Triassic Times AI-use disclosure

Hello hello, in 2026, it’s probably a good idea to offer some writing transparency. And tell you what we use AI for on this website, and what we do ourselves.

We started this publication because we really like reading and writing about fossils, paleontology and dinosaurs. But we didn’t realise how much design and website management goes into it (WordPress can be dictatorial about SEO efficiency), and it’s damn near impossible to design a website without any prior experience, despite what anyone says. So we use AI assistants to help run Triassic Times. AI assistants are basically just that – assistants.

Broadly speaking, the human-AI role separation is this:

  1. We (as in, the people behind Triassic Times) do all of the story selection, writing, editing, and tweeting. And we write all our own jokes. Always.
  2. We have no interest in keyword selection, URL linking to internal pathways, excerpt extraction, slug selection, writing out tags, spam detection, character limit restriction, changing passive voice to active voice so that the SEO score improves, making sure that no sentence is more than 20 words long because “readers don’t read long sentences”, making sure no paragraph is more than 4 sentences long because “readers don’t read long paragraphs”, checking copyright status of the photos we use, and other such mindless writing-related admin. AI does that.
  3. None of us have science degrees, and we are writing for people who also don’t have science degrees. We write our articles based on the original papers (or their abstracts) by the published scientists, paleontologists and researchers, instead of paraphrasing stories written by other much better websites/magazines. AI helps us understand the paper and summarise the findings, methodologies, analysis and conclusions of the authors of the original scientific papers (which we always link in our article). We then write the story on the basis of this understanding. And we do two layers of fact-checking, one by ourselves, and one by the AI assistant.

The below describes our current workflow in more detail – AI assistants keep improving, and our budget to afford the good ones keeps changing, so this disclosure document will be a dynamic document. This disclosure was itself researched and drafted with AI assistance (lol).

What the AI assistants do

  • Run a sweep of all the scientific journals to find new articles – We get a daily brief of new papers that have been published. We select the stories that we find interesting or fun. There’s clear biases here. We might miss stories that have wider importance than we thought, or have much lower value than we thought. We just found them interesting, so we decided to write about the stories that we decided to write about.
  • Summarise the article – An AI Assistant summarises the paper in words that we understand. This is an iterative process. We fear that the AI Assistant still thinks very highly of us, and there’s lots of technicalities (particularly when the skeletal remains are being described, or when phylogenetic relationships are being explained) where we get lost, and then the AI Assistant has to explain/simplify further. We then write the articles on the basis of this summary. Sometimes we like the way a particular concept is explained by the AI Assistant, and we retain that language.
  • Fact-checking – The AI Assistant runs a claim-by-claim check against the available sources. Assistants check quotations, attribution, dates, measurements and the strength of conclusions. They flag unsupported claims, conflicting evidence and things they could not verify. Some examples:
    • In this story, we described the finding as the earliest sign of life. The AI Assistant flagged that that is too absolute, and added a qualifier – that the paper’s claim covers only “directly dated” signs of life. We changed it.
    • Here, we had written that environmental factors “triggered” the disease. The AI Assistant flagged that the paper said “influenced”, not “triggered”. It’s two completely different things. We changed that as well.
  • Help with images – While we have access to lawyers, we have strict instructions from them not to involve them unless we really need to. Publishing a story on the internet has zero value these days if there’s no accompanying image, and given that cameras weren’t a widespread technology when our subject species were alive, we basically have to scrounge the internet to find royalty-free copyright-safe images. We get AI to do that, and we get AI to sort out the attributions and legal compliances.
  • Prepare for publishing – Writing the article on Google Doc is one thing (good), publishing it on the website’s backend is something else altogether (bad). An AI Assistant formats approved copy in WordPress, adds source links, prepares excerpts and search descriptions, organises categories and tags, and places cleared images. It prepares unpublished drafts and previews before publication.
  • Makes hashtags for X – We prepare our own tweets for X, but we are now too old to sit and think of the best hashtags to use. AI does that.
  • Review and maintain the website – An assistant audits search-engine optimisation and makes recommendations. AI assistance implements specific website changes we want to the look and feel of the website, including homepage layout and navigation. These are things that we would never have been able to do ourselves in 65 million years.

What we do ourselves

We choose which stories to pursue and when to run them. We retain complete control of which story to publish, how to order the stories, how to categorise the stories and when to publish the stories.

We read, write, edit and rewrite article drafts. That includes the headline, opening, structure, explanations and plain language. Some AI wording survives, if we liked the AI version in the pre-summary. Remember, all jokes are human-generated. That’s why they’re crap.

We make the final editorial judgement on facts, uncertainty, attribution, images and corrections. An AI Assistant’s fact-check is advisory, by the way, we don’t always accept the suggestion. We decide whether a draft needs more research, should be changed or should not run.

We review what is proposed for publication and give the final written approval. We remain responsible for what Triassic Times publishes, including any mistakes by an AI Assistant.

Limits and safeguards

Nothing is to be published, posted or sent without human explicit written approval. Permission to prepare a draft or preview is not permission to make it public. Website changes also require human authorisation.

We link to sources and check primary papers where accessible. Some checks can only use an abstract, repository record or press release because the full paper is unavailable (we complain about the death of the freedom of knowledge in these instances).

The images reviewed are credited photographs, scientific figures, maps and reconstructions sourced from research repositories, Wikimedia Commons or public-domain material. Reconstructions and comparison images are labelled as such.

Errors and corrections

If you spot a factual error, a misleading image or a missing credit, please contact thetriassictimes@gmail.com. Please include the article link and the passage or image concerned. We are responsible for reviewing the issue.

Oh, and we have been using em-dashes since 2014. We’re not going to stop now. Em-dashes rule!