News is where the llms.txt format earns its keep. A news site publishes dozens of stories a day, and the average AI tool that fetches your homepage and a handful of links will reach your next-day-old stories only by luck. An llms.txt file fixes that by telling AI tools exactly where your best reporting lives — organised by beat, anchored by standing pages, and refreshed by the timestamp that tells a model what is new.
Publishers that invest a few hours in this file give AI answers a short, curated path to their work. Those that skip it leave their discoverability to whatever the crawler happens to find. This guide shows you how to build the file for a newsroom: the sections to include, how to handle beats and standing pages, and how to keep it fresh without adding a new editorial chore.
Why news sites need llms.txt more than most
Most websites change slowly. A news site changes by the hour, and that is the problem an llms.txt solves. An AI tool that reads your homepage gets today's top stories once, and then it is guessing which section pages matter, which explainers are still accurate, and which older investigations still deserve attention. A well-built llms.txt removes the guesswork. It says: here are our beats, here are the pages that explain each topic, here is where live coverage lives, and here is how fresh this file is.
That curation matters because AI answers about news are heavily cited — or heavily plagiarised — depending on whether the model found your page. A citation is a link back to your site and a signal that you are a source worth reading. Without one, your reporting ends up summarised with no credit and no traffic. For local and mid-sized publishers, that difference is existential.
Build the file around beats, not your navigation menu
News sites usually organise llms.txt by beat: politics, business, sport, culture, local government, environment, health, technology. Under each beat, list the three to five pages that define your coverage — the investigation, the data page, the live blog, the explainer, the definitive guide. Then add a short note about what is not listed: "For the latest on the mayoral race, see our politics live blog."
Beats work better than your site navigation because your navigation is designed for humans browsing, while llms.txt is designed for a tool looking for the best answer to a specific question. Your "Opinion" section might be fourth in the nav; in llms.txt, opinion does not get its own beat at all, because AI tools should be directed to reporting first and commentary second.
Write one line per link, in this pattern: a Markdown link with the story title, then a colon, then a sentence of context — what the story is, when it was published, and why it matters. Example: "[City Hall budget passes after late-night vote](/politics/city-hall-budget-2026): Passed 5–4 on March 3; includes line-by-line spending breakdown and councillor reactions." That one line does more for citability than a bare URL ever will.
Standing pages are the backbone of a publisher's llms.txt
Every newsroom has pages that should never go stale: the election results hub, the team pages, the stadium guide, the who's who in local government, the ongoing cost-of-living tracker. These standing pages — sometimes called evergreen or hub pages — are the highest-value entries in a publisher's llms.txt, because they are the pages an AI answer can cite for days and weeks at a time.
Give standing pages their own section, prominently placed, and keep it short. A dozen links is plenty. For each one, include the last-updated date in the description. "Updated weekly" or "Last updated: November 3, 2026" tells the AI tool that this page is maintained and safe to cite. A standing page with no date is a risk: the model cannot tell whether your 2024 election hub is still accurate in 2026.
Do not list every article you have ever published on a topic. Three or four definitive stories per beat, plus the standing page, is a strong llms.txt. A hundred links is a sitemap, not a curation — and AI tools treat it as one, skimming without prioritising anything.
Freshness: the hard part, and how to make it easy
An llms.txt that was excellent in January and untouched since is worse than no file, because it keeps sending AI tools to last season's coverage. Publishers need a freshness routine that fits into existing editorial work, not a new job for an already stretched team.
The practical answer is automation plus one line of discipline. Generate your llms.txt from your CMS: a query that pulls the latest stories in each beat, your current standing pages, and their dates. Regenerate it on a schedule — daily for busy newsrooms, weekly for smaller ones — and set the file's own date line to the generation date. Then make one human check part of the publishing workflow: when an editor publishes a major story, they add or swap one line in the llms.txt source. One line, thirty seconds, and the file stays trustworthy.
Also include a "last updated" line at the top of the file itself, and a short note about your corrections policy with a link to it. AI tools notice these signals, and so do readers who find the file. A file that says "Updated November 3, 2026. Corrections are published here: [link]" tells everyone — model and human — that this newsroom stands behind its reporting.
The optional llms-full.txt for deep reporting
Journalism's longest work — investigations, data projects, long-form features — deserves the full treatment: an llms-full.txt that includes the full text of your best pieces, not just links to them. A link tells the tool a story exists; the full text lets it actually quote and understand the reporting without hitting your paywall or your page.
Do not put everything in llms-full.txt. Choose the pieces you would be proud to have quoted: investigations with documented sources, data explainers, and features with original reporting. A practical file stays under a few hundred kilobytes, which usually means a dozen long pieces rather than your whole archive. And keep it updated with the same schedule as the main file — an llms-full.txt that includes your 2025 investigation without the 2026 follow-up is actively misleading.
Syndication, wire, and licensing — handle them deliberately
Most newsrooms use some syndicated or wire content, and your llms.txt needs to say what is yours and what is not. Link to your own reporting first. If you include a wire story, label it as such in the description so the AI tool attributes it correctly — to the wire service, not to you. Getting this wrong can mean your site is cited for content it did not produce, which helps no one and risks licensing disputes.
The same logic applies to licensing. If parts of your archive are licensed to AI companies under specific terms, say so in the file or in a linked policy page. Transparency here protects your commercial position and your reputation: publishers who are clear about what AI tools may use are the ones whose terms get respected.
How to check it's working
You cannot watch every AI answer, but you can measure whether the file is doing its job. Start with server logs: requests for /llms.txt should appear regularly, and the user-agents fetching it should include AI crawlers and known AI tools. A file that nobody ever requests is not broken — it may just be new — but a file with no requests for weeks means you need to mention it somewhere an AI tool can discover it: in your robots.txt, in your site footer, or in your sitemap as an additional resource.
Then check citations. Search the AI answers for queries your beats cover, and see whether your site is cited and whether the cited pages are the ones your llms.txt points to. If the citations point to your most important stories, the file is working. If they point to random pages or to competitors, your llms.txt probably needs better curation — or the model simply has not fetched it yet, in which case give it time and recheck monthly.
Finally, validate the file's syntax and reachability with the free llms.txt generator, then confirm that each link in it resolves with the robots.txt tester. You can also check AI crawler access to your whole site with the AI bot checker.
FAQ
Where should llms.txt live for a news site?
At your domain root, /llms.txt, exactly like robots.txt. The root is the first place AI tools look; a file buried under a section directory will be found inconsistently, if at all.
What beats should a local publisher include?
Your main beats — local government, schools, crime and courts, business, sport — plus a standing-pages section and a breaking-news link to your live coverage. Leave out sections you rarely update; a stale beat is worse than no beat.
How often should a newsroom update its llms.txt?
Daily for high-volume newsrooms, weekly at minimum. The update does not have to be manual: generate the file from your CMS so new stories flow in automatically, and let an editor review it as part of the weekly desk routine.
Does llms.txt help with Google rankings?
Not directly — llms.txt is not a Google ranking factor. Its value is in AI answers: better citations, more accurate summaries of your reporting, and fewer AI hallucinations built from your outdated pages.
Should a paywalled publisher use llms.txt?
Yes, carefully. Point AI tools at your freely readable explainers and standing pages, and include a short note about your paywall and what registered readers can access. That builds citations without giving away the reporting your subscribers pay for.
Does the file need to list every story?
No. Curate ruthlessly: three to five definitive stories per beat, plus standing pages. A short, accurate file outperforms a long, stale one every time.
