This page sets out whether and how automated and artificial intelligence (AI) tools are used across this website, eve.gd. It applies to the whole site: blog posts, essays, my biography, the publications list, and any other content published here.

What is and is not written by AI

The textual content on this site is 100% human-written. I write my blog posts, essays, and biographical pages myself. These are not generated by AI tools. I do not draft with AI, produce with AI, and I do not use AI correction tools.

I do use voice assistance tools (automated transcription) as disability aids, particularly when my rheumatoid arthritis makes typing too painful. These tools sometimes use “AI” (machine learning) systems for their transcription (such as the NVIDIA Parakeet nvidia/parakeet-tdt-0.6b-v2 model).

There are three significant pieces of automation on the site. My publications list is populated automatically from the Birkbeck Institutional Repository (BIROn) through an automated feed, rather than being typed out by hand. The code that generates this feed and that interacts with the repository is called eprintsToCV and the code is openly available under a GNU Affero General Public License 3.0 license. I wrote this code by hand (and it’s not the neatest, but it does the job!). This is a main place where an automated process contributes to what you see. The second is that the references list at the end of each post on the site, where I have made reference to external objects and writings, is generated by an automated process. The third is the Now page, which is assembled automatically every time the site rebuilds: it gathers my most recent posts, my latest short thought and Today-I-Learned entry, and my current listening statistics, and slots them into place. No text on that page is generated by AI — everything there is either my own writing drawn from elsewhere on the site, factual metadata (dates, titles, track names), or the hand-written sections I add to the page directly.

The Site & Policies pages listed in the footer (e.g. this page) were created from an amalgam of other policies on other sites around the web. AI may have been used on those pages.

The HTML and CSS code for this site were partially coded by hand and partially coded with AI assistance (Anthropic Claude Opus 4.8 and Claude Fable 5.0 driven by Claude Code). The build process for this static site has been crafted using the same models and tools. Automated repository depositing was also developed with the assistance of these models and tools. Metadata for posts and images (for accessibility) has been enriched using these same models and tools. For instance, the FAIR metadata that is embedded in each more recent post was sometimes (but not always) generated with AI assistance and then checked by me, manually. Categories and tagging of posts was aided by AI toolsets (same models, same tools).

If I use AI assistance for anything else published on the site, I will disclose it. I have no plans whatsoever to abandon writing by hand and brain. I strongly believe that we need to continue asserting human voices, amid the cacophony of “slop”. If we stop writing and thinking for ourselves, then what is the point? So: if I have not disclosed AI involvement on a given page, that page or post was written by me.

Scope, limitations, and safeguards

The specific tasks that may involve automation are: retrieving and formatting my list of publications from BIROn; generating the reference lists at the foot of posts; and assembling the Now page from the site’s own content streams. Everything else is human work.

Automated output is checked by me before it appears on the site. The publications feed in particular is reviewed before it goes live, so that errors in metadata, classification, or formatting can be caught and corrected. I wrote the code that pulls this from BIROn and formats it into a PDF CV myself, by hand, back in the days before AI.

Oversight, review, and accountability

Automated and AI outputs on this site, including the auto-populated BIROn publications feed, are reviewed by a human before publication. That human is me, Martin Paul Eve.

The review process for the CV, for instance, works like this: the automated feed proposes entries drawn from BIROn; I review those entries for accuracy and appropriateness; and only reviewed content is published. I am accountable for all content on this site, whether written by hand or drawn from an automated feed.

How the publications feed is ordered and audited

The publications list is ordered by date, with the newest items shown first. Each entry’s classification is auditable through the BIROn repository itself. In the list, a green download link indicates a green open-access accepted manuscript, and a gold link indicates a gold open-access version. Because the underlying records live in BIROn, the selection and the green-versus-gold classification of any item can be checked against the repository. There are sometimes some problems with the metadata, such as published items remaining in “In Press” status. I fix these as I notice them.

Coding of the site

Large language models have been used for programming in creating the structure and design of the site. In particular, I have used LLMs (Anthropic Claude Opus 4.8) to improve page metadata. As above, none of the blog content or other pages were written by AI.

Reference lists on posts

The “References” section that appears at the foot of blog posts is generated through an automated process. External links in each post were extracted and checked to see whether they still resolve; citation metadata (titles, authors, dates, venues) was then harvested from the linked pages and from DOI registration records, with large language models assisting in labelling and normalising that metadata. The text of the posts themselves was not altered by this process.

Because much of this metadata was gathered automatically, reference lists may contain errors or incomplete entries, and some references can show only a link. These lists are provided purely as assistance for finding the items mentioned within a post; they are not a hand-checked scholarly apparatus, and each list carries a note to that effect.

human.json

This site publishes a machine-readable declaration of human authorship at /human.json, following the human.json specification. That file also lists “vouches”: other sites whose content I believe, from personal knowledge of their authors, to be human-written. The claim that human.json makes about this site should be read alongside this policy: the writing here is mine, while the site’s build tooling (including the small script that maintains the human.json file itself) was produced with AI assistance, as disclosed above.

Last reviewed: July 2026