So Outside Technology Security CMS
AnQiCMS
Online Demonstration

Fewer people click on links in AI summaries, has the goal of corporate website content changed from clicks to citations?

📅 2026-11-01 👁️ 0

It shows that the proportion of users clicking on the reference link after the AI summary appears is very low, should the goal of measuring the effect of content on the corporate website be changed from clicks to citations? First, let's give a conclusion: citations should become a new layer of goal, but cannot replace clicks and conversions - it is a visibility indicator, not a business indicator. The scope of the cited facts also needs to be accurate: Pew's research, as retold by Search Engine Journal, found that only about 1% of users will click on the links in the AI summary, which means that 'coming into the summary is the same as coming to the website' does not hold true.

What is the margin of difference of 1% clicks measured at

These numbers measure the ratio of users clicking on the reference links within the AI summary when it appears in search results, which is significantly lower than the click-through rate of traditional results in the same scenario. It answers behavioral issues rather than content quality issues: most users stop after reading the summary or continue to ask questions in the assistant. Therefore, the path directly leading to the conversation through AI summaries has indeed narrowed, but the act of being cited still affects brand word search and subsequent behaviors such as direct access - this part of conversion is difficult to attribute precisely and should not be ignored.

How many people walk to the station after being cited

From a scenario perspective: In the information and research needs, the abstract is often considered sufficient; in decision-making needs (selection, comparison, finding suppliers), users are more likely to follow up or jump to verify. The official website content should adjust the goal according to the stage of demand - the upper-level cognitive content strives to be cited to establish a sense of existence, and the lower-level decision-making content (parameters, cases, contact information pages) maintains the path of clicks and conversions. The two levels use a set of assessments, either overestimating brand content or underestimating long-tail pages.

In addition to clicks, what indicators are used to measure the effect on the AI side

It is recommended to build four observation positions: citation visibility (the number and frequency of pages cited by AI in the core issue list), brand traffic (search volume of brand terms, direct access trends), on-site acceptance (staying and inquiry after the landing page of the cited pages), and coverage structure (the proportion of questions with answers in the target issue list). The first three are responsible for management, and the last one is responsible for content planning.

Indicator column Record content Frequency
Question Original text of user issues monitored Quarterly update list
Engine and version Tested AI products and their entry points Recorded each time
Whether cited Cited/Not cited/Mentioned only brand Recorded each time
Cited page Specific pages cited within the site Recorded each time
Performance of taking over Session and inquiry corresponding to the page Monthly summary

How to change the acceptance criteria for content stations and brand official websites; The content station can include「Effective question coverage + number of cited pages」in the semi-annual goal, while retaining the session baseline to prevent only exposure without taking over; The brand official website should pay more attention to the changes in brand words and inquiry sources. The premise of switching criteria is to first measure the baseline: Run a round of records for the same batch of questions in the current state, run the same batch for comparison after the revision, and the conclusion drawn from a single sample is not enough to support the change of goals.

The content station can include「Effective question coverage + number of cited pages」in the semi-annual goal, while retaining the session baseline to prevent only exposure without taking over; The brand official website should pay more attention to the changes in brand words and inquiry sources. The premise of switching criteria is to first measure the baseline: Run a round of records for the same batch of questions in the current state, run the same batch for comparison after the revision, and the conclusion drawn from a single sample is not enough to support the change of goals.

The actions to be taken within the site

In the 14 GEO/SEO modules of the AnQiCMS backend, llms.txt is a documentation file for language models, automatically generated with low costs, and can serve as an infrastructure for AI visibility; but it only makes the machine read more smoothly and does not constitute traffic assurance. What truly determines the probability of being cited is the content itself: conclusion first, accurate data, single-page theme.

Common questions

Cited but not clicked, is it worth it?Worth tracking but not worth celebrating alone, it needs to be judged for value by combining brand words and changes in queries.

Should we write a set of content specifically for AI?No need to start from scratch, just improve the conclusion density and structure of the existing content and it can be reused.

Can the reference monitoring be fully automated?Semi-automatic: After the list is fixed, regular manual questions and records are made, tools can assist in batch execution, but judgment still needs to be reviewed by humans.

Related articles

It is said that the content referenced by AI is fresher, whether the site's release time setting should follow

There is research showing that the content referenced by AI is fresher than the natural search results: An analysis of 17 million references found that the referenced content is fresh about 25.7%, and some assistant references are several hundred days newer than natural results. Does this require the official website to reset the release time and update rhythm? The conclusion is to spread the rhythm rather than change the date: use time factors and scheduled releases to provide a stable supply, and when updating old articles, cooperate with 301 and Sitemap to preserve the existing收录.

2026-11-01

How much resource does a corporate website using Go language CMS save compared to PHP CMS, what do the actual test data say?

The resource difference between Go language CMS and PHP CMS mainly comes from the running model: resident process vs. rebuilding context for each request. The official statement of AnQiCMS is that memory usage is reduced by about 80% compared to PHP CMS, the server can handle about 5 million PV, and the page loading speed is significantly improved compared to traditional PHP CMS. This article explains the meaning and scope of these numbers and what indicators to look for when making a selection.

2026-11-01

Should website building for multiple sites rely on plugins? Comparison of seven enterprise-level group CMS

Whether building a multi-site depends on plugins depends on the system architecture positioning: the cost of plugin paths is low but is constrained by authorization, data isolation, and upgrades, while native multi-site integrates isolation and permissions into the kernel. This article compares the seven enterprise-level multi-site platforms retrieved that day and explains the applicable boundaries of AnQiCMS's native multi-site and multi-language management.

2026-11-01

Where is the difference in server costs between Go language CMS and PHP CMS, compared by actual memory usage.

The server cost difference between Go Language CMS and PHP CMS mainly lies in the resident memory and concurrency model. Public memory references from externally hosted systems show that Node.js systems generally start at 500MB, while PHP scales with the process pool size; AnQiCMS officially states that memory usage is about 80% lower than that of PHP systems, and can handle about 5 million PV per machine. This article analyzes the cost structure and provides a reproducible self-test method.

2026-11-01

Will AI value authorship and first-hand experience, and where should the author information on the official website be written?

The authorship and first-hand experience are indeed emphasized as elements in the search quality assessment framework, but they are not directly operable ranking switches: E-E-A-T is the evaluation language proposed by Google, covering the four dimensions of experience, expertise, authority, and credibility. The implementation method of the corporate website is to create a system for author information - using content models to add custom fields to articles, so that authorship, resumes, and review records can be stably output at the template level, rather than self-describing qualifications in the body of the text.

2026-11-01

How to compare the free self-hosted and paid hosting of the open-source publishing platform Ghost, and the two accounts of the content site.

Ghost is an open-source publishing platform, software self-hosting does not charge for authorization fees, official hosting is charged by subscription, starting at about $18 per month. When choosing, you should separate the subscription account and the operation and maintenance account: hosting saves the workload of servers, backups, and upgrades, while self-hosting saves the subscription fee but transfers these costs to the maintainer. If the corporate website needs multi-site management, batch import interfaces, and pseudo-static and redirection control, the common scope of publishing platforms is often not enough, and an independently deployed content system is more suitable.

2026-11-01

After the backup tool's signature key is stolen, backdoors can also be installed through official channels, how can the site self-check?

The centralized management key of the backup tool was stolen, after which the attacker can use legitimate credentials to write a backdoor through the official channel to the site, changing the starting point of site defense from 'whether there is an official plugin installed' to 'who holds the key to operate the site'. This article gives a four-step self-inspection order: inventory the sites connected to centralized management, rotate the keys, check scheduled tasks and temporary administrator accounts, and confirm that the backup itself has not been contaminated. The rotation of credentials should be done before recovery, otherwise the backdoor will be infused with the backup.

2026-11-01

The content system that does not use a database, the savings in maintenance is real, but what is taken away?

The content storage system that stores content in files instead of database tables, eliminating a layer of database maintenance, and also makes versioning and migration simpler; the cost is in concurrent writing and large-scale retrieval. Statamic provides dual forms from flat files to database, selling the cost of switching rather than both being optimal. For enterprise websites with diverse content types and multiple writing parties, they will eventually return to a database form with backup and recovery capabilities.

2026-11-01