AI Citability: What Author Boxes, Tables of Contents and Other On-Page Features Achieve

Why are some websites cited as sources by ChatGPT, Google AI Overviews and other AI systems, while others barely appear despite good content?

Besides the quality of a text, its structure, identifiable authorship and timeliness apparently also play a role. A recent study by SISTRIX with 96,418 URLs provides interesting insights on this.

Particularly striking: websites with a table of contents are cited 1.7 times more often in the dataset examined. For an author box with photo, the factor is 1.4. Article schema, FAQPage schema and tables with headers also show clear correlations with citation frequency.

This provides a comprehensive empirical data basis for a discussion that in the GEO industry has so far often been shaped by individual tests and assumptions.

But what do these figures actually mean? Which measures make sense and where should you be cautious?

1. What makes a website citable for AI systems?

A good position in classic Google search does not automatically mean that a website also appears as a source in an AI-generated answer.

AI systems need to find relevant information, classify the content of a page and decide which sources are suitable for a specific answer.

Several aspects play a role here:

• Content relevance: The website answers a specific question or deals with a clearly defined topic.
• Clarity: Statements are formulated comprehensibly and professionally classified.
• Information structure: Headings, tables, lists and tables of contents facilitate orientation.
• Traceability: Authorship, professional competence and sources are identifiable.
• Timeliness: Information corresponds to the current state of knowledge, as far as the topic requires.

These requirements are not new. They belong to good editorial and technical design of websites.

Therefore, the interesting question is whether certain features are also reflected in the actual citation frequency by AI systems.

A study by SISTRIX provides concrete figures on this.

2. What the SISTRIX study with 96,418 URLs shows

On 22 October 2026, SISTRIX published the study "GEO Myths in Data Check: What 96,418 URLs reveal about AI Citations". The study was updated on 29 October.

43 on-page features of websites that were used as sources by ChatGPT, Google AI Mode or Google AI Overviews were examined.

The data basis:

• 96,418 URLs
• 17,380 domains
• 48,527 cited pages from the SISTRIX AI Index
• Data from the DACH region, second quarter 2026
• 43 examined on-page features

How was it measured?

The methodological approach is particularly interesting.

SISTRIX compared cited pages with non-cited pages from the same domain. This was intended to reduce the influence of general domain and brand strength.

Thus, the study focused more on which features at page level are associated with higher citation frequency.

The statistical evaluation was carried out using conditional logistic regression. In addition, a correction for multiple testing according to Benjamini-Hochberg was applied.

This is a significantly more meaningful basis than a simple comparison of domains of different strength.

The most important results at a glance

Word count is not an independent optimisation goal. SISTRIX points out that longer texts often also contain more headings, tables and lists. The correlation can therefore partly reflect the better structure of extensive content. Artificially stretching texts to 1,000 words cannot be derived from this.

The title and H1 evaluations were added afterwards and should be regarded as initial indications. In particular, the value for the year is not proof of an isolated title effect.

Source: SISTRIX, GEO Myths in Data Check (2026).

What does a citation factor of 1.7 mean?

The factors describe the relative citation frequency in the comparison examined.

A factor of 1.7 for the table of contents means that, of two compared pages from the same domain, the page with a table of contents was cited 1.7 times as often as the page without this feature.

Technically, these are odds ratios. With the citation rate of approximately 1.4 percent mentioned by SISTRIX, this roughly corresponds to the ratio of citation frequencies.

Important: A factor of 1.7 does not mean 70 percentage points more visibility nor a guaranteed increase for a single website.

The study shows statistical correlations but no proven cause-effect relationships.

3. Author box: Making trust and expertise visible

An author box is more than just a name under an article. It makes visible who is responsible for content and what professional experience stands behind a publication.

The SISTRIX study provides a concrete finding:

• Author box with photo: factor 1.4
• Author box without photo: factor 1.3
• Author description or bio: factor 1.3

Pages with an author box and photo are cited 1.4 times more often in the dataset examined than pages without this feature.

This is an interesting correlation. Until now, the author box was often seen as an editorial element that primarily serves to build trust with human readers.

The study suggests that traceable authorship also plays a role in AI citability.

What belongs in a good author box?

An author box should not list as many job titles as possible. What matters is that the information fits the respective topic and is comprehensible.

Meaningful components are:

1. Full name of the author
2. Authentic author photo
3. Relevant professional qualifications
4. Traceable professional experience
5. Thematic focus areas
6. Link to a more detailed author profile

An example: For a specialist article on technical search engine optimisation, many years of SEO experience and knowledge in web development are more meaningful than a general list of all professional stations.

Structured data can also support the assignment. An author profile can be described, for example, with Person schema and linked to an article.

Important: An author box does not create expertise. It makes existing expertise visible.

Author box and E-E-A-T

E-E-A-T is often mentioned in this context:

• Experience
• Expertise
• Authoritativeness
• Trustworthiness

Google describes E-E-A-T as part of its quality evaluation. Trust plays a central role here.

However, an author box is not an independent ranking factor that is activated by inserting a specific element. It can help to make relevant information about the author of content transparent.

Google explicitly recommends making it clear who created content and what further information about the author is available, where appropriate.

For AI citability, the SISTRIX study now also provides an empirical correlation.

4. Tables of contents: Structure is more than user-friendliness

The table of contents is one of the most interesting results of the study.

With a citation factor of 1.7, it achieves a higher value than the author box with photo.

Only about 14 percent of the pages examined had a table of contents with jump links.

This is remarkable because a table of contents is often considered solely as a convenience feature for readers.

In fact, it fulfils several functions:

• It makes the topic structure of an article visible.
• It shows which sub-questions are addressed.
• It links headings to specific text sections.
• It facilitates orientation in extensive content.

For AI systems, such a structure can also be helpful because the organisation and thematic connections of a page become easier to recognise.

What should a good table of contents look like?

A table of contents should reflect the actual structure of the article.

This includes:

• Clear and meaningful headings
• A comprehensible H2 and H3 hierarchy
• Working jump links
• No unnecessary or content-redundant points
• A sensible order of topics

A table of contents with many similar headings is of little use. It should already convey to the reader before reading which questions the article answers.

Especially for longer specialist articles and guides, it is therefore a sensible editorial addition.

5. Other factors for higher AI citability

Author boxes and tables of contents are only two components of a larger context.

The SISTRIX study also shows interesting results for structured data, tables, lists and update information.

Article schema and FAQPage schema

Article schema achieves a citation factor of 1.6, FAQPage schema a factor of 1.5.

This is interesting because the importance of structured data for GEO is often controversially discussed.

Schema.org makes it possible to describe content and their relationships in a machine-readable way.

For Article schema, for example, the following information can be marked up:

• Article headline
• Author
• Publication date
• Update date
• Publisher
• Featured image

The FAQPage schema describes questions and their corresponding answers.

The marked-up content must actually be present on the website. Markup should never pretend information that is not visible or not accurate for the visitor.

The results suggest taking Article and FAQPage schema seriously for suitable content. However, they are not proof that every schema markup automatically leads to more AI citations.

Tables and lists

The presentation of information also plays a role.

SISTRIX determined the following factors:

Tables are particularly suitable for comparisons and structured data. Lists help to separate individual features, steps or requirements.

It is important that the presentation provides an actual informational benefit.

A table should not be included just because it might have a positive citation factor. It should make information clearer and more understandable.

Question-oriented headings and direct answers

At least one question as an H2 heading achieves a factor of 1.4.

This is an interesting finding because many search queries and AI interactions are formulated as specific questions.

An article about website loading time could, for example, include the following heading:

“How does loading time affect user experience?”

The question should then be answered as directly as possible and explained in more detail later.

This has a practical advantage: the respective section is understandable even if viewed independently of the rest of the article.

A short summary at the beginning of an article (TL;DR) also achieves a factor of 1.3.

Update date and source references

Maintenance and traceability of content are also reflected in the results.

A visible update date achieves a factor of 1.3. At least ten external links also 1.3. A scientific source achieves 1.2.

This does not mean that as many external links as possible automatically produce better content.

The quality of sources is crucial. Statements should be supported where they are important for understanding or professional classification.

A visible update date is particularly useful for topics whose facts can change. These include, for example, technical standards, prices, legal requirements and current developments.

A date without actual content update is not a quality feature.

6. Not every schema and GEO measure brings an advantage

The SISTRIX study is also interesting because it does not only provide positive results.

Some frequently discussed measures show no measurable correlation with higher citation frequency.

A factor of 1.0 means that no difference was observed in the comparison examined.

This is an important distinction: structured data are not inherently an advantage for AI citability. It depends on what information they describe and in what context they are used.

The much-discussed llms.txt file also showed no measurable advantage in the study for AI citations.

No general ineffectiveness for every conceivable use case can be derived from this. However, a high priority for AI visibility cannot be justified based on these data.

The combination of different features counts

Another particularly interesting result is:

Pages that combine many of the examined optimisation features are cited 2.1 times as often as pages with few of these features.

Pages in the mid-range achieve a factor of 1.6.

This suggests that citability should not be reduced to a single page element.

An author box alone does not make a good specialist article. A table of contents does not replace well-founded information. And Schema.org cannot compensate for missing content.

Only the interplay of understandable content, traceable authorship and a sensible information structure creates a convincing overall picture.

Important: The factors of individual features must not simply be multiplied. The combined factor of 2.1 comes from a separate evaluation of several features.

7. Differences between AI systems and website types

Not all AI systems weight the examined features equally.

SISTRIX therefore also looked at differences between the three platforms examined.

ChatGPT

For ChatGPT, question-oriented content and tables stand out in particular:

• Title in question form: factor 1.9
• Table with header row: factor 1.8
• Question as H2 heading: factor 1.6

This speaks for a clear presentation of questions and answers as well as well-structured information.

Google AI Mode

In AI Mode, the basic structural features stand out:

• Table of contents: factor 1.8
• At least five H2 headings: factor 1.8

Here, a comprehensible structure and good information architecture apparently pay off.

Google AI Overviews

For AI Overviews, there is a particularly clear correlation with Article schema:

• Article schema: factor 2.0
• Product schema: factor 0.5

Informative content is more represented in this study than pages primarily focused on sales.

These results are not fixed rules for the respective platforms. Rather, they show that good information architecture should consider several requirements simultaneously.

Website type also plays a role

The study also shows that certain features vary depending on the type of website.

Examples:

• For editorial offerings, pages with a table of contents are cited 2.5 times more often.
• A visible update date achieves a factor of 1.9 there.
• For shops, a visible update date is associated with fewer citations (0.7).
• For SaaS providers, Product schema achieves a factor of 1.7.

This is an important practical hint: GEO optimisation should not work with a universal checklist.

A product page requires a different information architecture than a specialist article, an editorial offering or a corporate website.

One limitation remains: The study examined ChatGPT, Google AI Mode and Google AI Overviews. Perplexity, Claude and the standalone Gemini Chat were not part of the evaluation. Therefore, transfer to these systems is not readily possible.

8. What does this mean for practical GEO optimisation?

Concrete measures for your own website can be derived from the results.

Not every element should be forcibly included on every page. A specialist article requires a different structure than a product page or a corporate presentation.

The following checklist offers guidance.

Author information and trust

• Clearly name the author
• Add author photo and brief professional bio
• Link author profile sensibly
• Present relevant qualifications and experience
• Provide sources for important statements

Structure and information preparation

• Use table of contents with jump links
• Use clear H2 and H3 headings
• Consider concrete questions as headings
• Prepare complex information in tables
• Use lists for steps and enumerations
• Formulate important answers as directly as possible
• Add a summary for extensive articles

Technical implementation and maintenance

• Implement Article schema correctly
• Use FAQPage schema for genuine FAQ content
• Link author information structurally
• Mark publication and update dates correctly
• Check content and sources when relevant changes occur

This list is not a guaranteed catalogue of measures for more AI citations. It combines the results of the SISTRIX study with sensible editorial and technical quality requirements.

What should you not do?

The results should not encourage inflating websites with artificial elements.

In particular, these approaches are not advisable:

• Create author boxes without actual expertise
• Fill tables of contents with redundant headings
• Create FAQ sections just for additional schema markup
• Artificially extend texts to a certain word count
• Adjust update dates without real content changes
• Use structured data that does not match visible content

Google explicitly recommends user-oriented content, traceable authorship and genuine content quality instead of measures aimed solely at search engines.

9. Rankmio: Consider citability already during content creation

The results show that GEO optimisation does not start only when a text is published.

Already during content creation, consideration should be given to how authorship, structure and technical markup are implemented.

In the Rankmio Content Studio, various page elements can be integrated into templates for this purpose.

These include, among others:

• Author boxes with individual details
• Tables of contents with jump links
• FAQ sections
• Quizzes
• Structured data such as Article and FAQPage

When an author is selected during content creation, the appropriate author box can be automatically inserted at the designated position.

This reduces editorial effort and ensures that recurring requirements do not have to be implemented anew for each article.

However, what matters is not the number of elements built in, but their meaningful use in terms of content.

An author box should convey actual expertise. A table of contents should reflect the structure of the article. Structured data must match the visible content.

Exactly this connection of content quality, technical implementation and control is interesting for a sustainable GEO workflow.

10. Conclusion: Citability begins with good content and clear structure

The SISTRIX study with 96,418 URLs provides an important data basis for the current GEO discussion.

Especially the results on tables of contents (factor 1.7), Article schema (1.6) and author boxes with photo (1.4) show that certain on-page features are clearly associated with citation frequency by AI systems.

However, the study does not prove that retrofitting these elements automatically leads to more citations.

Nevertheless, sensible editorial consequences can be drawn from this:

If you want your content to be considered as a source for AI answers, you should not focus solely on keywords and text length.

Traceable authorship, clear information architecture, well-prepared answers and clean technical markup also belong to this.

It is not about ticking off as many GEO checklists as possible. It is about creating content that is understandable for people and clearly classifiable for technical systems.

Good content must not only be found. It must also convince as a source.

11. Use Rankmio elements

The described elements can also be implemented practically. In the Rankmio Content Studio, author boxes, tables of contents, FAQ sections and quizzes can be integrated as freely configurable elements in templates.

When an author is specified during content creation, the appropriate author box can be automatically inserted at the designated position in the article. This allows editorial standards to be considered already when creating new content.

First publication: 09.11.2026

FAQ: Frequently asked questions about AI citability

Sources and further information

1. SISTRIX: GEO Myths in Data Check: What 96,418 URLs reveal about AI Citations
2. Google Search Central: Creating helpful, trustworthy, user-focused content
3. Google Search Central: Introduction to structured data
4. Google Search Central: AI features and your website
5. Google Search Central: Guide to Google Search Ranking Systems
6. rankmio - All template elements. From author box to quiz