69 to 1,190 Indexed in a Week: What Actually Moved It

69 to 1,190 Indexed in a Week: What Actually Moved It

Seven days ago Google had indexed 69 pages of our blog. Today Search Console reports about 1,190 indexed and 215 not indexed. Nothing was added to the corpus in between beyond twelve posts, no links were bought, and no console setting was flipped. What changed was the shape of the site's internal linking and the schema derived from the pages themselves.

This is the follow-up to a post that ended with "none of this has moved anything yet". It had. Below are the numbers, the two changes that plausibly caused them, and the honest list of what indexing still does not buy you.

What was the starting state?

The corpus went from 84 posts to 1,360 over five days, across six languages. The morning after the last batch, Search Console read: 69 indexed, 43 discovered and not indexed, the rest not discovered at all. Publishing had not been the bottleneck; being reachable was.

Two things were wrong, and both were structural rather than editorial. Every post was reachable only from one hub page that listed all of them, so authority was divided by the size of the list and no crawl path led from one post to another. And the engine's own FAQPage support was fed by a frontmatter field that 40 posts out of 1,360 had filled, which is close enough to nobody.

What changed, exactly?

Two changes, both mechanical, both shipped in one day.

The first replaced the flat hub with a ring. Sort each translation group, then let post i link to posts i+1 through i+6, wrapping around at the end. Every post ends up with six outgoing and about six incoming links, and no page is an orphan by construction rather than by care. Internal links per page went from 10 to 16.

The obvious alternative, a "top six related posts" block, would have been worse than nothing: six pages would collect a thousand inbound links and the rest would collect zero. That is a funnel, not a mesh.

The second derived the FAQ schema from the rendered article instead of from a field nobody filled. Only a subheading that actually ends in a question mark qualifies, and the answer is the first paragraph under it, so both halves are visible on the page as Google requires. That produced 1,157 pages carrying 8,217 questions, none of it invented.

What do the numbers look like now?

The line worth staring at is the last one. Average position 22.4 means page three. Everything is in the index and almost nothing is on page one.

Did the consoles do anything?

Barely, and it is worth saying plainly because a lot of advice suggests otherwise.

Resubmitting the sitemap made Bing aware of 1.4K URLs where it had known 11. IndexNow accepted a large batch in seconds. Google's request-indexing tool caps at roughly ten URLs a day and Yandex's recrawl queue at 150, so neither can move a corpus of this size. All of that is discovery. None of it is authority, and the difference between the two is the whole story.

The honest reading is that the consoles told Google the pages existed and the ring told Google they were worth walking through. Only the second one changed the indexed count.

Did the three engines behave the same way?

Not remotely, and the difference is itself useful.

Google took the corpus: about 1,190 pages indexed, 879 impressions, average position 22.4. The queries that reach us are mostly not in English, which is what multilingual long-tail produces.

Yandex has no complaints at all. Its diagnostics report zero errors, it found no meaningful duplication of titles or descriptions, and 73 percent of our impressions land on the first page of its results. The clicks are few but real, and they arrive on sensible queries. Its search updates add a few of our pages a day.

Bing is furthest behind: 14 impressions and zero clicks against 1.4K URLs it knows about from the sitemap. It is also the only one of the three that states the cause in its own interface, in one line: not enough inbound links from high quality domains. These are not three different problems. It is one problem, and one of the three decided to say it out loud.

What does indexing not buy you?

Three things, and pretending otherwise would waste your time.

It does not buy position. Nine hundred impressions at position 22 is a page that Google will show when almost nothing else matches, which is exactly what long-tail multilingual coverage produces at a young domain. The queries where we do appear are narrow and specific: hotel site builders in Spanish, "crm selbst erstellen" in German, vacation rental pages in Spanish. Real queries, real intent, third page.

It does not buy clicks. Fifteen clicks against 879 impressions is a 1.7 percent click-through rate, and at position 22 that number is mostly a measure of how far down people scroll, not of how good the titles are.

It does not buy links. Bing Webmaster Tools states the remaining problem in one line: not enough inbound links from high quality domains. That is the same conclusion the numbers reach from the other direction, and no amount of publishing changes it.

What would you do differently from the start?

Build the mesh before publishing the corpus, not after. The ring costs a few lines in the build script and it is the difference between a thousand pages Google walks and a thousand pages Google files away as "discovered, currently not indexed". Retrofitting it worked, but a week of crawl budget went into a shape that did not deserve it.

Check what your own engine already supports before adding anything. The FAQ schema had been in the pipeline from day one, fed by a field that 40 posts out of 1,360 had filled. A capability that 97 percent of your own content walks past is close enough to a missing one, and finding it cost nothing.

And measure the thing you are claiming. Every number above came out of Search Console, Bing Webmaster Tools and the build script, not out of a feeling that things were improving. If you want the method rather than the result, how to write a good prompt is the same discipline applied to a different problem: say the specific thing, then check what came back.

Does this apply to a site smaller than a thousand pages?

The ring does, and it costs nothing. Any site where posts only connect through a category page has the same shape at a smaller scale, and the fix is identical: give every page a handful of real neighbours rather than one link home.

The schema does too, with a caveat. Derived FAQ markup is only honest if the questions and answers are genuinely on the page. Turning a statement into a question to qualify would be marking up something the page does not say, and that is the kind of shortcut that works until it does not.

The part that does not transfer is the scale of the payoff. A hundred pages will not show a 17-fold jump because a hundred pages were probably never crawl-budget limited in the first place.

What comes next here?

Links, and there is no clever version of that answer. The corpus is indexed, the mesh is built, the schema is real and the position is 22. The next move is not more pages; it is being cited by places that already have authority, which is slow, manual and unrelated to anything a build script can do.

We write these up while building Zugo, which turns a written description into a working game, site or app with every build booted in a sandbox before it is called done. If you want the earlier half of this story, with the code for the ring and the schema regex, it is in what AI builders cannot do and the posts it links to.