To start with who we are: we are Beijing Siwen Lide, based in Shunyi, and we have spent 14 years on enterprise IT Operations Services, Network Equipment Packages and Low-Voltage & Cabling Projects.This is the 35th post in our SEO/GEO series. This one has to admit a fact that is not very flattering: over the past four weeks we published 36 articles, and our index count on Toutiao Search has stayed flat at 48 — not one more.
In the previous 34 posts we looked at rankings, at citations, and at why AI does not recommend us. Today we are looking at an earlier link in the chain: the content went out — did the search engine actually take it in? If this link is broken, everything downstream is spinning in place.
Three numbers, read together
| Metric (Toutiao Webmaster Platform) | Late August | Mid-to-late September | How to read it |
|---|---|---|---|
| Indexed pages | 48 | 48 | Not one added in four weeks |
| Daily crawl count | 87 to 104 | 27 to 49 | Roughly halved |
| Impressions | A few scattered times in August | Last data point stops at the end of August | Almost never shown |
What we were doing over the same period: our sitemap holds 341 URLs and 275 articles, 36 of them published since August 27, and every one was submitted on our own initiative after publishing. We did the submitting; the indexing did not follow.
The most direct check: we took the titles of the five service pages published this week and searched them on Toutiao Search exactly as written — not one of them returned us. Searching our company name as a control, we rank 3rd, which shows that both the search engine and the account are working normally. Zero is zero; it is not a search mistake.
Submitting, crawling, indexing and impressions are four separate things
Many people treat “I submitted it” as “they took it in.” In fact there are four gates in between, and if any one of them is blocked, nothing downstream happens:
- Submit: you tell the search engine the URL. This step is entirely under your control — done is done.
- Crawl: it sends a crawler to read your page. The webmaster platform shows how many crawls happen each day. Ours fell from around 100 a day to roughly 40.
- Index: after reading the page, it decides whether to store it. This step is its choice, not yours, and it is where we are stuck.
- Impressions: when someone searches a relevant term, your page is placed in the results. Being stored does not mean being shown.
In our case: submitting is fine, crawling is falling, indexing is flat, impressions are almost nil. The halving of the crawl count is itself worth noting — crawlers come less often usually because they have concluded there is nothing new worth storing when they do, which is a negative feedback loop.
Why the number is 48
The figure 48 is telling. Our sitemap has 341 URLs; only 48 are indexed. Going through those 48, they are almost all old pages that have existed for years, including the one that has long ranked on page one for our main keyword and still carries its old title. New content has hardly been picked up at all.
There are a few possible reasons; we can only say which ones we have ruled out and which remain uncertain:
- It is not a submission problem: the site and the sitemap were submitted back in late April, crawling happens every day, and each new article was submitted individually after publishing; the platform returned “submitted successfully” every time.
- It is not an accessibility problem: after each article went live we visited it once as a browser to confirm a 200 response, and we checked every internal link one by one.
- It may be the content format: the old indexed pages are mostly service introduction pages, while the newly published ones are mostly long-form articles. But this is only a guess, with no evidence behind it.
- It may be site authority: there is an inherent ceiling on how many pages a site is “worth storing” in its view, and we have never broken through it.
What this means for AI recommendations
Toutiao Search is one of Doubao’s retrieval sources. If the index does not move, the direct consequence is: what we newly write is largely unusable to Doubao. This week we deliberately asked Doubao and one other AI questions matching two freshly published service pages, and the result was that neither AI cited a single article we published this week. The other AI returned 10 cited sources for each of the two questions — 20 in total, not one of them from our official website. All were business directories, classified listing sites and merchant pages from years ago, the oldest published in 2014.
A measurement bias we noticed along the way: it recognized who we are
Today, when we asked the monitoring question, Doubao’s process record contained these two lines: “I notice the Siwenlide official website appears in the search results,” and “The Siwenlide official website shows that the company itself provides low-voltage and security system projects, and I will note this in my conclusion.” It then recommended several other companies and left us off the list.
Four days earlier, asked a similar question, it said something comparable: it found that the company itself is a low-voltage projects provider, and that if you were fitting out your own office, going through your own in-house system would be the smoothest route.
This means part of our own measurement data is noise. We asked our questions while logged in with the company account, so the AI knew who was asking and treated our pages as “your own material” rather than as a candidate answer. A real prospective customer does not ask that way, and what they see may differ from what we see.
This applies to anyone testing their own company: the answer you get from your own account is not the answer your customers see. A closer approximation is to ask again while logged out, or from a device that has never been logged in, and compare the two.
This bias does not affect the main conclusion of this article. An index count stuck at 48, crawl volume halved, and five new article titles that cannot be found — all three come from the webmaster platform and ordinary search, and have nothing to do with login status.
It also explains a pattern that recurred in earlier posts: when Doubao answers questions about Shunyi it cites pages we published in early September, but when answering on new topics it cites only other people’s old pages. It is not that it dislikes what we write — it simply does not have it.
What we plan to do next
- Install the automatic indexing code: the Toutiao webmaster platform provides a snippet to place on the page, which pushes the URL again every time a user opens the page — more timely than batch submission. This is the only lever we can pull right now against “cannot get into the index,” and we plan to install it first.
- Stop trying to solve it with volume: 36 articles in four weeks bought us zero index growth; piling on more makes no sense. We will put our effort into a small number of terms that people actually search for.
- Treat the old pages as assets: since the indexed pages are the old ones, we will correct the titles and content of those few dozen old pages so they answer the right questions, instead of hoping new pages will squeeze in.
- Look at these three numbers again in two weeks: index count, daily crawl count, impressions. Whether it works, the numbers will tell us.
If you publish content too, these three numbers deserve a weekly look. Watching only “how many posts went out” and “what rank I hold” leaves out the step in between: publish all you like, but if nothing gets stored in the index, ranking never enters the picture. Most search platforms show indexed page counts and crawl counts in the webmaster console, and after you launch a site it takes ten minutes to connect it once.
Want to know how many pages of your own website are indexed?
Siwenlide has spent 14 years in enterprise IT operations and network equipment packages, and this series is our own hands-on record of doing GEO, warts and all. If you need help with enterprise networking, low-voltage & cabling, or IT operations, let’s talk — call 400-686-2011.
Earlier in this series:Dropped off page one on Friday, ranked first on Monday (Post 34) · Six business lines, but search engines recognize only one · Search your own article titles exactly as written · Five silent website failures that look fine but no longer work
