Everyone in the industry is going through the same thing right now.
Your content team provides a strong asset, but traffic remains stable or decreases. You view the analytics, see the line move sideways, and conclude that the AI content has stopped working.
This verdict is often wrong. The data may be accurate, but the numbers do not reflect the qualities of the content.
By changing what you measure, you may find that your content performs better than your analytics suggest.
The problem
During the first four months of 2026, 68% of Google searches in the United States end without a clickbased on SparkToro analysis. This is up from 60% in 2024. The increase in zero-click searches is largely due to AI insights and people doing more searches without leaving Google.
Ahrefs published an analysis of click data which initially AI previews are estimated to have reduced clicks to the first result by 34.5%. When they re-evaluated with new datathe prediction for first place was 58%.
The question now is what teams do about it.
Why data stopped working
For 20 years, traffic was a good indicator of successful content. When a page was interesting, Google directed people there and your analytics recorded that visit. Since value and traffic are often correlated, you can infer one from the other.
Now traffic and value are disconnected. Search Console shows clicks, impressions, and average position, but it Does not differentiate between clicks from a traditional search result, AI preview, or AI mode.
Google expanded its AI results with more link options without providing websites with visibility insight. Therefore, when clicks decrease, it’s unclear whether AI previews are absorbing traffic, rankings are dropping, or people are reading a summary of your content without clicking.
A drop in click-through rate does not necessarily indicate that fewer people are engaging with your content. Seer Interactive discovered that although the brand cited Overview CTR dropped 61% quarter-over-quarterthe number of clicks on these pages remained almost unchanged. The decrease in rate is because impressions are increasing faster than clicks.
Google refers to the clicks eliminated by AI Overviews as “bounce on clicks“, or quick visits where people find a fact and then leave. Liz Reid, head of search at Google, may be right. The problem is that Google measures how often people return to search, reflecting Google’s retention, not the value of your content. Publishers don’t have a way to measure clicks from AI surfaces, and until Google provides one, any positive spin-off is just an assertion.
What we do know is that when an AI preview is shown, people click on a result about 8% of the time, compared to 15% without that result. Only 1% click on a link in the preview itself. These are real losses, but the same searches still expose your work to people who never click.
What actually happens to “failed” content
A analysis of approximately 846,000 search sessions found that people slow down when a preview is shown. They scroll, go back, revisit listings, and carefully consider the options on the results page before making a decision. The search results page now performs more functions that were previously handled by a landing page.
A randomized field experiment found that when AI previews appear, they reduce outbound organic clicks by 38%, but self-reported satisfaction remains unchanged whether the preview is shown or removed. If the lost clicks had only been low-value bounces, satisfaction would likely have dropped when the summaries disappeared. This is not the case.
The seer discovered that the cited pages receive approximately 120% more clicks per impression compared to non-cited pages in the AI preview results. GWI data indicates that frequent users of AI searches also tend to click on sources more often. Half of daily users click on a quote, compared to just 14% of occasional users.
The results of these studies describe an audience that evaluates, compares, and sometimes makes decisions in areas where your analyzes cannot track them.
What to measure instead
Keep an eye on branded query volume and direct traffic, as these are signs of influence that don’t necessarily translate into clicks. When your content generates demand that you can’t capture immediately, it can often show up later when someone searches for your name or visits your URL.
Next, monitor your presence on AI surfaces. When available, Search Console’s generative report shows impressions for AI previews and AI mode, while third-party tools estimate citation points. Google combines the clicks generated by these features with your total search data. So you can’t separate them. An impression occurs when your page link appears, but this does not indicate whether your content influenced the response.
Since fewer people are clicking, those who do are more likely to be further along in their decision-making process. Measure their actions after landing, such as reading depth, repeat visits, newsletter signups, and conversions, as these metrics provide more information than just the number of sessions. For example, a page with half the usual traffic but twice the conversion rate achieves its goal.
Rand Fishkin’s advice is to create a correlation dashboard rather than a single traffic KPI. This involves plotting your posting schedule based on branded searches, direct conversions, and conversions, and observing how they evolve together. It’s gentler than a clean attribution number and closer to how influence currently spreads.
Measuring influence is a separate task from reporting it. At Search Engine Journal, we recently looked at the reporting aspect, explain to a financial director why traffic is low but revenue remains stable. This article is about the previous step: seeing what your content did in the first place.
To report on content that contributed to revenue, you first need visibility into it, but Google’s default dashboard doesn’t show this information.
There is not yet a clear way to measure influence, and to claim otherwise would be misleading. You get triangulation, multiple imperfect signals that, when combined, offer more information than traffic data alone.
What it looks like by category
AI insights primarily handle information and search queries. Branded, local, and high-intent transactional searches perform best in organic search, categories that SparkToro says it still benefits from SEO.
In ecommerce, buying guides and “best of” pages are impacted the most because they directly answer the overview question, while product and category pages continue to convert.
Publishers face the most challenges, and discussing influence seems meaningless when reduced traffic leads to lower advertising revenue. The most vulnerable visitors are those who rely on search, while loyal readers who visit directly or through your app are less likely to be attracted by the preview.
How to answer
If people click from a preview to a page with the same summary they’ve already seen, they may leave. Based on GWI AnalysisThis helps add an extra layer to your pages, something that AI can’t produce from your existing text. It could be an interactive graphic, a video, or a free download.
Another way to win is to write content that people will remember after they leave. Memorable content is great for inspiring brand searches later, even if it doesn’t capture immediate conversions.
Be careful not to delete pages based solely on traffic. Before deleting a page that has lost clicks, check if it is still referenced and if the brand request has evolved during its period of activity. A page can decrease in traffic while still maintaining its usefulness.
What to do next
You can’t choose what to write next based on a number that doesn’t reflect the success of your content. You have to change what you measure, which is more difficult than interpreting a line on a graph.
New signals are more chaotic, partially underdeveloped and move more slowly. However, they answer the right question: whether your efforts actually reach people and influence their actions.
Join our live webinar on Thursday, July 9 at 2 p.m. ET to learn more about what to do when your content no longer works.
More resources:
Featured Image: Cagkan Sayin/Shutterstock
