In short. ai.txt names several separately authored proposals whose formats, intended uses and publication statuses differ.

Definition: ai.txt proposals are separately authored formats for declaring how AI systems may use website content. Their authors propose different file locations, permission fields and processing methods, so the filename alone leaves the intended format ambiguous.

Key facts

As of 17 October 2026, the IETF Datatracker record lists Kayla Cardillo's AI.TXT as an active individual Internet-Draft with no formal standing in the IETF standards process.
Spawning's announcement of 30 May 2023 describes a root-level file whose permissions are checked when media is downloaded.
Guardian News & Media's draft paper of 9 August 2024 proposes permissions for individual pages and site sections.
Yuekang Li and coauthors' research preprint, submitted on 2 May 2025, proposes element-level rules and natural-language instructions.

What does the IETF Internet-Draft propose?

Cardillo's AI.TXT draft, published on 12 June 2026 with an intended Informational status, requests registration of /.well-known/ai.txt and /.well-known/ai.json. Its stated expiry date is 14 December 2026. An Internet-Draft is a working document that can change or expire.

The text format uses colon-separated fields. The draft's minimal example is:

# ai.txt
Spec-Version: 1.0
Site-Name: My Blog
Site-URL: https://myblog.com
Training: deny

The proposed production location uses HTTPS and the content type text/plain; charset=utf-8. Separate fields cover training, scraping, indexing and caching; agent blocks specify overrides for named agents. The draft describes the declaration as advisory and assigns access-control enforcement to servers.

How does Spawning's proposal differ?

Spawning's generator documentation describes machine-readable permissions for commercial text and data mining, meaning automated analysis of content. Its generator initially opts all content out; selecting allow changes the permission for the selected media type. The installation FAQ mentions both / and /.well-known/, while its browser verification example uses /ai.txt.

Spawning's launch announcement describes checking permissions at media-download time, including downloads through links already held in datasets. It presents Spawning's API as the channel that communicates those permissions to researchers and partners. The same announcement characterizes ai.txt as a convention for declaring permissions for text and data mining.

This differs from Cardillo's site-level fields such as Training and Indexing. A website implementation therefore needs to identify which author's format its intended reader supports.

What does the Guardian paper add?

Matt Rogerson's Guardian News & Media paper favors the existing robots.txt structure in a purpose-specific file. Its example requests exclusion of the entire site:

User-agent: * # any AI
Disallow: /   # disallow everything

The paper's example of finer control permits news home pages containing headlines while denying article content. It also suggests extensions for snippet length, attribution and training permissions. The Datatracker listing identifies this document as material for the IAB Workshop on AI-CONTROL. Its publication form is a workshop draft paper.

What does the research proposal add?

Li and coauthors' arXiv preprint introduces a domain-specific language, a language designed for a particular task. Here, that task is expressing rules for AI interaction with web content at the level of individual page elements.

The authors describe an editor with automatic XML generation, where XML is a structured markup format. They propose programmatic enforcement through XML and integration of natural-language instructions into prompts. Their reported evaluation consists of preliminary experiments and case studies, giving this proposal the status of a research approach.

FAQ

Which location should I use?

Follow the proposal supported by the intended reader. Cardillo's draft specifies /.well-known/ai.txt; Spawning's generator FAQ gives /ai.txt as its browser check.

Does publishing the file enforce my policy?

Cardillo's draft explicitly treats ai.txt as advisory and requires servers to enforce access controls independently. Spawning's generator terms likewise make effectiveness dependent on recognition and respect by developers and data scrapers.

Sources

Check how a page reads to answer engines. Open the GEO score analyzer