After one month of using a personal side project as an SEO test bed, I had 632 organic visitors. The number was encouraging, but the most useful lesson came from a different number: on October 15, the site reached a daily peak of 83 visitors, and part of that peak did not look human.
I was preparing for a larger SEO initiative at work and wanted practical experience rather than another round of theory. The side project gave me a place to make changes, watch the data, and be wrong without turning every observation into a production decision.
That distinction became important quickly. A metric can move at the same time as a change without proving that the change caused it. A traffic spike can look like growth until you inspect its sources. And a session replay can reveal something strange without telling you exactly why it happened.
What 632 organic visitors actually told me
The first month produced 632 organic visitors. That is the result I can confirm. What I cannot claim from that number alone is that I had found a repeatable SEO formula, that every change improved rankings, or that the trajectory would continue.
For a learning project, the value was in having enough real search traffic to observe. I could make a change, monitor what moved, and then ask a more precise question. That was much more useful than treating SEO as a checklist where every “best practice” is assumed to produce a ranking gain.
The font-size change taught me to be careful with causality
One of the things that surprised me was a correlation between font-size adjustments and performance metrics. In the original version of this story, it was tempting to summarize that as “small UX details can affect rankings.” That is stronger than the evidence supports.
What I actually observed was that I changed the font size and later saw related metrics move. I did not run a controlled experiment that isolated font size from every other change, and I do not have evidence that a search engine changed rankings because of that specific CSS adjustment. Font size can change layout, wrapping, element positions, and therefore some user-experience or rendering measurements, but the exact mechanism in my case remains unproven.
A safer rule for experiments like this is: record the observation first, explain it second. “Metric X moved after change Y” is data. “Y caused X” is a hypothesis until I have enough evidence to defend it.
The 83-visitor day looked better before I opened Webvisor
On October 15, the site reached its highest daily total so far: 83 visitors. This was not necessarily 83 organic visitors; it was the daily traffic peak I was looking at. When I broke the traffic down, a significant share came from China and was classified as direct traffic.
“Direct” is easy to misread as “someone typed my URL.” In Yandex Metrica it can also mean that the referrer was not passed, among other cases. Yandex documents this in its traffic-source documentation. So the source label by itself does not tell me who the visitor was or how they really discovered the page.
I then opened Yandex Webvisor and watched the suspicious sessions. The page appeared to be accessed with CSS disabled. Based on the pattern I saw, I estimated that roughly 20 of those “users” were probably bots.
CSS disabled was a clue, not proof
That estimate is still an estimate. There are at least two broad explanations for what I saw.
- The requests really were automated. A scraper may fetch HTML while ignoring stylesheets, images, or other assets it does not need. Saving bandwidth is one plausible reason, but I did not confirm that this was the motive.
- The replay may not perfectly reproduce the original session. Session-replay systems reconstruct what happened from recorded data. Yandex itself documents cases where the recorded page can look different from what the visitor actually saw, including CSS-related replay mismatches after stylesheet changes.
So “Webvisor showed the page without CSS” is something I observed. “The visitor deliberately disabled CSS” is an interpretation. “Therefore it was a bot” is a further inference. The pattern made bot traffic plausible to me, but one replay characteristic is not enough for a confident classification.
What I would verify in the logs
The server logs are the next place I would trust more than the visual replay. They can help turn a suspicion into a stronger case. For traffic like this, I would compare:
- the User-Agent values and whether many sessions share the same unusual fingerprint;
- IP addresses or network ranges, while remembering that one IP does not equal one person;
- request timing and repetition: humans and automated crawlers often leave very different request patterns;
- whether the client requested only HTML or also fetched CSS, JavaScript, images, and other assets;
- requested paths, status codes, referrers, and repeated navigation sequences;
- whether the same behavior appears in analytics dimensions such as browser, JavaScript support, region, session duration, and landing page.
None of those signals is perfect in isolation. Together, they can support or weaken the bot hypothesis much better than “the replay looked strange.”
The bigger lesson was about measurement, not rankings
I started this side project because I wanted practical SEO experience before taking on a larger initiative at work. After a month, the useful result was not just the 632 organic visitors. It was learning how quickly an attractive story can form around incomplete data.
I could have looked at the 83-visitor peak and called it growth. I could have looked at the font-size change and called it a ranking factor. I could have looked at CSS-less sessions and called every one of them a bot. Each version would have been simpler. None would have been as defensible.
For future experiments, a more defensible workflow would be to separate traffic sources, keep a change log, compare before and after without assuming causation, inspect anomalies at the request level, and treat analytics tools as measurement systems with their own limitations.
For me, that has been the most valuable part of the experiment so far. SEO gave me the traffic to study; the harder skill is deciding what the data actually proves.
I am still digging through the logs, especially around the direct traffic from China. If I can identify a repeatable request pattern behind those sessions, I will have a much stronger answer to the original question: whether those roughly 20 visits were really bots, and whether skipping CSS was part of how they operated.