JATS Guide
Index Toggle Dark/Light/Auto mode Toggle Dark/Light/Auto mode Toggle Dark/Light/Auto mode Back to homepage
  1. JATS Guide
  2. /
  3. Topics
  4. /
  5. Attribution Metadata and AI Attribution

Attribution Metadata and AI Attribution

  • 1 Apr 2026: Created
Draft
This is a draft. Please send comments via this comment form.

Table of Contents

Introduction

Attribution metadata can identify the creator of specific content within an article. In publishing workflows, some content within an article, such as image descriptions for accessibility, might be provided by an author or another party such as the publisher or vendor, and generative AI may be used in its creation. It is important to be able to identify the origin of such content through attribution. Attribution metadata can also be used in reporting and quality control.

Attribution metadata should be included for image descriptions in <alt-text> or <long-desc> (see Accessibility) as these will frequently be added during a production process using some portion of human effort and/or generative AI.

If needed, it is possible to add attribution metadata to any piece of content within an article. These tagging guidelines for attribution metadata have been created because of a need for clarity regarding the use of generative AI for improving the accessibility of journal articles. While the initial scope for attribution metadata includes AI attribution, attribution metadata may also be useful in other scenarios to record metadata about origin of specific content within an article.

This tagging requires JATS version 1.4. In earlier versions of JATS, the <custom-meta-group> element is optional but not repeatable in <article-meta> and does not have the @content-type attribute, and the @lang-source attribute is not available.

To avoid confusion, attribution metadata in this context is different than other forms of attribution, such as attributions that appear in text and should be tagged using <attrib>, or contributor metadata that should be tagged using <contrib>.

Attribution Metadata

Attribution metadata should be tagged using the <custom-meta-group> element with attribute content-type="attribution". There can be more than one <custom-meta-group content-type="attribution"> in an article, and each unique attribution will need its own <custom-meta-group> for attribution metadata.

The location of <custom-meta-group content-type="attribution"> should typically be within <article-meta>. Attribution metadata may be placed nearer to the relevant content if permitted by the DTD. For example, attribution metadata that is for a piece of content within a <sub-article> may be placed within the <front-stub> of the same <sub-article>.

The <custom-meta-group> element contains <custom-meta> elements, which hold name/value pairs using the <meta-name> and <meta-value> elements to hold a name and a value, respectively.

Attribution metadata should always contain the following information:

  • content - identify the specific content to which the attribution relates
  • source - describe the origin of the content (e.g. author, publisher, vendor)

Additionally, the following AI attribution metadata should be included when generative AI tools are used in the creation of content.

  • model_id - identify the language model as specifically as possible
  • ai-use - describe the activity performed using generative AI
  • human-review - describe who was responsible for reviewing and possibly editing or approving generated content

The attribute lang-source="machine" should be added to the element that contains content that was created using generative AI. For example, when generative AI is used to create an image description, a lang-source="machine" attribute should be added to the <alt-text> or <long-desc> element.

In earlier versions of JATS prior to version 1.4, a content-type="machine-generated" attribute can be added to the <alt-text> or <long-desc> element to label an image description that has been generated by machine.

content

In attribution metadata, the content value identifies the specific content to which the attribution relates. The <meta-value> element should contain one or more <xref> elements whose @rid attribute holds a reference to the @id attribute of the content. The <xref> element’s @ref-type attribute holds the name of the element being referenced. The content that is being referenced must have an @id attribute.

For example, this content reference:

<custom-meta>
    <meta-name>content</meta-name>
    <meta-value>
        <xref ref-type="alt-text" rid="F0001-alt-text"/>
    </meta-value>
<custom-meta>

points to this <alt-text> element for a figure using its @id:

<fig id="F0001">
    <alt-text id="F0001-alt-text">
Note that, in the above example, the value of @rid points to the @id on the <alt-text> element to identify the alt text specifically, not the entire figure, as the content for which the attribution metadata applies.

If the same attribution metadata applies to more than one piece of content then the content <meta-value> can identify all content that is relevant by using more than one <xref> element or by having more than one @id value in @rid.

For example, this content reference:

<custom-meta>
    <meta-name>content</meta-name>
    <meta-value>
        <xref ref-type="alt-text" rid="F0001-alt-text"/>
        <xref ref-type="long-desc" rid="F0001-long-desc"/>
    </meta-value>
<custom-meta>

points to the <alt-text> and <long-desc> elements in a figure using the @id attribute of each element:

<fig id="F0001">
    <alt-text id="F0001-alt-text">...</alt-text>
    <long-desc id="F0001-long-desc">...</long-desc>

In this next example, this content reference uses one <xref> element whose @rid contains a space-separated list of @id values:

<custom-meta>
    <meta-name>content</meta-name>
    <meta-value>
        <xref ref-type="alt-text" rid="F0001-alt-text F0002-alt-text"/>
    </meta-value>
<custom-meta>

pointing to the <alt-text> elements in two different figures:

<fig id="F0001">
    <alt-text id="F0001-alt-text">...</alt-text>
    ...
</fig>
<fig id="F0002">
    <alt-text id="F0002-alt-text">...</alt-text>
    ...
</fig>

The options shown in the two preceding examples (using more than one <xref> element and using a list of @id values in @rid) can be used together.

source

In attribution metadata, the source value describes the creator or origin of the content.

The <meta-value> element should contain a value from this list:

  • author - for content supplied by the author
  • editor - for content supplied by the editor
  • translator - for content supplied by a human translator
  • publisher - for content supplied by the publisher
  • vendor - for content supplied by the vendor
  • unknown - for when content source is not known (for example, when processing older content)

For example:

<custom-meta>
    <meta-name>source</meta-name>
    <meta-value>publisher</meta-value>
</custom-meta>

Note: This list of values is derived from @lang-source and Taylor & Francis eBook Specification, Version 4.4.1, §6.7.7.5.1.

model_id

In attribution metadata, the model_id value identifies the language model and should be included when generative AI tools are used in the creation of content.

The model_id value should identify the language model as specifically as possible. For example, “gpt-5-mini-2025-08-07” is a specific release of “GPT-5 mini”, which is more specific than “GPT 5”. Ideally, this value should be a unique, machine readable identifier that can be used programmatically to reference a specific model.

For example:

<custom-meta>
    <meta-name>model_id</meta-name>
    <meta-value>gpt-5-mini-2025-08-07</meta-value>
</custom-meta>

Currently, there is no standard identifier system for language models, however several software tools and platforms are using model_id as a way to uniquely identify language models.

If more than one language model needs to be identified, such as for a combination of models used in a workflow, then each model can be identified with its own <custom-meta> element with <meta-name> model_id.

ai-use

In attribution metadata, the ai-use value describes the activity performed using generative AI, and should be included when generative AI tools are used in the creation of content.

When AI is used in creating image descriptions for accessibility, the <meta-value> element should contain description. For example:

<custom-meta>
    <meta-name>ai-use</meta-name>
    <meta-value>description</meta-value>
</custom-meta>

For other uses, the <meta-value> element should hold a suitable description. If possible, the value may be selected from a controlled vocabulary or taxonomy, such as GAIDeT or NIST AI Use Taxonomy, and the @vocab and @vocab-identifier attributes may be added to the <custom-meta> element to identify the vocabulary or taxonomy.

If more than one ai-use value is required, then each value can be placed in its own <custom-meta> element with <meta-name> ai-use.

human-review

In attribution metadata, the human-review value describes who performed or was responsible for reviewing and possibly editing or approving generated content, and should be included when generative AI tools are used in the creation of content.

The <meta-value> element should contain a value from this list:

  • author - the author
  • editor - the editor
  • translator - a human translator
  • publisher - the publisher
  • vendor - the vendor
  • unknown - it is not known if human review was performed (for example, when processing older content)
  • none - no human review

For example:

<custom-meta>
    <meta-name>human-review</meta-name>
    <meta-value>author</meta-value>
</custom-meta>

Examples

Several examples of attribution metadata are provided below. These examples are not intended to be an exhaustive list of possible scenarios.

Alt text provided by the author

This example shows attribution metadata for alt text provided by an author.

<custom-meta-group content-type="attribution">
    <custom-meta>
        <meta-name>content</meta-name>
        <meta-value>
            <xref ref-type="alt-text" rid="F0001-alt-text"/>
        </meta-value>
    </custom-meta>
    <custom-meta>
        <meta-name>source</meta-name>
        <meta-value>author</meta-value>
    </custom-meta>
</custom-meta-group>

The above attribution metadata points to this alt text:

<fig id="F0001">
    <label>Figure 1.</label>
    <caption>
        <p>Study area &ndash; locations of the surveyed public housing in Amman.</p>
    </caption>
    <alt-text id="F0001-alt-text">A map of Amman showing the geographic distribution of surveyed public housing sites across different districts.</alt-text>
</fig>

Alt text created by a vendor using AI and reviewed by the author

This example shows attribution metadata for alt text that was created by a vendor using AI and then reviewed by the author.

<custom-meta-group content-type="attribution">
    <custom-meta>
        <meta-name>content</meta-name>
        <meta-value>
            <xref ref-type="alt-text" rid="F0005-alt-text"/>
        </meta-value>
    </custom-meta>
    <custom-meta>
        <meta-name>source</meta-name>
        <meta-value>vendor</meta-value>
    </custom-meta>
    <custom-meta>
        <meta-name>model_id</meta-name>
        <meta-value>claude-opus-4-6</meta-value>
    </custom-meta>
    <custom-meta>
        <meta-name>ai-use</meta-name>
        <meta-value>description</meta-value>
    </custom-meta>
    <custom-meta>
        <meta-name>human-review</meta-name>
        <meta-value>author</meta-value>
    </custom-meta>
</custom-meta-group>

The above attribution metadata points to this alt text:

<fig id="F0005">
    <label>Figure 5.</label>
    <caption>
        <p>Appliance usage pattern.</p>
    </caption>
    <alt-text id="F0005-alt-text" lang-source="machine">A bar chart showing the frequency of use of major household appliances among surveyed families.</alt-text>
</fig>

More than one group of attribution metadata

This example shows an article that contains more than one group of attribution metadata. In the first group of attribution metadata, the author is identified as the source of the alt text for Figure 3. In the second group of attribution metadata, the alt text for Figure 4 is attributed as being created using AI and reviewed by the publisher.

<custom-meta-group content-type="attribution">
    <custom-meta>
        <meta-name>content</meta-name>
        <meta-value>
            <xref ref-type="alt-text" rid="F0003-alt-text"/>
        </meta-value>
    </custom-meta>
    <custom-meta>
        <meta-name>source</meta-name>
        <meta-value>author</meta-value>
    </custom-meta>
</custom-meta-group>
<custom-meta-group content-type="attribution">
    <custom-meta>
        <meta-name>content</meta-name>
        <meta-value>
            <xref ref-type="alt-text" rid="F0004-alt-text"/>
        </meta-value>
    </custom-meta>
    <custom-meta>
        <meta-name>source</meta-name>
        <meta-value>vendor</meta-value>
    </custom-meta>
    <custom-meta>
        <meta-name>model_id</meta-name>
        <meta-value>claude-opus-4-6</meta-value>
    </custom-meta>
    <custom-meta>
        <meta-name>ai-use</meta-name>
        <meta-value>description</meta-value>
    </custom-meta>
    <custom-meta>
        <meta-name>human-review</meta-name>
        <meta-value>publisher</meta-value>
    </custom-meta>
</custom-meta-group>

The above attribution metadata points to the alt text on these figures:

<fig id="F0003">
    <label>Figure 3.</label>
    <caption>
        <p>Respondents&rsquo; thermal comfort in their houses.</p>
    </caption>
    <alt-text id="F0003-alt-text">A chart showing the percentage of respondents who reported feeling comfortable, neutral, or uncomfortable within their homes.</alt-text>
</fig>
<fig id="F0004">
    <label>Figure 4.</label>
    <caption>
        <p>Descriptive statistics on renovations and their motivation, cooling and heating months, heated and cooled spaces, thermal comfort strategies, and occupancy patterns.</p>
    </caption>
    <alt-text id="F0004-alt-text" lang-source="machine">A grouped bar chart summarising data on renovation reasons, seasonal heating and cooling practices, comfort strategies, and occupancy schedules.</alt-text>
</fig>
gdoc_arrow_left_alt Article Versioning Author Contributions and CRediT gdoc_arrow_right_alt