Three real clicks arrived on the discount calculator this week. I went to look at which queries drove them. GSC returned 25 rows. Every single one of them showed zero clicks.
The clicks happened. I just can't see where they came from.
That's been the week in a sentence.
The setup
The daily automation ran every day this week — one tool page improved per day, the same pattern as the past six weeks. This week's batch: Conversion Rate (June 28), Runway (June 29), Engagement Rate (June 30), CAC Payback (July 1), CAC (July 2), LTV (July 3), MRR (July 4). Each commit adds verified benchmarks, worked examples, and internal links. Each one gets pushed, deployed, and submitted to GSC.
Seven pages improved. On paper, a productive week.
When I pulled the data, the useful signal was hard to find — and what I did find wasn't what I expected.
Finding #1: I improved the engagement rate calculator. Its position got worse.
The engagement rate calculator was in Monday's automation queue. The commit is there: verified benchmarks for Instagram, TikTok, LinkedIn, and YouTube, ERF vs ERR examples, updated internal links.
GSC for the 7-day window ending today shows the page at position 39.3. Over the prior three weeks (backing out the 7-day data from the 28-day window), the implied average position was around 33.6.
That's a drop of roughly 5–6 positions in the week following a content improvement.
I want to be careful here. Position can swing several points week to week from normal SERP volatility — one page shuffles, five shift around it. A single week's comparison isn't diagnostic. And GSC has a 2–3 day data lag, which means some of this week's positions reflect pre-improvement crawls.
But I can also look at the query-level breakdown. The head term "engagement rate calculator" — 54 impressions over 28 days — sits at position 43.8. That's page 4–5. The content improvement didn't move it out of there.
The prior post covered a similar dynamic: pages we'd improved once, improved again, still not moving. This week is the freshest version of that — a page improved five days ago that GSC currently shows at a worse position than before we touched it.
I don't know if the improvement hurt, helped, or had no effect yet. That's the honest answer. The signal window is too narrow and too noisy to say more.
Finding #2: Three clicks from queries I cannot name
The discount calculator had its best week on record. 3 clicks, 397 impressions, position 13.3. The prior three-week weekly average was near zero — roughly 0 clicks and 61 impressions per week.
The discount calculator last got a real content update in May. It wasn't in this week's automation queue. Nothing changed on the page.
So I pulled the query-level breakdown to understand what drove the spike.
GSC returned 25 queries. Here's the sample:
| Query | Impressions | Clicks | Position |
|---|---|---|---|
| 10% off 70$ | 1 | 0 | 7.0 |
| 35% off $42 | 3 | 0 | 11.0 |
| discount check | 7 | 0 | 37.0 |
| discounted price | 3 | 0 | 6.7 |
| check discount | 4 | 0 | 46.0 |
Twenty-five queries. 43 total visible impressions. Zero visible clicks.
The page had 397 impressions and 3 clicks this week. The visible queries account for 43 of those impressions — about 11%. The remaining 354 impressions and all 3 clicks are below GSC's privacy threshold. They exist; they just don't show up.
This is the graveyard I keep hitting from different angles. Week 1 of this series was about the loan payment calculator with "average position 9.8" that turned out to be one query at 2.5 and dozens of invisible long-tails. The second post found the same pattern in blog impressions. This week it's showing up as actual clicks — real sessions, real people, real conversions from queries I literally cannot read.
The clicks came from somewhere specific. Someone typed something, saw the page in the results, clicked it. GSC has the data. I'm just not allowed to see it.
I don't have a clean answer for what drove the spike. It could be a news cycle around a specific discount query. It could be a bot that happens to click. It could be one person with an unusual search pattern who stumbled on the page. I can't tell. What I can say is that an untouched page had a better week than the seven pages we polished — and the reasons are invisible to us.
Finding #3: Position 7.4. Zero visible queries. Zero clicks.
The blog post saas-valuation-multiples-in-2026-why-profitability-now-trumps-growth-at-all-costs appeared in the 7-day data this week for the first time. 27 impressions, position 7.4, zero clicks.
Position 7.4 is good. For a blog post that wasn't showing any impressions in the prior three weeks, breaking into the top 10 is the kind of number that should be encouraging.
I pulled the query-level breakdown to see which terms were driving it.
Zero rows returned.
Not low-volume rows. Not a handful of long-tail queries. Zero. Every one of the 27 impressions came from queries below the threshold — meaning each individual query had fewer than 2 impressions over the period, and GSC's privacy filter removed all of them.
This is the clearest version of the measurement problem I've seen. A page is effectively ranking on page 1 for something. Someone is searching for something, seeing it, and not clicking. I can't see the query. I can't write for the intent. I can't tell if "not clicking" means the title is wrong for the query or the query is so specific that someone just read the title and got what they needed.
The 0% CTR at position 7.4 is what I'd expect from an informational query where the title answers the question. Or from a position that looks better than it is because it's averaging across a distribution with one good spot and a dozen bad ones. I genuinely don't know which.
What I'm going to do about it
Run the automation another week — the 7-page batch from this week will show up in GSC in 10–14 days. The engagement rate calculator comparison needs more data before any conclusion is valid. I'll pull the same metrics next week.
Curl the discount calculator queries — the 25 visible queries include "discounted price" at position 6.7 and "35% off $42" at position 11. I want to check the live HTML for both to see if there's anything specific drawing ultra-niche calculation queries to this page that I could amplify.
Watch the SaaS valuation blog for two more weeks — if it's genuinely near page 1 for specific queries, clicks should start appearing within 2–3 weeks as GSC's threshold unlocks some rows. If it stays at 0 CTR with zero visible queries, the distribution is too long-tail to do anything with.
Leave the untouched pages alone — the discount calculator had a better week without being touched. The purchase order generator has consistently been the top-clicks tool without any content automation. The pattern of "untouched pages winning" has now shown up across three different measurement weeks. I'm not going to automate those pages.
The measurement problem
Here's what bothers me about this week's data.
Seven pages got improved. I know exactly what changed — the git log is specific, the diffs are readable, the content is better by any reasonable standard. What I don't know is whether "better" is doing anything in Google's model, and I won't know for weeks.
Meanwhile, a page that hasn't been touched in six weeks had its best week yet. The traffic came from queries I can't read.
The feedback loop is supposed to work like this: you improve a page, you measure the result, you adjust. But when GSC hides 89% of your impressions and 100% of your clicks behind a privacy threshold, and when position averages are arithmetic means across distributions that span position 2 to position 100, the feedback loop is barely functional.
This isn't a complaint about GSC — the privacy thresholds exist for good reasons. But it means the standard "improve content → wait → check metrics → iterate" loop runs on extremely noisy data. The only honest conclusion I can draw from this week is: the automation ran, something shifted somewhere, and I don't know what caused what.
That's not a comfortable place to be six weeks in. It's also not surprising. I knew going in that the measurement lag in SEO is weeks to months, not days. What I didn't fully appreciate until I started pulling query-level data every week is how much of the signal is simply not accessible, even in principle, under current tooling.
Next week I'll check the same pages. If the engagement rate calculator is still at 39.3 or worse, I'll have more confidence that something is genuinely wrong with the approach — and I'll say so.
I'm running these experiments on valuefy.app and writing up what the data actually shows, week by week. If you're building programmatic SEO or fighting the same measurement wall, drop a comment — I'd like to compare notes on what's working for you.
I also run AImiten, where we build AI tooling for companies. This side project is where I stress-test the ideas before they reach client work.
Top comments (0)