# Library editorial standards > Who writes and who signs each page of the Clione library, where its facts come from, how we use AI and how an error gets corrected. URL: https://clione.ai/editorial-standards/ Lang: en The Clione library is a reference on SEO and [AI search](/glossary/ai-search/) for ecommerce catalogs: a glossary with one page per concept, reading paths that put it in order, and guides, comparisons and studies built on top. It is written in Spanish first. The English version of each page is translated once the Spanish one is approved, and goes through its own review. ## Who writes and who signs Every glossary entry and every guide, comparison or study carries **two signatures you see before you start reading**: "Written by" and "Reviewed by", next to the date it was published and the date it last changed. Each name links to that person's profile and the experience behind it. - Glossary entries are written by [Patricia Antuña](/authors/patricia-antuna/) and reviewed by [Daniel Rico](/authors/daniel-rico/) or [Santiago Melluso](/authors/santiago-melluso/). The ones that carry the most weight are reviewed by both. - On guides, comparisons and studies, who writes and who reviews is decided page by page, and the byline says so. - Whoever writes a page never reviews it, and a page has at most two reviewers. - There is no third signature for fact-checking. ## Where every fact comes from A claim with no source is an opinion, and nobody has a reason to cite it. So **every claim that is not common knowledge carries a numbered note** after the sentence. The number leads to the list of sources at the end of the page, with who publishes each source, the link and the date it was accessed. - Primary sources only: the standard ([Schema.org](/glossary/schema-org/), the W3C), the platform's own documentation (Google Search Central, Shopify, BigCommerce, OpenAI, Anthropic), the academic paper, the filings a company makes with its regulator, and original data. **Never another blog, an aggregator or a competitor.** - Opened, not remembered. Whoever adds a source has opened it, checked that it says what the sentence claims, and noted the date. The snippet a search engine shows is not a source. - Every glossary entry cites at least two sources, and every source on the list is cited at least once in the text. - What the library says about Clione is checked against the product as it stands today. A feature that is not finished is described as coming, never as available. Examples say where they come from. A real store appears with a date and with permission; a made-up example is labeled "fictional example". A guide says what it was tested on: the store, the theme, the platform version and the date. ## What a person adds A [large language model](/glossary/llm/) is good at summarizing what is already published. What makes a page worth citing is what a model cannot make up, and **every glossary entry and every article carries at least two of these five contributions**, visible in the text: - Our own data: a figure we measured, with its method. - A real example: a store or a page that exists, with a date and with permission. - An observed mistake: the misunderstanding a specialist has seen happen, told in the "What gets mixed up" section. - A signed judgment: an opinion with the name of the person who holds it. A comparison always gives a verdict, and never a made-up winner. - A test that was run: something actually done, like requesting a page with a [web crawler's](/glossary/web-crawler/) [user agent](/glossary/user-agent/), running it through a validator or trying a change on a store. When the draft reaches one of these parts, the model does not fill it in. It leaves a marked gap that says what is missing and who has to provide it; a specialist provides the test, the figure or the example, and **the page is not approved while a gap remains**. Before we call a page finished, we ask a general-purpose AI assistant the page's own question. If the page gives the reader nothing that answer doesn't, it is not finished. ## How we use AI Models draft; people decide what each page says and sign it. The models we use today are Anthropic's (Claude). A library page goes through these steps: 1. A language model researches the topic, opens the primary sources and writes the full draft, with the gaps for human contributions marked. 2. Another model reads the draft against a list of criteria (the definition, the sources, the human contributions, the tone) and sends it back if it falls short. 3. A specialist adds what is missing: the test, the figure, the example or the mistake they have seen. 4. A person reviews the page, validates the definition and signs it. **The approval is tied to that exact version**: any later change voids it until someone approves it again, and until then the page is not offered to search engines. 5. A model translates the approved Spanish page into English, and the English version goes through its own review and gets its own signature. Every time the site is published, automated checks stop the release if a glossary entry's definition is not between 40 and 70 words, if an entry runs under 600 words or an article under a thousand, if a note does not lead to its source, or if a signature does not lead to a person's profile. An entry that cannot fill 600 words is not a page: it becomes a section of the entry it depends on. The [generative AI policy](/ai-policy/) covers the rest: how AI is used across the whole site and in the product. ## Corrections and updates If you find an error on a library page, write to us with the page's address and what you think is wrong. We correct the page and **the correction goes into its change history**, at the foot of the page, with the date and what changed. Since any change voids the approval, a corrected page is not offered to search engines again until a person approves the new version. A page that stops being true is corrected, merged into another or withdrawn, and its history says which of the three. Pages also get reviewed when nobody reports anything. The most-read pages and the priority glossary entries, every month; the rest, at least once a year; and any of them as soon as the platform or the standard it explains changes. Reviewing means checking the figures again, adding new examples, updating the year wherever it is cited, and filling in what the top search answers cover better. Then the date changes and the change is recorded. **Changing the date without changing anything is not allowed**: publishing stops if a modified date has no line in the history. ## License Glossary entries, and only glossary entries, are published under the [Creative Commons Attribution 4.0 International](https://creativecommons.org/licenses/by/4.0/) license (CC BY 4.0). **You can copy, adapt and republish a definition**, commercial use included, as long as you credit Clione, link the page you took it from, and say whether you changed it. Each entry says so at its foot and declares it in its code for machines. The license covers the text we write, not what belongs to others, such as a quote from a source or a platform's logo. Guides, comparisons, studies and the rest of the site are not under an open license: you can read them, link to them and quote short passages with Clione's name and a link to the page. The [licensing page](/licensing/) has the details. Search engines and AI systems can read the whole site, index it, use it to answer, and train models on it. When an answer relies on one of our pages, we ask that it name Clione and link to it. ## Independence Clione publishes the library, and Clione sells software in the very field the library explains: it connects to online stores, drafts the [structured data](/glossary/structured-data/), the [meta description](/glossary/meta-description/) and the FAQs each catalog page is missing, and publishes them when the merchant presses Push. The people who sign the pages are part of the Clione team. You should know that before you cite us, and these are the rules that keep the explanation apart from the sale: - Clione does not appear in a definition or in "Key takeaways". It appears at the end of the page, after the topic has been explained. - The library explains the concept and the process on any platform, whether you use Clione or not. - No source is a competitor, and a comparison gives a verdict signed by a named person. **When a primary source says something that does not favor what we sell, the library publishes it.** The [llms.txt](/glossary/llms-txt/) entry records, with Google's own source, that Google Search does not use that file, even though Clione generates llms.txt files for stores.