14,826 impressions, 273 clicks: an SEO audit of our own blog

After years of auditing websites for clients, we noticed something a little embarrassing one day:We had never properly audited our own blog.

89 Chinese-language posts, written over six months. Was anyone reading? Where were they coming from? Did Google like them? We had hunches, not numbers.

So this post does one thing:We treat ourselves as the client and run the full audit, start to finish.Every number is public, including the unflattering ones.

As usual for this series, here's how it was done: the data comes from two official Google tools (explained below), Claude ran the analysis and made the changes, and I made the final call on every "should we change this, and how."

1. The numbers first

Data period: GA4 from 2026-07-28 (the day we installed it) to 11/7; Search Console through 11/5.

Put another way: just over three months, fewer than 900 views a month.This is not a high-traffic blog.We want to be clear about that up front, so the analysis that follows makes sense.

2. Two sets of numbers, two different questions

Many people read GA4 and Search Console side by side and get confused when the numbers don't match. In fact, they measure two different things:

• GA4 (Google Analytics) measures "someone arrived."When someone opens a page, the tracking code on it reports a visit, whether they came from Threads, LINE, or Google.
• Search Console measures "seen on Google."Each time your page appears in someone's search results, that's one "impression," even if they never click. Think of it as a passerby noticing your shop sign.

So "273 clicks from Google" and "313 visits from search engines recorded in GA4" won't match, and that's normal: one is counted on Google's side, the other on the page itself. Browsers that block tracking and people who leave before the page finishes loading both create gaps.

Only by cross-checking the two can you see where the problem lies.That's the first key takeaway from this audit.

3. Where readers come from: 60% from social

Take "Direct" with a grain of salt: when people tap links inside the Threads, LINE, or Facebook apps, the app often doesn't tell the site where the visitor came from, so it's logged as direct. Social's real share is likely even higher than the 60% shown here.

One number is even more extreme:A single post, "The Big Six agencies' methodology: half real expertise, half sales talk," accounts for 36% of all Chinese-language views, and the top 5 posts together account for 54%. It took off on Threads on August 16–17, drawing nearly 900 views in two days.

This tells us one thing:Right now, our blog is a "social" blog.When people share, traffic comes; when they don't, it goes quiet. Google brings in only about 10%.

4. The real problem: seen, but not clicked

Looking at GA4 alone, you'd conclude "just post more on Threads." But open Search Console, and the picture is completely different:

Impressions grew 38x over five months, while clicks have been stuck at 50–70 a month since July.

In plain terms: Google is ranking our posts more and more often, but people see them and don't click.

Then we looked at which posts. The most extreme was "What is hallucination in Chinese?": seen 1,750 times over the period, clicked 8 times, a 0.5% click-through rate. Average position 9.2, which is the very bottom of page one.

By the textbook, the fix here is "rewrite the title so people see the answer at a glance." But first we did something the textbooks often skip:We actually searched Google ourselves to see what searchers see.

5. Search it once, and the answer is right there

Searching "hallucination 中文" returns this layout:

1. At the very top:a Google Translate box, with "幻覺" in large type.
2. Next:an AI Overview, explaining the difference between medical hallucinations and AI hallucinations in three paragraphs. (Our post is one of the sources it cites.)
3. Only then do the regular results appear: Cambridge Dictionary, us, Wikipedia…

What searchers want to know is simply "what's the Chinese word for hallucination."The answer is given on the first screen. They have no need to click any site.

The industry calls this "zero-click search": the results page itself answers the question. In the last 7 days, our site had 520 impressions and 1 click for this term.

That finding changed our expectations for the post. No matter how good the title, it won't help someone who already has their answer.It's not that the post is badly written. It's the nature of this kind of query.

(Note: we searched while logged in, and results vary somewhat by person. But for "what is X in Chinese" queries, the translation box and AI Overview appear almost every time.)

6. So we sorted our posts into four types

Search-type examples: "The real distance of AI video generation" got 50 visits from Google, from people searching "LibTV pricing"; "Google TimesFM lost to a 1994 formula" got 12, from searches for "TimesFM 3.0." What they have in common:The answer searchers want can't be given on the results page. They have to click through to get it.

The social-type example is that methodology post: 1,025 views, but only 84 Google impressions. Forcing a search-style title onto it would only make it harder to share on Threads.

7. What we changed

We picked 10 posts with 150+ impressions and a CTR under 2%, and changed only two things:the titleandthe description shown under it in search results. The body text stayed the same.

There was only one rule:The first phrase of the title directly answers the question people are searching for.

The other 5 posts got the same treatment, including two English versions.

What we deliberately didn't do

This section matters more than the table above.

• Some people found our post on how LLMs work by searching "I'm just a language model," but that phrase doesn't appear in the post.We didn't stuff it into the title. If a title promises something readers can't find after clicking, they leave immediately, and Google notices.
• Some people found our model comparison by searching "Claude human customer service," but the post doesn't cover that.We didn't add it either.
• Some people searched "Claude person's name," but that post is actually about the Chinese pronouns 他/她/它 and doesn't explain where the name came from.So the description simply says "Claude is the name of Anthropic's AI assistant, not a real person," and promises nothing more.

The most common SEO mistake isn't doing too little. It's promising too much.Luring people in with a title gets you one click and one disappointment.

8. Writing our predictions down first

On the day of the changes, we used Search Console's URL Inspection tool to confirm one thing:All 10 pages were already indexed by Google, but last crawled between mid-October and early November.In other words, Google is still showing the old titles. The new ones won't appear until it crawls again.

Baseline (10/9–11/5, last 28 days): these 10 pages had 2,737 impressions and 9 clicks combined, a click-through rate of 0.33%。

Broken down:

• The Hallucination and Embedding posts (zero-click type): 2,053 impressions, 4 clicks, 0.19% CTR.
• The other 8 posts: 684 impressions, 5 clicks, 0.73% CTR.

Our predictions:

1. The other 8 posts: in the 28 days after Google recrawls, CTR rises from 0.73% to 2% or higher.
2. Hallucination and Embedding: barely move, staying below 0.5%.If they also rise sharply, our "zero-click" diagnosis was wrong.
3. Overall impressions won't drop as a result.Changing titles can prompt Google to re-evaluate rankings. That's the risk.

We'll come back in mid-December to publish the results.Win or lose, we'll publish.

Why write it down in advance? Because anyone can write a success story after the fact. Writing predictions first and checking back later is the only way you can judge whether what we say holds up.

9. For anyone who wants to audit their own site

1. Read GA4 and Search Console together.Look at only one, and you'll get a complete but wrong story.
2. Start with pages that are "seen but not clicked."It's the cheapest improvement: no rewriting, just new titles and descriptions.
3. Before changing anything, search Google yourself.If the answer already appears on the results page, a new title won't help. Your effort is better spent elsewhere.
4. Don't force social posts to become search posts.They have their own job to do.
5. Don't use titles to promise what the content doesn't deliver.
6. Write down the results you expect in advance.Otherwise, three weeks later, whichever way the numbers go, you'll feel it "worked."

One honest closing note: a blog with 900 views a month is no SEO success story. But that's exactly why we wanted to make this public—This is what most small business websites look like.You don't need a high-traffic site to make an audit worthwhile. What you need is to understand your own numbers, and to know what to change next and what to leave alone.

10. FAQ: GA4, Search Console, and SEO

Q: How are GA4 and Search Console different? Do I need both?

Yes, and both are free. GA4 covers what happens after people arrive: where they came from, which pages they viewed, how long they stayed, regardless of source. Search Console covers what happens before they arrive: how often your pages appear in Google results, where they rank, how many clicks they get, and whether Google has added them to its index. One watches inside the shop; the other watches the storefront.

Q: Why do the two sets of numbers never match?

Because they measure at different points. Search Console counts clicks on Google's side; GA4 only counts a visit once the page loads and the tracking code runs. Ad blockers, users who decline tracking, and people who close the page immediately show up on only one side. The default time zones also differ (Search Console uses US Pacific Time). A 10–20% gap is normal; just check that the trends line up.

Q: What does "average position 8.5" in Search Console mean?

It's the average of where you ranked each time you appeared, weighted by impressions. It doesn't mean you always ranked 8th: you might have been 3rd sometimes and 15th at others. Also note that positions 1–10 roughly make up page one, but ads, AI Overviews, and translation boxes often sit above them, so position 8 might be two or three screens down.

Q: What's a normal click-through rate?

There's no universal benchmark. CTR depends on ranking position, whether the results page has an AI Overview or other features, and what the query is. Rather than comparing against online "industry averages," compare your own pages at similar positions: if both rank on the lower half of page one and page A gets 3% while page B gets 0.3%, page B's title and description deserve a look.

Q: Lots of impressions, few clicks: did our SEO go wrong?

Not necessarily. There are two common causes: either you're at the bottom of page one and people don't scroll that far, or the results page already gives the answer (translation box, AI Overview, featured snippet), so searchers don't need to click. The first can be improved with better titles and content; the second you should accept first, then decide whether the term is worth the effort. The fastest way to tell them apart: search it yourself.

Q: After changing a title, how long until we see results?

You have to wait for Google to recrawl the page, usually a few days to two weeks. You can check the last crawl date with URL Inspection in Search Console, and click "Request indexing" to ask Google to come sooner. Once the new title appears, collect another 2–4 weeks of data before comparing. Also, Google sometimes rewrites titles or descriptions before showing them, so what you wrote isn't always what users see.

Q: Does the meta description (the text under the title in search results) affect rankings?

Google officially says it doesn't directly affect rankings, but it does affect whether people click once they see it. Write it like ad copy: answer what the searcher is looking for in the first sentence.

Q: Why is there so much "Direct" traffic in GA4?

"Direct" really means "source unknown." Besides people typing the URL or using bookmarks, links tapped inside apps like LINE, Threads, and Instagram often carry no source information and end up here too. The fix is to add UTM parameters to the links you share (a tag at the end of the URL that marks the source, for example ?utm_source=threads), so GA4 can recognize them.

Q: Why do the keywords in Search Console add up to fewer clicks than the total?

To protect privacy, Google hides queries with very low search volume, counting them in the total but not listing them individually. The more niche the site and the more long-tail queries it gets, the larger the hidden share.

Q: If I install GA4 today, can I see past data?

No. GA4 only records from the day it's installed, so the sooner the better. Search Console is different: Google has been collecting that data all along, so once you verify your site, you can see up to 16 months of history.

Q: Can I see traffic from ChatGPT and AI Overviews?

Partly. Visitors who click through from services like ChatGPT or Perplexity are usually recorded in GA4 as referral traffic from that site, and newer versions of GA4 group them under "AI assistants." Google's AI Overviews, however, are counted within Search Console's regular search data and aren't broken out separately yet.

If you'd like to see how your own site looks on Google and in AI search, we offer afree AI visibility audit。