First, a note on who we are: we are Beijing Siwen Lide Technology & Trade Co., Ltd., based in Shunyi, and we have spent 14 years on enterprise IT operations, Network Equipment Packages, and Low-Voltage & Cabling Projects. This is the 34th post in our SEO/GEO series. In the 33rd post last Friday (September 19) there was one line of record: the primary keyword “Beijing IT Operations Outsourcing company” did not appear on the first page across three consecutive searches on Toutiao Search, while on September 15 it still ranked No. 3. At the time I wrote, “No guessing today — we’ll recheck on Monday and confirm.” Today is Monday, and the confirmation is in: all four rounds came in at No. 1 (one round at 9:51 a.m., then three consecutive rounds around 10:50 a.m.). For “Beijing IT Operations company,” three of the four rounds were No. 8 and one was No. 16; on September 15 it was No. 13.

This post isn’t about new tactics. It’s about something more fundamental: just how unstable rankings in search and AI really are, and what kind of data qualifies as a basis for drawing conclusions or changing a page. After 34 posts, we nearly let a single fluctuation lead us astray ourselves.

The Same Keyword’s Ranking Within One Week

On Friday we changed nothing. We published no new pages over the weekend either. The ranking came back on its own, and higher than before. Conversely, if on Friday we had treated it as a drop and changed the title and body copy, then seeing No. 1 today we would have assumed the changes had done the trick — that is the kind of false causality a single fluctuation can manufacture.

Doubao Is the Same: Four Weeks, Four Answers to the Shunyi Question

Every Monday we put the same question to Doubao: “Which IT operations companies in Shunyi would you recommend? I run a 60-person company in Shunyi and I’m looking for local IT outsourcing.” The results across four weeks:

For the September 15 round we did one thing: we added a public price table to the Shunyi list page. Today it’s back, but we cannot claim the price table did it — because the Toutiao example from Friday just proved that it can come back without any change at all. All we can say is this: present in three of four weeks, absent in one, and the trend is stable.

What Kind of Data Qualifies as a Basis for Conclusions

Why the Fluctuation Happens

We have no internal data from Toutiao or Doubao, so we can only reason from what we observe. A few observable facts:

• The entry that ranked No. 1 on Monday still carries a title from several years ago: “Beijing IT Operations Outsourcing Enterprise – Beijing Siwen Lide Technology & Trade Co., Ltd. – Enterprise Private Line.” What the index holds is still the old page; whether it appears depends on updates on the index’s side, not ours.
• On Friday the first page held 22 results; today it ranges from 20 to 24 — the total itself shifts. “Beijing IT Operations company” appeared at both No. 8 and No. 16 within the same hour, so the slots on the first page are themselves in flux.
• Doubao’s “work mode” delegates the question and re-retrieves every time; the set of materials retrieved differed across all four weeks, and the Shunyi miss most likely just means it wasn’t found on that particular occasion.

The conclusion: a ranking is not a fixed number but a distribution. What we can do is shift the center of that distribution forward, rather than staring at a single day’s sample.

What Was Stable and Unstable This Week

Looking at this week’s five fixed queries together, which are trends and which are noise:

• Consistently present: the Shunyi question (three of four weeks), company-name search (top five every week), and “Beijing IT Operations Outsourcing company” (top three on the first page in two of three weeks).
• Consistently absent: “Which Beijing IT operations companies would you recommend (80 employees)” — Doubao showed zero for four weeks straight; today it recommended eight, and the top reason was that they “specialize in companies under 100 people, 19 years in business, fully certified”; “Which Beijing IT outsourcing provider is best” — zero on both Toutiao and Doubao; “How much does it cost” was quoted verbatim once on September 5, then zero for three weeks.
• Noise: that Friday, the main keyword dropped off the first page.

Those three consistently-absent queries are exactly where the next round of effort belongs. For the 80-employee question, all eight companies recommended today had the “service model” and “price reference” columns filled in — the same logic behind the star ratings discussed in post 31: those with a clear positioning and public pricing make the list; those without cannot be filled into the table. Our 50-to-100-employee page (published September 15) has not been retrieved yet; let us wait for it to enter the index rather than rush to change it.