“We want to be cited by AI, but we can’t keep up with site maintenance” — How we reviewed our operations so generative AI can correctly recognize our website

“We want to increase the number of articles cited by AI, but we don’t know where to start.”

Recently, we have been hearing more comments like this from people involved in website operations and content publishing.

They want their company’s website information to be correctly recognized by generative AI tools such as ChatGPT and Gemini, not just optimized for SEO. But in practice, it is hard to know whether improving article content is enough, or what needs to be checked on the website side.

We felt the same challenge ourselves.

When looking at access logs for our own website, we noticed that generative AI crawlers had started visiting at a level we could no longer ignore. According to crawler access statistics published by a major CDN, OpenAI’s GPTBot access share reportedly increased from around 5% in May 2024 to around 30% in May 2025.

Website visitors are no longer limited to the “humans” and “search engine crawlers” we had traditionally assumed.

So, as a production company, what kind of operations should we build so that generative AI can correctly recognize our website?

In this article, we explain the maintenance workflow we actually reviewed and how we incorporated MONJI+’s “Website Anomaly Detection & Improvement” feature into our operations.

---

Why this issue occurs

When trying to create articles that are cited by generative AI, attention naturally tends to focus on the “content” and “structure” of the article.

Of course, accuracy and readability matter. But through actual operations, we realized that the conditions for AI citations are not completed by the article alone.

If the condition of the website where the article is hosted has deteriorated, even a well-written article may become difficult for AI to read correctly.

Background 1: AI crawlers are becoming more visible

Until now, web publishing has mainly focused on SEO, with search engine crawlers such as Googlebot in mind.

But today, generative AI-related crawlers such as GPTBot, ClaudeBot, and Perplexity’s crawler are also visiting websites.

One of the goals of the media we operate is not only to rank in search results, but also to be cited and referenced in responses generated by AI.

That is why we now care more than before about how often AI crawlers visit our website, and what condition the site is in when they crawl it.

Background 2: Even if you publish articles, site-side issues can prevent the content from being understood

We have also experienced this failure ourselves.

We created articles with the aim of being cited by AI, but a few months after publication, we found that some related links within those articles were broken. The articles themselves still existed, but the past articles and resource pages referenced inside them had become 404 pages.

We also found articles where image alt attributes had been left blank.

Even if the text says, “Please see the figure below,” AI may have difficulty understanding what the figure means if there is no alt attribute explaining the image. In other words, the visual evidence supporting the article’s argument was not being properly conveyed to AI.

In addition, on a client website, we found that old robots.txt settings had been left in place, unintentionally blocking some AI crawlers.

If AI cannot read a page, the possibility of that page being cited by AI does not arise.

In other words, when thinking about AI citations, it is necessary not only to create articles, but also to continuously maintain the condition of the website itself.

---

Steps to solve the issue

In parallel with writing articles that are more likely to be cited by AI, we established operations for regularly checking site-side anomalies.

Here, we introduce the flow we actually adopted as a production company in three steps.

STEP 1: Visualize the current state

The first thing we did was identify site-side anomalies that could interfere with AI citations.

In our work, we pay particular attention to the following items:

• Broken links
• Image loading failures
• Missing image alt attributes
• Missing titles or descriptions
• Missing tags
• Outdated robots.txt settings

Each of these may seem like a small issue on its own.

However, if broken links prevent AI from reaching related information, images are not explained, or missing meta information makes it difficult to understand the page topic, these issues can accumulate and make the website harder for AI to read.

That is why we started using MONJI+’s “Website Anomaly Detection & Improvement” feature to regularly check key pages that AI crawlers are likely to visit.

The pages we check include article pages, case study pages, and service description pages—pages that are more likely to be cited. We check both PC and smartphone displays for broken links, missing images, missing titles, missing alt attributes, and missing tags.

STEP 2: Record and list issues as part of monthly operations

Next, instead of treating detected anomalies as one-off fixes, we turned the process into a monthly operation.

Our workflow is as follows:

1. Automatically crawl key pages on a regular schedule
2. List the detected results
3. Create correction requests for items that require action
4. After correction, check the affected scope through integration with Google Analytics
5. Store recurrence-prevention checklist items in the Wiki

The important point is not to stop at “finding anomalies.”

By connecting detected issues directly to correction requests and managing post-fix confirmation in the same flow, it becomes easier to track the condition of articles after publication.

We are also now able to see, in chronological order, which articles had what kind of issues and when.

As a result, the team has developed a stronger awareness that publishing an article is not the end. We need to keep the article in a state where AI can continue to read it properly.

STEP 3: Improve based on several months of accumulated results

When detection, correction, and recording are continued monthly, several months of history accumulate.

This makes it possible to see trends that would not be visible from one-off error handling.

For example, we may notice that alt attributes are often missing in a particular template, that links to older articles tend to break at certain intervals, or that meta information is often overlooked in a specific group of pages.

Based on these trends, we review future checklist items and production workflows.

For us, this operation is not merely anomaly detection. It functions as a system for regularly maintaining the quality of the entire website so we can continue publishing content aimed at AI citations.

---

Changes we actually saw

Before incorporating MONJI+ into our maintenance operations, our content publishing efforts were focused almost entirely on “writing good articles.”

We carefully selected topics, created article structures, refined the text, and published the articles. But in all honesty, we were not fully tracking what condition those articles were in after publication within the website.

After changing our operations, our perspective expanded from “article quality” to “the quality of the entire website where those articles are hosted.”

If a broken link appears, it is detected and a correction request is created. If an image is missing an alt attribute, it is listed. Missing meta information also becomes part of the monthly check.

As a result, we can now continuously check the necessary preconditions for aiming for AI citations.

Of course, this operation alone does not guarantee that AI will cite our content.

But at the very least, it has become easier to build a foundation that prevents us from leaving the site in a state where AI has difficulty reading it.

---

Notes and limitations

There are areas this operation can and cannot cover.

What MONJI+’s anomaly detection feature can detect is mainly issues that can be judged mechanically.

For example, broken links, whether images load successfully, whether tags exist, and whether meta information is missing are items that can be detected automatically.

On the other hand, qualitative judgments still require human eyes and hands. These include whether the article content itself is worthy of being cited by AI, or whether structured data markup is appropriate.

In addition, analyzing AI crawler access behavior in detail or measuring how often AI actually cites the content requires separate efforts, such as access log analysis and AI citation log measurement.

What we are currently working on is preparing the “preconditions” for AI citations on the website side. It is important to understand that this does not directly change AI-side behavior.

To be cited by generative AI, multiple layers need to be in place, including robots.txt settings, content quality, site structure, the absence of broken links, and the condition of images and tags.

That is why it is important to create a system that can continuously run detection, response, and knowledge accumulation.

---

MONJI+: A website operations support platform born from the field

To publish content aimed at AI citations, it is necessary to maintain not only the quality of the article itself, but also the condition of the website where the article is hosted.

MONJI+ is a website operations support platform created to make it easier for teams to continue that kind of operational work in the field.

MONJI+ brings together the “people” involved in website operations into one “team” and helps solve issues across every phase of website operations.

We are not trying to develop a “completed product” from the beginning.

The real issues we need to face exist in the field of website operations. That is why we have continued to listen to the real voices of people working there.

By listening to each voice, resolving small points of friction through repeated updates, and developing new features from actual field challenges, MONJI+ has grown into a platform supported by users in 77 countries around the world.

▼ Learn more about MONJI+ here
https://monji.tech/ja/plus/

A world where people working in website operations can proudly say, “I love this work.”

▼ We look forward to hearing your voice to help make that world a reality
https://monji.tech/ja/plus/co-creation/

---

Conclusion

As generative AI crawlers visit websites more frequently, the assumptions behind website operations are gradually changing.

To create articles that can be cited by AI, it is not enough to improve only the content and structure of the writing. If site-side issues such as broken links, missing image alt attributes, missing meta information, or outdated robots.txt settings remain, AI may have difficulty recognizing the site correctly.

We incorporated MONJI+’s “Website Anomaly Detection & Improvement” feature into our maintenance workflow and began regularly checking the condition of key pages.

Detected results are connected to correction requests, the affected scope is checked with Google Analytics, and recurrence-prevention checklist items are stored in the Wiki.

By running this monthly cycle, we are now able to continuously maintain not only individual articles after publication, but also the quality of the entire website where those articles are hosted.

When publishing content aimed at AI citations, “whether the article remains readable by AI” is just as important as “what the article says.”

As a production company, we will continue improving the foundation of our content publishing by looking not only at each article, but also at the operational quality of the entire website.