Alex Rostovtsev is an SEO at Elfsight and an AI visibility specialist at Beamtrace. Alex combines hands-on SEO practice with original research into how AI systems evaluate and surface content. In this article, he breaks down Google’s NavBoost system – what it measures, why it matters, and what it means for anyone trying to rank in the age of AI Overviews.
In 2023, a federal antitrust trial forced Google to publicly acknowledge a re-ranking system that had been quietly reshaping search results for nearly two decades. A May 2024 API leak then confirmed the data structures behind it. Here’s what those two events revealed – and why it matters for anyone creating content in the AI search era.
For years, the SEO playbook was relatively simple: if someone clicks your result, that’s about it – mission accomplished. Turns out Google has been dismantling that logic for at least two decades – but quietly, and without a playbook update. It took a federal antitrust trial, an accidental API leak, and a series of unusually candid interviews from Google’s Head of Search to make the mechanics visible. Here’s what we now know: Google doesn’t just count your clicks – it grades them.
The System Google Didn’t Want to Talk About
In 2023, Google’s then-VP of Search Pandu Nayak took the stand in the DOJ antitrust trial and confirmed the existence of NavBoost – a re-ranking system that adjusts search positions based on how users behave after they click.
NavBoost has been running since at least 2005. It processes a 13-month rolling window of click behavior data and tracks five distinct signal types:
- goodClicks — user stayed on the page (positive)
- badClicks — user returned to search results quickly (negative)
- lastLongestClicks — the final click in a session with the longest dwell time (strong positive)
- squashedClicks — normalized click counts, smoothed to prevent manipulation
- unsquashedClicks — raw counts before normalization
The DOJ’s argument was telling: they claimed the click behavior data is Google’s real competitive advantage – the thing no competitor can replicate, because no one else has two decades of user engagement data at that scale.
A year later, the May 2024 API leak confirmed the data structures. The internal documentation showed these exact click categories as fields in Google’s ranking systems. NavBoost wasn’t a theory anymore, it was now shown as code.
What “Quality Clicks” Actually Means
In August 2025, Liz Reid – Google’s VP and Head of Search – published a blog post with a title that deserves close reading: “AI in Search is driving more queries and higher quality clicks.”
The phrasing is deliberate: not “more clicks” – “higher quality” clicks. Google defines these as clicks where users don’t quickly click back; what Reid calls “bounce clicks” in a separate interview.
In that interview, she was more specific: Google has “expanded beyond this concept of spam to sort of low-value content – this content that doesn’t add very much, kind of tells you what everybody else knows.”
That last point is worth thinking about. Google isn’t just fighting spam anymore. They’ve created a category below spam – content that’s technically fine but adds nothing; content that exists to rank, not to be useful. And they’re measuring this through post-click behavior at scale.
The Math Behind the Grading
NavBoost aggregates patterns across millions of users, then normalizes the data to prevent gaming. Google patented the underlying formula in 2006: LCIC – Long Click divided by total Clicks.
The processing pipeline runs in five stages: collection, classification, squashing (normalization), 13-month aggregation, and re-ranking adjustment. NavBoost doesn’t create initial rankings – it modifies positions of results that have already qualified through Google’s core algorithms. Think of it as a quality filter that sits on top of everything else.
The practical implication: you can have perfect on-page SEO, strong backlinks, and excellent topical authority – and still get demoted if users consistently bounce back to Google after clicking your result. Conversely, a page with modest traditional signals can climb if it consistently earns lastLongestClicks.
This Now Feeds AI Answers Too
NavBoost-style engagement signals now influence what gets cited in AI Overviews and AI Mode. The pages most likely to be cited in AI-generated answers are the same pages that have already demonstrated strong click performance in traditional search.
This creates a flywheel. Pages that earn quality clicks in organic search get cited in AI Overviews. Being cited in AI Overviews drives more quality clicks. And more quality clicks strengthen the page’s ranking signal for both pipelines.
It also explains a data point that’s been puzzling SEOs: brands cited in AI Overviews earn 35% more organic clicks and 91% more paid clicks compared to uncited competitors. The AI citation isn’t just visibility – it’s a trust signal that drives engagement on every other surface.
The Citation Decoupling
One more data point that changes the strategy: in mid-2025, roughly 76% of pages cited in AI Overviews also ranked in the top 10 organically. Six months later, that number dropped to 38%.
Google’s AI is increasingly pulling from sources that don’t necessarily dominate the blue links – pages with strong engagement signals, topical depth, or unique data that the model finds useful for generating answers. This means the old assumption – “rank in the top 10 and you’ll get cited” – is becoming less reliable. You need both: strong organic rankings AND the kind of content that earns quality engagement signals that feed into FastSearch.
What This Means in Practice
1. Measure what happens after the click, not just the click itself. Google Analytics engagement rate, scroll depth, time on page, and pages per session are now the signals Google is actually measuring through NavBoost. If your pages have high CTR but high bounce rate, that’s a NavBoost red flag – not a win.
2. Stop publishing content that restates what’s already known. Liz Reid was explicit: Google has expanded the definition of low quality beyond spam to include content that “doesn’t add very much, kind of tells you what everybody else knows.” If your page is a rephrased summary of the top 10 results, it will earn badClicks and get demoted, regardless of how well it’s optimized on-page.
3. Make the first 30 seconds count. NavBoost’s lastLongestClicks metric rewards sessions where the user stays engaged.
4. Invest in content that can’t be summarized by an AI Overview. The clicks that survive the AI filter are the ones where users want more than a summary. Original data, interactive tools, detailed case studies, expert commentary, community discussion – these are the content types AIO can’t absorb.
5. Track your AI citation presence alongside organic rankings. With citations decoupling from top-10 rankings, a page can rank well but never get cited, or rank modestly but appear in AI Overviews regularly. AI visibility tools like Beamtrace are built specifically for this – tracking brand mentions and citations across AI platforms – giving you visibility into the engagement flywheel that now connects organic performance and AI answers.
The Uncomfortable Bottom Line
Google has been grading clicks since 2005, but the stakes changed recently. When AI Overviews filter out 58% of clicks, the remaining 42% carry disproportionate weight in NavBoost’s calculations. Every click is more consequential, and every bounce becomes more damaging.
Liz Reid summarized it cleanly: people want content from a human perspective – the unique thing you bring to it. The old SEO model optimized for visibility: get seen, get clicked. The new model optimizes for satisfaction: get clicked and deliver something the user couldn’t get from the AI answer alone.
