Why your top-ranking pages are losing traffic (and what to do about it)

Your rankings haven’t moved. Your content output is up. Your traffic is down anyway. If that sounds familiar, you are watching the citation economy replace the click economy in real time, and most B2B teams are still measuring the wrong thing.

This is not a penalty. It is not a content quality problem. It is a structural shift in how Google and other AI-driven search systems deliver answers, and it is hitting informational and research queries hardest. Mid-market B2B sites, which depend heavily on top-of-funnel explanatory content, feel it first.

The decoupling of rank and traffic

For two decades, ranking position predicted traffic with reasonable accuracy. Higher placement, more clicks. That relationship is breaking.

Seer Interactive measured organic click-through rate dropping from 1.41% to 0.64% on queries where AI Overviews appeared, even when the same ranked listings still showed on the page. Search Engine Land, citing Ahrefs data across roughly 300,000 keywords, reported the click-through rate for the first organic result fell by 34.5% when AI Overviews surfaced. BrightEdge data, also reported through Search Engine Land, showed enterprise search impressions rising 49% year over year while clicks dropped 30%.

Read that last one twice. Your content is being shown more. It is being clicked less.

Search Engine Land also reports that 27.2% of U.S. searches ended without a click in March 2025, up from 24.4% a year earlier. That number will keep rising as Google expands AI Overviews coverage.

So when a marketing manager pulls up Search Console and sees impressions climbing while sessions slide, the instinct is to assume something broke. Usually nothing broke. The page is doing exactly what it always did. The results page is doing something new.

What the system is doing instead

Traditional search retrieves documents. Generative search retrieves meaning. Google’s AI Overviews assemble explanations from multiple sources and present them above the ranked list, with citation links shown as supporting references rather than primary destinations (Google Search Central, 2025).

The retrieval happens at the passage level, not the page level. OpenAI documentation describes this as semantic retrieval, where the system matches the meaning of a query to the meaning of stored text segments, even when the words differ. A long page that covers five topics might contribute one paragraph to one answer and nothing to four others. A shorter, sharper page where each section addresses one question cleanly can show up across many answers.

This is why your best-performing pillar page may be quietly failing as it expands its footprint in impressions. The system is reading it, pulling from it, citing competitors alongside it, and giving the user enough to move on without clicking.

Authority still matters. It just works differently. Authority now depends partly on whether your explanation matches how the topic is commonly described across trusted sources. Pages with idiosyncratic phrasing, narrow framing, or unusual definitions tend to get passed over during synthesis because the model cannot reconcile them with the broader pattern.

Most agencies get this wrong. They keep optimizing for keyword density and exact-match headings on pages that are losing traffic for reasons keyword tools cannot see.

Diagnosing your actual problem

Before changing anything, separate the noise from the signal. The first thing we check on a client site is whether the impression-to-click gap correlates with AI Overview presence on the target queries. If impressions are flat and clicks dropped, you have a different problem (intent shift, SERP feature change, seasonality). If impressions are up and clicks are down on informational queries, the pattern fits generative search behavior.

Pull the queries where the page ranks in positions 1 through 5 and CTR has fallen meaningfully over the past 6 to 9 months. Run those queries manually. Note which ones now trigger an AI Overview, who gets cited, and how the answer is structured. This 30-minute exercise tells you more than any rank tracker.

A few patterns we see repeatedly:

• Definition queries (what is X, how does X work) lose the most click volume and are the hardest to recover through ranking alone.
• Comparison queries (X vs Y) sometimes recover when the page presents a structured comparison the Overview can cite directly.
• Bottom-funnel and navigational queries (pricing, login, specific product names) are largely unaffected.

If your traffic loss is concentrated in the first two categories, the work ahead is content design, not link building.

On-page changes for generative visibility

Page-level rewrites should target retrieval readiness. A few things move the needle.

Lead each section with a direct answer. If the heading is “What is SOC 2 Type II,” the first sentence should define it plainly. Do not warm up. Do not bury the definition under three paragraphs of context. Generative systems evaluate passages independently, and a passage that requires surrounding context to make sense gets weighted lower or skipped entirely.

Use headings that name the topic, not headings that tease it. “SOC 2 audit timeline” works. “Everything you need to know about getting audited” does not. The system has to match the heading to a query intent, and clever phrasing breaks that match.

Keep terminology stable across pages. If your guide calls it a “buyer committee” and your blog calls it a “buying group” and your product page calls it “stakeholders,” you are training the system to treat your content as three disconnected sources. Pick one term per concept and use it everywhere.

Place definitions before examples. Examples reinforce a defined concept; they cannot substitute for one. OpenAI’s documentation notes that well-structured text improves retrieval quality because models align related passages more easily during synthesis.

Section length matters less than section coherence. A 120-word section that answers one question fully beats a 600-word section that answers three questions partially. Splitting bloated pillar pages into focused subsections, each with its own H2 and direct answer, usually produces measurable lift in citation frequency within 60 to 90 days.

This is where the website itself becomes the constraint. WordPress sites with deep, messy heading hierarchies, inconsistent template usage, and slow rendering give retrieval systems less to work with. Part of what we do at 321 is rebuild the underlying site architecture so each page presents clean, extractable sections, with schema and internal linking that reinforce entity relationships rather than scatter them. Content fixes alone won’t carry you if the site structure works against retrieval.

Reporting that matches the new behavior

Sessions and CTR still matter for transactional and navigational queries. They no longer describe what happens upstream. If your reporting stops at clicks, you cannot see whether your content is influencing buyers who never visit.

Citation presence is the cleanest leading indicator. Track which of your pages appear as cited sources within AI Overviews across your priority topic set. Tools like Profound, Otterly, and AthenaHQ are reasonable starting points, though manual spot-checks on twenty or thirty target queries each month will tell you most of what you need to know. Pair that with branded search volume in Search Console (filtered to queries containing your company name), which tends to rise when your brand is being mentioned in generated answers users do not click through.

Seer Interactive found that organic CTR rose from 0.74% to 1.02% when a brand appeared in an AI Overview compared with appearing only in traditional results, and paid CTR rose from 7.89% to 11% under the same condition. Inclusion produces better engagement on the clicks you do get. That is the asset worth tracking.

A note on attribution. Demo requests, branded direct traffic, and sales conversations referencing topics you publish on are all assisted-influence signals. They are messy. They are also more honest than a CTR chart that pretends nothing changed.

Where this goes

The decoupling of rank and traffic is not a temporary glitch in the SERP. It is the new baseline. Teams that adjust their content design, site structure, and reporting in the next two quarters will hold visibility as AI Overviews expand to more query types. Teams that keep optimizing for the 2019 model will watch their best pages quietly stop performing, even as their rank tracker shows green.

If you are looking at a Search Console chart that does not make sense and a content calendar that does not seem to be moving the business, we are happy to take a look. We do this work for mid-market B2B teams every week, and the diagnosis is usually faster than people expect. A short conversation will tell you whether the problem is your content, your site, or your measurement.