llms.txt — AI Search Discovery

SyteOps can generate and serve an llms.txt file at your site root — a plain-text map of your most important pages that AI assistants (ChatGPT, Claude, Perplexity) look for when deciding what to cite.

Enable it

Go to the Content Pipelines tab and, on the GEO (llms.txt) provider card, click Turn on. Once enabled, your llms.txt is served at https://yoursite.com/llms.txt and refreshes automatically when you publish content (and daily).

Native generator or LLMS Amplifier

By default the built-in generator produces your llms.txt — no extra plugin required. If you install and enable the LLMS Amplifier integration, it takes over llms.txt generation instead (an either/or choice), and the built-in generator steps aside. Switch between them from the GEO (llms.txt) card — Use LLMS Amplifier engine / Use native engine — with no separate setup step. See LLMS Amplifier.

What's in it

A Markdown list of your published Posts (most-recently-updated first), each with its title, URL, and a short excerpt — plus your site name and tagline.

Choosing which content types are covered

Content Pipelines → Review Portal settings → Answer-engine publishing → Content types decides which content your AI files describe. It covers all of them at once: llms.txt, the questions feed, FAQ structured data, and the LLMS Amplifier index.

Out of the box only Posts are covered. Pages are usually site furniture — your contact page, your privacy policy, a landing page — rather than content that answers a question worth citing, so listing them mostly pads the file. Tick Pages, or any other public content type, if your site is the exception.

At least one type is always covered, so clearing every box puts you back to Posts.

Changing this rebuilds your AI files right away, so the change is live within a couple of minutes rather than at the next scheduled rebuild.

:::note If you use LLMS Amplifier with a separate front end LLMS Amplifier can serve a decoupled ("headless") front end, in which case it works out each article's public address from its own public routing settings. A content type with no route there has no address Amplifier can publish, so its entries appear without a link. SyteOps tells you when that happens — beside the Content types setting, and in Amplifier's own health panel — so you can either add the route in Amplifier or untick the type here. :::

Publishing the questions your articles answer

When an article goes through the Review Portal, the GEO analysis records the specific questions it answers well — and an answer for each, drawn from the article. Those question-and-answer pairs are what an AI answer engine actually quotes, so SyteOps can publish them in five places.

Each one has its own switch under Content Pipelines → Review Portal settings → Answer-engine publishing:

Switch	What it does
Questions in llms.txt entries	Lists each article's questions and answers directly under its llms.txt entry. On by default.
Site-wide questions section	Adds a Questions this site answers section to llms.txt, combining every article's questions into one deduplicated list with the pages that answer them. On by default.
FAQ structured data	Adds FAQ structured data to the published article, which Google and Bing read today, and shows the same questions and answers in a visible block on the page. Off by default — see the warning below.
Questions feed	Publishes the whole question set as JSON at https://yoursite.com/llms-questions.json. On by default.
Questions in the LLMS Amplifier index	Adds the question set to the FAQ section of LLMS Amplifier's llms-index.json. On by default. This switch appears only when LLMS Amplifier is installed and switched on. Any FAQ entries you uploaded to Amplifier yourself are kept, and take precedence over these.

Both llms.txt options work the same whether the built-in generator or LLMS Amplifier is producing your llms.txt. The questions feed and the FAQ structured data are served by SyteOps either way, so they keep working no matter which generator you use.

:::note None of these switches publish anything by themselves Turning a switch above on does not put an article's questions and answers on that surface right away. They stay held back until a reviewer confirms them — see Reviewing the answers before they go live below. Once confirmed, they appear on every surface you've switched on, without you doing anything else. :::

Leaving an article out

The Review Portal lets you exclude an individual article from llms.txt. That exclusion now applies no matter which generator you use — including every file LLMS Amplifier produces, such as its knowledge bundle. Previously it applied only while SyteOps generated the file, so an excluded article reappeared once LLMS Amplifier took over.

If you use LLMS Amplifier, note that its own Include metadata option controls whether the per-article block appears — with that switched off in Amplifier, there is no metadata section for the questions to attach to.

:::warning Check before turning on FAQ structured data Many SEO plugins (Yoast, Rank Math, Squirrly) can already add FAQ structured data to a page. Two FAQ blocks on one page is invalid and search engines may flag it. Turn this on only if nothing else on your site is adding FAQ markup. The visible questions on the page use the same reviewed answers as the structured data — they are not a separate toggle. :::

Reviewing the answers before they go live

The questions and answers appear in the GEO panel of the Review Portal, where a reviewer can edit any of them, add a question the analysis missed, or remove one. Answers are shown in full — they are never cut off — and can be up to 2000 characters.

• Analyze GEO never overwrites your work. It keeps the wording and answers you have written, fills in only the blanks, and adds anything new it finds.
• Draft answer writes an answer to a single question straight from the article, kept in full so you can review it before it is published. If the article genuinely does not answer that question, it tells you instead of inventing something — edit the article, reword the question, or remove the row.
• If an answer already looks cut off from an earlier review, edit it or tap Draft answer again. Re-running the analysis keeps your existing wording, so it will not replace a cut-off answer on its own.
• I have reviewed these answers confirms someone has checked them — and this is also what releases them. An article's questions and answers stay held back from all five answer-engine surfaces until this box is ticked. Tick it and SyteOps rebuilds your AI files right away, so the answers appear without you doing anything else; untick it and they're pulled back out. Re-running the analysis changes the answers, so the confirmation clears automatically and the answers go back to waiting for review until someone checks them again.

By default, Approve & Publish also waits for this box before letting an article publish — that requirement is skipped automatically when no answer-engine option is switched on, or when an article has no questions, and you can switch it off entirely under Pre-publish requirements. Doing so only removes the check from the publish step; it does not change what appears on your answer-engine surfaces. An article's answers still wait for review before they show up anywhere, whether or not the pre-publish check is switched on.

The Answer-engine publishing settings card shows how many articles currently have answers waiting for review — for example "3 articles have answers waiting for review. Until someone checks them, those answers are held back from every surface below." On a very large backlog the check stops counting once it has seen enough, and shows "At least 200 articles…" instead of an exact number. If you use LLMS Amplifier, its own health panel shows the same count.

What gets published

Only information that is already public: the question, its answer, and the URLs of published articles. Draft articles, password-protected articles, and any article you have excluded from llms.txt are all left out, and the internal readiness score is never published.

An article shows up on these surfaces only after its GEO analysis has run and a reviewer has ticked I have reviewed these answers. Articles analyzed before this release have questions but no answers — they will gain answers the next time they are analyzed.

Pillar articles

The AI files can only hold so much, and by default they keep whatever was published or edited most recently. That means a routine news post can push your definitive guide off the end of the list.

Ticking Pillar Article (Cornerstone) in the portal's GEO panel says which articles your site's authority actually rests on. It does three things:

• Its questions and answers are kept ahead of other articles' when the index has to choose which to publish, and when several articles answer the same question, the pillar is the one an AI tool is pointed to. This applies whichever generator you use.
• With LLMS Amplifier generating your files, a pillar is listed first — and in the index file (llms-index.json) that means it survives the file's size limit rather than being trimmed. In llms.txt itself the ordering gives it prominence, but the size limit is applied before SyteOps can intervene, so it does not rescue an article from trimming there.
• If you have an SEO plugin connected, the same tick marks the article there too — cornerstone content in Squirrly SEO and Yoast SEO, pillar content in Rank Math.

Use it for a handful of pages, not for everything — the point is to say which few to trust first.

Keeping the files current

SyteOps rebuilds your AI files shortly afterwards, on its own, whenever something changes what they should say: editing an article's answers, reviewing or un-reviewing an article's answers, marking it as a pillar, excluding it from llms.txt, publishing a page, or unpublishing, trashing or deleting an article. That last group matters most — without it, a retracted article's title, address and answers would stay in your AI files indefinitely.

Rebuilds are batched, so a run of edits produces one rebuild rather than one per save, and they happen in the background — you never wait for one. If something else rebuilt the files in the meantime, SyteOps skips its own pass rather than repeating the work.

When another plugin also makes an llms.txt

Only one plugin can answer https://yoursite.com/llms.txt. Some SEO plugins — Squirrly SEO from version 14.2 onward, for example — have added their own llms.txt feature, and because of when it runs it takes the address before SyteOps gets a chance. When that happens, your visitors and any AI tool asking for the file receive that plugin's version instead of the one SyteOps builds for you, and nothing on screen would otherwise tell you.

So when GEO is switched on and SyteOps is generating your AI files, SyteOps switches the other plugin's llms.txt feature off for you. Your file is the one that gets served. Nothing else in that plugin is changed — its own AI usage rules are left exactly as you wrote them, and every other setting is untouched.

If you later switch GEO off, or deactivate SyteOps, the other plugin's feature is switched back on automatically, so your site is never left without an llms.txt at all. And if you had already turned that feature off yourself, SyteOps leaves it that way — it only gives back what it took.

If the switch cannot be made for some reason, you'll see a warning naming the plugin on the Answer-engine publishing card, so you can turn it off yourself.

When another plugin removes your FAQ answers

The same kind of clash affects structured data. Several SEO plugins offer a setting along the lines of "remove other Json-LD from page", which deletes every other plugin's structured data from the page before it reaches a search engine. If that setting is on, the reviewed questions and answers SyteOps publishes are deleted too — so FAQ schema looks switched on in SyteOps, is built correctly, and never arrives anywhere.

When SyteOps is actually publishing FAQ structured data, it switches that setting off for you as well. It only touches the "remove other people's data" setting — the other plugin keeps publishing its own structured data exactly as before. Turn FAQ schema back off, or deactivate SyteOps, and the setting is handed back.

SyteOps tells you when it does this

Each time SyteOps switches one of these settings off, you get a message in your dashboard saying which setting it was and where to find it. If someone later switches it back on, SyteOps switches it off again and shows you a warning, so a change that would quietly stop your AI files working is never silent.

You'll also see a note on the other plugin's own settings screen explaining why the switch is off and what to do if you want it back — turning off the matching SyteOps feature, rather than fighting the switch.

When another plugin serves your robots.txt

Some SEO plugins also take over robots.txt, and for the same timing reason they replace the version WordPress builds rather than adding to it. That quietly drops the lines pointing AI crawlers at your answer-engine files — including the one that tells them where your questions and business details live. Your files are still there and still correct; nothing is telling anyone about them. One of those two lines is a Sitemap: line, pointing at the sitemap of your AI files — not your ordinary search-engine sitemap, which is a separate file that another plugin looks after.

SyteOps does not switch that plugin's robots feature off. Your robots.txt holds your own crawling rules, and they are yours to keep — turning the feature off would take them out of the served file along with everything else.

Instead, the Answer-engine publishing card shows you exactly which lines are missing and offers to add them for you. SyteOps adds only those lines. Every rule you wrote stays exactly where it is, and there is an Undo this change button that puts the file back the way it was.

If you would rather do it yourself, the same card links straight to that plugin's robots editor.

A few things worth knowing:

• SyteOps checks what that plugin is already serving first, so it only ever adds lines you do not already have.
• If you later switch off the feature those lines point at, SyteOps takes them back out on its own — you are never left advertising a file that is no longer published.
• If you edit one of the lines yourself, SyteOps treats it as yours and leaves it alone.
• If one of your own rules blocks the file a line points at, SyteOps tells you, because adding the line would otherwise look like it worked while changing nothing.
• robots.txt is usually cached by your CDN, so SyteOps clears it where it can and tells you when it could not — a change at your site is not the same as a change a crawler can see.
• SyteOps has no opinion about your ordinary search-engine sitemap and never changes that setting. If it says nothing about your XML sitemap, that is on purpose, not an oversight — the only sitemap it ever adds a line for is the one holding your AI files. If your SEO plugin is serving robots.txt with its own sitemap feature switched off, so that file lists no sitemap at all, the Answer-engine publishing card will point that out and leave the fix to you.

A setup review when you switch answer-engine publishing on

The clashes above are all detected the same way, and none of them announces itself. So when answer-engine publishing is switched on and LLMS Amplifier is producing your files, SyteOps shows you a short review of what it found — in two halves, because they need different things from you.

Already handled lists what SyteOps has done on its own: taking back your llms.txt address, and switching off another plugin's "remove other Json-LD from page" setting so your FAQ answers survive. Both are reversible and both are handed back if you turn the feature off.

Needs your approval lists what SyteOps will not do without being asked: adding the discovery lines to another plugin's robots file, and switching on the LLMS Amplifier settings that publish the index file your answer surfaces write into. Each has a button.

The review never blocks anything — answer-engine publishing is already on by the time you see it, and closing it changes nothing. Not now is remembered against what the review said, not as a permanent "never show this", so if a new problem appears later you will be told about that one.

The Review answer-engine setup button on the Answer-engine publishing card opens it again whenever you want it. It stays on the card once answer-engine publishing is on with LLMS Amplifier, including when everything is in order — open it on a healthy site and it tells you so, rather than disappearing and leaving you to guess whether there was nothing to report or nothing loaded. The review only opens by itself when something is actually outstanding.

Notes

• If a real llms.txt file already exists at your site root (from another plugin or a static file), your web server serves that one instead.
• As of now, no major AI vendor has publicly committed to reading llms.txt — it's a low-cost, future-proofing step, not a guaranteed ranking change.