If your page takes 3 seconds to load, displays a cookie wall, or makes the user scroll past hero imagery before reaching the answer they searched for, you are generating badClicks at scale. Every one is a ranking vote against you, accumulated for 13 months.

What to do: load the page on mobile (375px width) and screenshot it before scrolling. Check that the H1, the first sentence answer, and a relevant visual element are all visible. No popups. No cookie banners blocking content. No "hero" section that pushes the actual answer below the fold.

2. Earn the lastLongestClick by answering the next question too

The page that answers the question, then answers the obvious follow-up question, then provides the context the user needs to take action, will earn the last click in more sessions than a page that answers partially.

This is the structural logic behind comprehensive content outperforming thin pages on competitive queries. Not because Google rewards word count. Because comprehensive pages earn more lastLongestClicks. Users get the full answer and end their search session on your page.

What to do: after answering your primary query, ask "what would they ask next?" and answer that on the same page. Then "what would they need to take action?" and provide that too.

3. Brand search is the cleanest possible click signal

When a user searches for your name directly and clicks your result, the entire session starts and ends with you. These are pristine lastLongestClicks that are essentially impossible to fake or misattribute.

This connects to E-E-A-T and Google's Brand Search Ratio. Building brand recognition that drives direct name searches is, in a very literal sense, generating the cleanest possible NavBoost signals.

What to do: invest in digital PR, podcasts, speaking, social presence. Anything that drives people to search your brand by name. Branded search volume is the most defensible NavBoost signal you can build.

4. Internal linking from your highest-engagement pages compounds the signal

The click distance patent (US8082246B2) establishes that proximity to highly-clicked, trusted pages passes authority via internal links. Audit which pages on your site have the highest engagement time in GA4. Then internal-link FROM those pages TO the pages you want to rank.

Most SEOs do this backwards. They link from low-engagement pages to high-engagement pages thinking that flows authority. The patent suggests the opposite is more powerful.

What to do: in GA4, sort organic landing pages by average engagement time descending. Take the top 10. Add internal links from those pages to the lower-traffic pages you are trying to push.

The bottom line

Mustang gets you onto the shortlist. NavBoost decides where you land.

You can have perfect on-page SEO, a strong backlink profile, and clean technical fundamentals, and still lose to a competitor with weaker traditional signals if their above-the-fold UX is better and they earn more lastLongestClicks for the query.

The 13-month memory cuts both ways. A bad UX update from 6 months ago is still demoting you today. A great UX win from a year ago is still paying ranking dividends.

Most SEOs spend 100% of their time on Stage 1 optimisation. The competitive edge in 2026 is treating Stage 2 as equally important. Build for the click, then build for the satisfied last click. That is how you beat sites with deeper pockets and stronger link profiles.

FAQ

What is NavBoost in Google's ranking system?

NavBoost is Google's click-signal re-ranking system that runs after the primary Mustang algorithm generates an initial ranked list. It uses 13 months of accumulated user click behaviour (goodClicks, badClicks, lastLongestClicks) to re-order results before they are served to users. Google's VP of Search confirmed under oath at the 2023 DOJ trial that NavBoost is "one of Google's strongest ranking signals" and has run since approximately 2005.

Does Google use clicks as a ranking signal?

Yes, definitively. Confirmed by a 2004 Google patent (US8595225B1), sworn DOJ trial testimony in 2023, and 14,000 pages of internal API documentation leaked in May 2024. The internal module that processes click signals is literally named Craps. For 20 years Google publicly denied this, but the production reality was always different from the public statements.

What is the difference between goodClicks, badClicks, and lastLongestClicks?

goodClicks are extended user engagements where someone stays on your page and reads. badClicks are short clicks where the user immediately returns to the SERP, indicating the page didn't satisfy them. lastLongestClicks are the strongest positive signal: the final extended interaction before the user ends their search session entirely. Earning lastLongestClicks for a query means Google interprets your page as the complete answer.

How long does NavBoost remember click data?

NavBoost maintains a 13-month rolling window of click behaviour. This means a bad UX update can suppress your rankings for 13 months after you fix it, and a strong UX win continues paying dividends for 13 months. The long memory makes manipulation expensive and makes legitimate UX investments highly defensive.

Can I optimise for NavBoost without optimising for traditional SEO?

No. NavBoost only re-ranks pages that Mustang has already shortlisted. If your traditional SEO fundamentals (technical SEO, content quality, links, on-page) don't get you into the initial ranking pool, NavBoost never sees your page. Mustang gets you onto the shortlist. NavBoost decides where you land. You need both.