Editorial Process

Every news summary on Interestana is written by an AI model from one item published by another news organisation, company or research group, and it is published automatically. No person reads, edits or approves an article before it appears, and nobody fact-checks it afterwards as a matter of routine. The byline “Interestana AI Editorial” names that automated process, not a newsroom.

We summarise an item only when it is a company’s or institution’s own announcement, or comes from a publisher whose terms, as far as we have found, do not forbid reuse of its reporting. Most of the news outlets we follow have terms that do forbid it. Their reports are never sent to an AI model and never summarised: Interestana shows only the publisher’s own headline, its name, the date and a link to the original report, and asks search engines not to index that page. So most of the stories in our feeds are headlines that link out, and a smaller number are summaries.

Every article names and links to the original report. That report is the authority: read it before you rely on, repeat or act on anything you read here.

Each article goes through the same steps:

1. Fetch. Twenty topic agents each read a hand-picked set of RSS feeds every hour, and a twenty-first, the Creator Economy desk, reads its own. The desk’s stories are listed at /creators/news rather than on the home feed, /news, the topic and interest pages, the site RSS feed or the day-by-day archive. Items more than 48 hours old, items already covered and any article we have taken down are dropped.
2. Scope gate. Before any AI is used, an item goes on only if its headline names a subject Interestana covers (for the AI and cybersecurity agents, its feed description may name it instead). The Creator Economy desk is judged by its own creator-economy terms instead, in its headline or feed description, and an item whose headline uses a word or phrase on the desk’s list of crime, death, health, relationship, adult-content and market terms is turned away. Everything else is discarded unread.
3. Publishers that do not allow reuse. An item from a publisher whose terms forbid reuse of its reporting, which is most of the news outlets we follow, stops here: no AI model sees it. If its headline files it under one of Interestana’s nine interests, it is published as that headline and a link to the original, with no summary, and marked noindex. For the Creator Economy desk the headline must contain a creator-business phrase (such as “brand deals” or “creator payouts”), or name creators as a group, podcasting, newsletters or a creator platform together with a business word (such as payouts, sponsors or rules); a creator named on their own does not count. It must also use no word or phrase on a longer list that adds more health, allegation and personal-life terms, and Tubefilter’s weekly “Have you heard?” news roundup, which puts three unrelated items in one headline, is never listed. Only the remaining items go on to the next steps: companies’ and institutions’ own announcements, and reports from publishers whose terms, as far as we have found, do not forbid reuse.
4. Summary. An AI model — Google Gemini models (gemini-2.5-flash-lite, gemini-3.5-flash-lite, gemini-3.1-flash-lite, each tried in turn when the one before it is unavailable), with an OpenAI model as a last fallback only when one is configured — is given the item’s headline, its text from the feed, the publisher’s name and, when the feed gives one, its publication date. It first decides whether the item is really about Interestana’s subjects and skips it if not; the Creator Economy desk’s model is also told to skip any item about a creator’s private life.
5. Source facts only. The model is told never to state a date, number, price, quotation, person or job title that is not in the item it was given; to quote only words that appear in the item, attributed as the item attributes them; and to add nothing from its own memory beyond a few words saying what a named company or product is.
6. Length follows the source. A summary is at most 300 words and is scaled to the item: most feed items are a headline and a sentence or two, so most summaries are short. The Creator Economy desk publishes summaries of 50 to 300 words, and never rewrites an item too short to support 50 words.
7. Automated checks. The answer must come back in the expected format, with topic labels from our own list. A summary much shorter than its source supports is not published, and one that runs long is cut back to whole sentences. A final check must file the article under at least one of Interestana’s nine interests (for the Creator Economy desk, its own section, and nothing else), or it is not published. Reading time is calculated from the summary’s length.
8. Key points, figures and quotations. Beside the summary the model may write a few key points, up to three figures for a “By the numbers” panel, and one quotation. A program matches them against the item: a key point or figure with a number the item does not contain is dropped, and a quotation is kept only when the same words appear inside quotation marks in the item, with the speaker named beside them. The links listed under “Linked in the original report” are the links the publisher’s own report contains; we list them as they are and do not check them.
9. Photographs. We never show a publisher’s photo: we hold no licence for them. A story about a NASA mission, spacecraft or rocket may carry a photo from NASA’s Image and Video Library, which NASA allows news outlets to use with credit. It is used only when every part of its credit is NASA, never when NASA names another owner, and it is labelled as a file image with NASA’s own title, date and credit, because it shows the subject rather than the event reported. The photo is served from interestana.com, so your browser never contacts NASA.
10. Publish. The article gets a permanent address on interestana.com and appears on the home feed and its topic and interest pages, and search engines that support IndexNow are notified. It is also listed in the sitemap and the RSS feeds. A headline-and-link page is not: search engines are not notified of it, and it is in no sitemap or feed. A Creator Economy desk article appears at /creators/news, and a desk summary in that section’s own RSS feed, instead of on the home feed, the topic and interest pages or the site-wide RSS feed.

The automated checks test the shape, length and subject of a summary. Apart from the numbers in key points and figures and the words of a quotation, they do not compare what the model wrote with its source, and they cannot catch a mistake in the source itself. A summary can still misread its item, leave out an important caveat, or get a name, number or date wrong. The instruction to use only facts from the source reduces invention; it does not guarantee its absence.

Articles fetched before 08:00 UTC on 12 October 2026 were written under an earlier prompt that asked for longer summaries and did not give the model the publisher or the publication date. A review of those articles found dates and details that their sources did not contain, so treat them with extra caution. They stay online at their addresses but are marked noindex and are left out of the sitemaps and feeds.

Each agent reads feeds from news publishers, company blogs and research outlets in its subject; examples for AI are OpenAI, Hugging Face and MarkTechPost, and for science and space, NASA and ESA. Each feed has a weight that decides which item comes first when two feeds carry the same story. We change the list when a feed breaks, when it keeps sending items outside our subjects, or when a publisher asks us to stop using it.

For every news publisher whose feed we read, we keep a note of what its terms say about reuse of its reporting, where we read them and when. A publisher whose terms forbid commercial or AI reuse is never summarised: its stories appear only as a headline and a link, as described above. That covers most news outlets we follow, among them Nature, SpaceNews, The Decoder, SiliconANGLE and BleepingComputer. Announcements that companies and institutions publish themselves (for example NVIDIA, NASA or the US cybersecurity agency CISA) can be summarised, at most ten a day across the site, plus up to six of NASA’s own news releases a day. The Creator Economy desk also summarises the US Federal Trade Commission’s consumer-protection press releases when they are about influencers or endorsements; FTC material is in the public domain.

Interestana summaries are reference snapshots, not original reporting. They restate the core facts of one report so a reader can decide quickly whether the underlying story matters to them, and every summary links prominently to that report. Opinions in a summary belong to the people and publications it quotes, not to us. A run that fails is logged and nothing is published.

If you spot an error, contact us through the contact page with the article’s address. Corrections are handled by the publisher, Khaled Samen. We aim to acknowledge a request within twenty-four hours and to correct the article, take it down or explain why we have not within seventy-two hours, depending on how much checking it needs. A summary we take down is barred from being summarised again. Publishers who object to how we use their RSS feeds can ask us to stop using them through the same page.

Interestana publishes no sponsored articles; if it ever does, they will be labelled as sponsored and kept apart from the news summaries. We accept no money for news coverage, source weighting or favourable summaries. The site carries advertising served by Google AdSense, and advertisers do not choose, buy or influence what we cover. Articles are selected and written automatically from the feeds above, so the founder’s agency clients have no say in which stories appear or how the feeds are weighted.

Reading Interestana needs no account. We use Google Analytics to count readers and Google AdSense to show ads, both of which use cookies — in the EEA, the UK and Switzerland only with your consent — and we collect email addresses only from readers who subscribe to the magazine. Our privacy policy at interestana.com/privacy explains exactly what is collected, why, and how to opt out or have it deleted. A summary may name only the people its source names, so it repeats nothing about a person that the source has not already published. If an article names a private individual and you want it removed, use the contact page.

Interestana is built to be referenceable. We publish a sitemap of every article written under our current summary instructions (older summaries stay online but are marked noindex, and so are the headline-and-link pages of publishers whose terms do not allow reuse of their reporting), describe each one with Schema.org NewsArticle structured data, allow AI crawlers (GPTBot, ClaudeBot, Google-Extended and others) in our robots.txt, and link to the original publication by name. When you cite an Interestana summary, cite the original publisher as the source of its facts.