AI abstracts will significantly reduce the click-through rate of the top position, should content sites still focus on ranking words?

Keep an eye on it, but change the object of attention. A third-party comparative study has provided such a criterion: in a sample covering 300,000 keywords, the average click-through rate decreased by 58% when AI abstracts appeared, and by December 2025, the first click-through rate was only 0.016. These numbers do not mean that 'ranking is useless', but rather that 'ranking is no longer equal to clicks' - the logic of directly converting keyword indicators into traffic has failed. Content sites should continue to maintain a keyword library, but the evaluation criteria should shift from 'rank to which' to 'whether it is cited, and whether there is conversion after citation.'

Where does the decrease in click-through rate come from?

This type of research usually performs a same-word comparison: the same batch of keywords, under two situations of displaying AI abstracts and not displaying abstracts, the natural click distribution of search results is counted separately. The differences are concentrated on the first screen, the abstract gives the conclusion first, and the user stops clicking after reading. To use this criterion correctly, it is necessary to pay attention to three points: it is an average value, the differences between words with different intentions are very large, and the extent of influence on transactional words and query words is different; it measures natural clicks, brand words and navigational queries are basically unaffected; it fluctuates with the change of abstract display strategy, and belongs to a阶段性 observation rather than a constant.

Using such numbers as the sole basis for decision-making can lead to problems, but ignoring them can also cause problems - if your content strategy is entirely designed to 'grab the top position', and the top position no longer brings the corresponding clicks, the return on investment changes.

Why does the ranking word index distort

The core of distortion is that the causal chain is broken in one link: before, ranking brought clicks, clicks brought conversion; now, ranking brings exposure, and whether to click after exposure depends on whether the abstract has finished the answer. Your page may be very high in position, but the answer is repeated in the abstract, and the click is zero; it may also be ranked low, but because it is directly quoted in the abstract, it gets a higher quality visit

The keyword library should be maintained according to intent

The keyword library is still infrastructure, the keyword library management of AnQiCMS is used to centrally maintain SEO keywords, its value is particularly evident in revision and writing new content. The thing that needs to be changed is the maintenance scope.

Step 1: Divide words according to intention. Define type (what is it), comparative type (how to choose among schemes), operational type (how to do it), and troubleshooting type (how to check for problems). Each layer should be written on its own page; do not let one page answer two types of intentions at the same time. Step 2: Prepare a segment that can be quoted independently for each word group, controlling the length to about two sentences, including the main entity, key restrictions, and the conclusion itself. This text should also serve as the first paragraph of the page. Step 3: Reflow the transformed data back to the word table, sort it by citations and effective visits, and eliminate words that have no long-term value instead of continuously expanding coverage.

The fourth step is easy to overlook: on-site association. Anchor text management makes keywords consistently point to the corresponding page, and there is a traceable path between the referenced page and the receiving page, where the user ends up after clicking depends on this structure rather than the ranking itself.

Accessibility and being cited are two different things.

Many sites treat 'Machine Read Me' as 'Machine Refer Me', which are actually two levels. The automatic generation of site maps and active push to Baidu, Bing solve the issue of discovery and indexing timeliness, which belong to the entry level; llms.txt automatically generates a list that organizes the site structure and capability口径 into a model-readable list, solving the issue of understanding cost. Neither guarantees that your brand name will appear in the answers.

Whether a citation depends on the form of the content: whether the conclusion is self-contained, whether the limiting conditions are stated clearly, whether specific steps and judgment criteria that cannot be easily described by others are given. General introduction paragraphs, even if included, are difficult to cite because they do not contain information increment.

On-site inspection rhythm

It is recommended to have a monthly convergence: confirm whether the new content is positioned according to the intention, check whether the cited word list has回流 into the keyword library, check the implementation status of the link push task, and view the display form of some high-value words. If a certain type of intention has not been cited for a long time, it is preferable to suspect the content structure rather than the number of publications.

Common questions

Should we still write a long article?Yes, but the structure needs to change. The value of a long article lies in its ability to carry a complete judgment, on the premise that each section is self-contained and can be cited separately, rather than having only one conclusion on a single page.

What does the decline in brand word traffic indicate?It indicates that the user has completed the need without leaving the search results page, in such cases, it is necessary to observe re-visit and direct access rather than just looking at single clicks.

Which one should we do first to see results faster?Add a conclusion that can be independently cited to the existing high-traffic page, which is usually faster than adding new content to see changes.

Should we clean the keyword library?It requires maintenance costs for words that have not been exposed or cited for a long time, and regular convergence is healthier than constant increase without decrease.