Skip to content
Help Center

Generating articles from a website source

Short answer: open Help Center → Website sources, pick a crawled website and click Generate articles. You tick the pages that may be converted, at most twenty at a time, and Sarrai builds collections, folders and articles from them.

What the action sets off

  1. Choosing. You're shown the crawled pages and tick which ones may be converted.
  2. Reading in. The chosen pages are prepared.
  3. Generating. The same AI pipeline as KickStart builds collections, folders and articles from them.

This is separate from the KickStart wizard: you don't need onboarding and the site is not crawled again. Sarrai reuses the text that's already been read in. The action only appears when the crawl produced usable pages, by the way; during a running crawl or on an empty source the button isn't there.

Read them afterwards, not before

The biggest quirk: the generated articles come in published and available to the AI right away, with no approval step. What the AI wrote is therefore live on your Help Center immediately.

Plan time for that after the conversion, not before. If you're converting a large site, do it at a moment when you can actually go through the articles.

You convert in batches of twenty

Per conversion you can select at most twenty pages. That limit is built in and isn't in the settings anywhere, so you can't raise it. A site of three hundred pages therefore takes fifteen consecutive batches.

That's deliberately made workable. On opening, the top twenty pages not yet converted are already ticked, so confirming once is enough for "the next batch". The Select top 20 button refills your selection to the limit, and Clear selection empties it. Tick beyond twenty and the extra tick is silently ignored; there's no message telling you you're at the limit.

Converted pages are remembered

Every page you convert is marked permanently. That has three effects you don't see.

Already converted pages are hidden by default in the list; a toggle brings them back with a converted badge. The pre-selection skips them, so each round you automatically get the next twenty rather than the same ones. And the reading-in itself skips already converted addresses too, which makes retrying safe: you get no duplicate articles if you repeat a failed attempt.

Watch one thing: that marking survives deleting the generated articles. Throw a generated article away and want that page converted again, and you have to make it visible through the toggle and tick it by hand. It won't come back into the pre-selection by itself.

What the list shows you

The pages are ordered largest first, measured by the number of sections the crawl pulled from that page. Large pages usually carry the most content and make the best candidates for a full article.

You see no text preview and no character count, only "x sections". That's a deliberate choice: requesting the actual text forced the database to read all page content off disk on large sites, which left the overview hanging for minutes at 500 pages and then failing. So you choose on address, title and size, not on content.

Limits and cost

Article generation sits behind your plan. If it isn't in there, the screen shows an upgrade hint instead of the page selection; see How does Sarrai's pricing work?

If the AI proposes more than a hundred articles, you're asked to confirm first, because this costs money. Above a very large amount of text the article in progress is finished and the generation stops with the status partially completed; you do the rest in a following round. Strongly similar pages are flagged before generation, with "use both" as the default choice. And all usage counts towards your reporting, see The AI agent: billing mode, limits, and usage.

Good to know

Addresses longer than eight hundred characters are silently skipped. They don't fit the index the conversion keeps, and you get no message about it: the page simply never shows up as an article.

Only one generation runs at a time per workspace. Start one while another is busy and you see the progress of the running one instead of a new selection. So you can't convert two sources at once.

Closing the window doesn't stop the generation. It keeps running in the background; reopen the action and you pick the progress back up. And if a run looks stuck, it's resumed rather than redone: Sarrai continues on the same run and skips articles that are already written.

Also searched as

"turn a website into articles", "fill the knowledge base from my site", "generate articles button", "why only 20 pages", "generated articles are live already", "regenerate without duplicates"

See also

Was this article helpful?

Your feedback helps us improve this article.