Now that AI summaries cover 80% of searches, what does this mean for the content side?
AI summaries now cover 80% of searches, referring to both sides of the same thing: search requests have not decreased, but they are digested by summaries at the display level. The review on the day was that AI Overviews appeared on more than 80% of search requests, while a year ago this ratio was about 50%, and the previous year about 15%; the percentage of zero-click searches has increased from 56% to 69%. This means that the evaluation criteria for the content side are changing from 'whether there is a ranking' to 'whether it can be abstracted' - rankings still exist, but clicks no longer occur automatically.
How is the 80% coverage rate calculated
These numbers all come from continuous observation of the same batch of query words, not a census of the entire site's traffic. The observation scope usually involves selecting a group of query samples, calculating the proportion of AI summaries that appear among them, and then comparing them over time. The type of sample determines the result: words with a commercial intent or informational questions have a high coverage rate; words with a navigation intent (looking for a specific site) have a low coverage rate. So when you see the number 80%, the correct usage is to treat it as a trend line, not to copy it into your monthly report as site data.
The trend itself is clear: from about half to eight in a year, indicating that this layer has already changed from supplementary function to default display.
What does the rising of zero-click mean for the content side?
The most direct feeling on the content side is that three things happen simultaneously. First, the absolute value of natural traffic decreases, but the decrease does not mean that the content is ineffective, because readers may have finished reading the conclusion somewhere you didn't count. Second, the search volume of brand words is often more stable than that of non-brand words, indicating that some readers return to the site through the names in the summary. Third, the structure of cited sources is changing: observations show that a considerable proportion of AI citations come from content on third-party sites, not just the brand's own website.
A common mistake is to attribute everything to 'the content is not written well', and then to discard all old drafts and rewrite them. A more efficient approach is to first distinguish which pages are naturally dependent on clicks for monetization and which are playing the role of being cited.
Four things that the writing layer can do
| Action | Purpose | The landing points within the site |
|---|---|---|
| Move the conclusion paragraph to the forefront | Allow paragraphs to be extracted independently from the text | Directly answer the question at the beginning of each article, and write the subheadings in question form |
| Judge the data and sources | Improve the probability of being quoted | Retain the口径 and time of the number, and avoid absolute expressions |
| Clear internal links and anchor texts | Discover related content together | Anchor text management to unify in-site keyword pointing |
| Site map and active push | Ensure that the page can be indexed | Automatic generation of Sitemap, Baidu push and Bing link push |
These four things, the first two are writing issues that anyone can do; the last two are site configuration issues that determine whether the content you have written can be discovered stably.The site already has these capabilities: Sitemap automatically generates the page list to be submitted, link push is responsible for actively informing search engines of new content, static URL and 301 redirect ensure the stability of the address, and anchor text management makes the direction of in-site keywords controllable.Another entry facing large models is also inside——LLMs.txt can be automatically generated by the site, making it easier for language models to read your content in a structured way, which is the other half of the on-site optimization engine (GEO).
How to verify whether you are cited
Don't rely on intuition. There are three reproducible checks: take your ten core question sentence patterns and ask them in AI search, note whether the source of citation includes your site; fix the same batch of questions and run them once a week, observe whether they enter or exit the list; at the same time, observe whether the natural entry from in-site brand words has increased. The cost of these three steps is not high, but it can avoid 'optimization based on impression'.
Common questions
Covered 80%, do we still need to do traditional SEO?Yes. The reference sources of the AI summary are still from the pages included and crawled, site maps and active push are not alternatives for these basic tasks.
Should a version for AI be written separately on the content side?No need to write another one. Clear the conclusion paragraphs and data with the口径, it can meet both readers and the model.
How should the traffic recommended by AI be evaluated?Observations show that the conversion rate of visitors recommended by LLM is not low, but it is not suitable as an independent channel for setting KPIs, it is more suitable as a supplementary indicator for brand search and lead quality.