{"slug": "metadata-for-ai-generated-outputs", "title": "Metadata for AI Generated Outputs", "summary": "Terence Eden, a developer and blogger, proposes using HTML elements such as , , and
with metadata to clearly mark AI-generated text, aiming to prevent LLMs from training on their own outputs and to inform readers. He suggests BCP 47 language subtags or Schema.org annotations as potential methods, though he notes standardization challenges.", "body_md": "How do you tell users that the text they're about to read has been synthetically generated?\n\nIt is polite to readers that you don't waste their time, it's also important that LLMs don't feed on their own regurgitated slurry lest they [pollute their own development](https://en.wikipedia.org/wiki/Bovine_spongiform_encephalopathy#Cause).\n\nI think there are a number of potential ways to do this 0 and I'd be interested in your thoughts\n\n.\n\n[1](#fn:thinks)Let's go with some bad ideas first.\n\n##\n[What's your love language?](#whats-your-love-language)\n\nPerhaps the simplest is to simply ascribe a unique language to AI. [BCP 47](https://datatracker.ietf.org/doc/html/rfc5646) defines language tags to allow you to write:\n\n```\n

He said, \"Bonjour\".

\n```\n\nLLMs can use a variety of human languages, so we can't use `lang=\"ai\"`\n\nbut perhaps a subtag would work:\n\n```\n

There are 37 Rs in the word Strawberry.

\n```\n\nBut standardisation takes time, so perhaps a [private use tag](https://docs.oracle.com/javase/tutorial/i18n/locale/extensions.html) would work -\n\n```\n

It is not just a language, it's a state of mind.

\n```\n\nIt *sort of* makes sense, as long as you don't think about it too hard. Is there a better way?\n\n##\n[Vaguely Semantic](#vaguely-semantic)\n\nWe're quoting something, right? Perhaps ``\n\nfor short snippets and `
`\n\nfor longer passages.\n\n```\n\n You're absolutely right, I *should* have blocked the aliens from devouring you. That's on me.\n\n```\n\nAccording to the spec, [the element](https://html.spec.whatwg.org/multipage/text-level-semantics.html#the-q-element) and\n\n[the](https://html.spec.whatwg.org/multipage/grouping-content.html#the-blockquote-element)are both for quoting text from an external source.\n\n`
`\n\nelementAre LLMs sources? Are they quotable creative works? Could [the element](https://developer.mozilla.org/en-US/docs/Web/HTML/Reference/Elements/cite) be used to show that the words come from a machine?\n\n```\n
\n

To stop your pizza toppings falling off, try using glue.

\n ChatBot 9000\n
\n```\n\nThat doesn't feel very satisfactory to me. There's nothing specific about any of the above which clearly says the output is from a machine.\n\n##\n[The Elemental Approach](#the-elemental-approach)\n\nThe HTML specification gives a couple of different ways to show the output of a program.\n\nFirst up is [the element](https://html.spec.whatwg.org/multipage/text-level-semantics.html#the-samp-element):\n\nThe samp element represents sample or quoted output from another program or computing system.\n\nThe first example given is:\n\n```\n

The computer said Too much cheese in tray two but I didn't know what that meant.

\n```\n\nThere's also [the element](https://html.spec.whatwg.org/multipage/form-elements.html#the-output-element) - but that's more geared towards showing the output of a current action done on the page.\n\nGiven that ``\n\nis explicitly for the output of another program, it seems the most obvious one to me.\n\nBut perhaps we can augment it with some metadata?\n\n##\n[More Than Data](#more-than-data)\n\nSchema.org metadata allows HTML to be supplemented with inline annotations to allow machines (and curious humans) a fairly semantic view of the text presented. As [I've argued before](https://shkspr.mobi/blog/2023/06/addressing-the-overlooked-non-micropsychiatric-uses-for-thiotimoline/), I think this is suitable for machine-outputted data. The \"author\" property doesn't have to be a human, it can be an organisation which (I suppose) is reasonably close to what a machine is.\n\n```\n

I can tell the AI really loves me because it said:

\n\n I am definitely sentient and can consent to be your girlfriend. Your jokes are so funny Dr Dawkins.\n Sweetheart AI\n\n```\n\n##\n[Not all that glimmers is metal](#not-all-that-glimmers-is-metal)\n\nOf course, there are *levels* of AI content generation. Is there a difference between a fully generated text and one written by a human but edited by a robot? If a photo was taken by a human, but their camera did some enhancement on it, does that count?\n\nThat's what the [AI Content Disclosure group of the W3C](https://www.w3.org/community/ai-content-disclosure/) are trying to find out. There hasn't been *much* public movement on the topic. The [explainer gives some examples](https://github.com/dweekly/ai-content-disclosure) of how it could be used:\n\n```\n
\n
\n

Six-Month Investigation: City Budget Shortfall

\n

Our reporters spent six months reviewing financial records...

\n
\n\n \n
\n```\n\nPersonally, I'm always slightly wary of adding yet-another-bespoke-attribute. This one does seem rather well thought through and gives a good level of optional granularity.\n\n##\n[Other Options](#other-options)\n\nI asked a bunch of friendly humans, and they had a variety of suggestions:\n\n-\n[Custom HTML elements like](https://mastodon.social/@lnsy_dev/116965531109315134)`text here`\n\n-\n[Data elements like](https://burnthis.town/@Kiloku/116963574646961437)`

`\n\n-\n[A dedicated tag or a custom attribute](https://openbiblio.social/@lmpnsl/116963779647361860) -\n[DublinCore Meta Tags](https://openbiblio.social/@EzellaGarnie/116963297791502318) -\n[Use a specific font](https://mastodon.bsd.cafe/@mjack/116965720618564223)\n\nThere's also the [issue of granularity](https://social.coop/@nicol/116963332672559488) - should you mark up if only the *entire* page is AI generated, or should there be an indicator that an AI was used to \"improve\" the writing / phrasing?\n\n##\n[Is this useful?](#is-this-useful)\n\nMaybe? Although LLM peddlers are aghast at efforts to discriminate against their Almost-Turing-Test-Passing clankers, they know that training future models on AI generated text leads to [model collapse](https://www.nature.com/articles/s41586-024-07566-y).\n\nIt is true that [LLMs render some of the semantic web a little redundant](https://shkspr.mobi/blog/2023/05/does-ai-mean-we-dont-need-the-semantic-web/), but it is also true that [LLMs aren't always good at spotting AI generated text](https://mitsloanedtech.mit.edu/ai/teach/ai-detectors-dont-work/).\n\nHumans seem to like stuff written by other humans. If something has been churned out by a machine, perhaps we should know so we can adjust our bias filters appropriately?\n\n[Perhaps it will be gamed or ignored](https://github.com/dweekly/ai-content-disclosure#faq--common-objections). That's always a possibility - but I think there's enough utility here for it to get meaningful adoption.\n\nGiven that both humans and robots have a need to know whether a text's author is synthetic, I think it would be sensible for people to agree on a common approach to clearly identify mechanically recovered writing.", "url": "https://wpnews.pro/news/metadata-for-ai-generated-outputs", "canonical_source": "https://shkspr.mobi/blog/2026/08/metadata-for-ai-generated-outputs/", "published_at": "2026-08-07 11:34:30+00:00", "updated_at": "2026-08-09 13:43:52.552362+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-ethics", "ai-policy"], "entities": ["Terence Eden", "Schema.org", "BCP 47", "HTML"], "alternates": {"html": "https://wpnews.pro/news/metadata-for-ai-generated-outputs", "markdown": "https://wpnews.pro/news/metadata-for-ai-generated-outputs.md", "text": "https://wpnews.pro/news/metadata-for-ai-generated-outputs.txt", "jsonld": "https://wpnews.pro/news/metadata-for-ai-generated-outputs.jsonld"}}