One calculator peaked at five clicks and position 26 in late July. I never changed a line of its code. Then it fell to position 43 and one click per week — and I still haven't touched it.
Meanwhile, a different page — one the automation committed to 14 days ago — just had its best week in the series: four clicks, 10.53% CTR, position 21.
Both facts are real. Neither adds up to a clean story. That's what this week looked like in GSC.
The setup
Since July 5, a daily Claude Code routine has been improving one calculator page per day. Benchmark update, worked examples, verified external sources, updated dateModified schema. Commit. Repeat.
42 commits later, the queue this week covered: Runway, Ad Spend, Conversion Rate, CAC, LTV, ARR, MRR. Every one of those pages has zero presence in Google Search Console. Not a single impression across all seven.
That part I expected. Pages committed today don't show up in GSC tomorrow. Or next week. The lag between on-page work and Google's response has been a constant in this series.
What surprised me was the pattern on the pages the automation touched weeks earlier — and the one it never touched at all.
Finding #1: The impression calculator peaked in July without telling me
The impression calculator is our highest-volume page by impression count — 2,007 impressions in the last 28 days. I haven't committed to that file in over six weeks.
Here's what the weekly data shows:
| Week | Clicks | Impressions | CTR | Position |
|---|---|---|---|---|
| Jul 18–24 | 4 | 555 | 0.72% | 26.4 |
| Jul 25–31 | 5 | 664 | 0.75% | 31.5 |
| Aug 1–7 | 1 | 474 | 0.21% | 32.7 |
| Aug 8–16 | 1 | 475 | 0.21% | 43.4 |
Something changed on or around August 1. The clicks fell from five to one. The CTR dropped from 0.75% to 0.21%. The position has been sliding since — from 31.5 to 43.4, a move of 12 positions over two weeks.
I went back through every commit from July 30 to August 2. Nothing touches ImpressionCalculator.tsx. The sitemap got a batch lastmod update on August 1 — every page's date bumped to August 2026 at once — but I can't confirm whether that caused a recrawl that hurt the page or whether it's coincidence.
The honest answer is I don't know what changed. Google knows. I don't.
Finding #2: Most of what I'd need to diagnose it is hidden
The impression calculator has 2,007 impressions in 28 days at the page level. When I pull the query-level breakdown for the same period, I get 25 queries covering 174 visible impressions.
That leaves 1,833 impressions — 91% of the total — in GSC's privacy graveyard. Queries with one or two impressions that don't meet the threshold for disclosure. The bulk of the traffic distribution is invisible to me.
I can see that "impression calculator" (the head term) sits at position 30.3 with 2 clicks and 49 impressions. I can see "calculate impressions from cpm" at position 46.4 with 35 impressions and zero clicks. But I can't see what drove five clicks in one week in late July, because those queries are buried below GSC's threshold.
What I can say: the position on the visible queries (30–47) hasn't moved dramatically. If the position degraded, it happened mostly on the long-tail queries I can't see. The click drop and CTR drop might be a position shift on the invisible majority, not on the queries I'm watching.
This is the same trap as the loan payment calculator in the first post in this series — aggregate position metrics hide a distribution, and the distribution hides the actual behavior.
Finding #3: The book-value calculator appeared from nowhere
The book value calculator had no GSC data for the week of July 25–31. Nothing. Not even impressions.
The automation committed to it on August 2: updated the dateModified field from June to August, changed "Last Updated: June 2026" to August, swapped one internal link for an external link to CSIMarket's industry aggregates, and updated benchmark text.
Here's what happened in the two weeks after:
| Week | Clicks | Impressions | CTR | Position |
|---|---|---|---|---|
| Jul 25–31 | — | — | — | — |
| Aug 1–7 | 2 | 25 | 8.00% | 21.9 |
| Aug 8–16 | 4 | 38 | 10.53% | 21.4 |
The position barely moved (21.9 → 21.4). The clicks doubled.
I'm not claiming the automation commit caused this. The changes were minimal — a date, a link, a few updated sentences. Position didn't move. But the page went from invisible to six clicks in two weeks, and that's notable regardless of what caused it.
When I pulled the query-level breakdown, I got five visible queries covering 11 total impressions:
| Query | Clicks | Impressions | Position |
|---|---|---|---|
| site:valuefy.app | 0 | 5 | 25.0 |
| book value calculator | 0 | 2 | 7.5 |
| reported book values | 0 | 2 | 82.5 |
| book value calculator online | 0 | 1 | 32.0 |
| net book value calculator | 0 | 1 | 70.0 |
That's 11 visible impressions. The page has 56 total impressions in 28 days. The six clicks came from queries below the privacy threshold — the invisible majority again. Whatever someone searched to find and click this page, I can't see it.
Finding #4: The og:title bug from the first post is still live
In the first post in this series, I found that every page on valuefy.app ships two og:title tags: one static tag baked into index.html, and one dynamic tag injected by React Helmet during prerender. I committed to fixing it "tonight." That was months ago.
I curled three pages today — the homepage, the book-value calculator, and the impression calculator. All three still show count=2:
<meta property="og:title" content="Valuefy - Free Business Calculators & Financial Tools">
...
<meta property="og:title" content="Book Value Calculator: BVPS, P/B Ratio & Tangible | Valuefy" data-rh="true">
The static tag in client/index.html (line 23) is still there. The fix — stripping the static og: tags and letting React Helmet be the only source — never happened.
To be clear about what this costs: og:title doesn't affect SERP snippets. Google uses the <title> tag and <meta name="description"> for search results. The duplicate og:title matters for social sharing previews — Twitter cards, LinkedIn posts, Open Graph renders. When someone shares a specific calculator page, some platforms pick up the generic site-wide title instead of the page-specific one.
That's a real bug. It's not why the impression calculator dropped positions. But it's been on the list for months, and it's still broken.
What I'm going to do about it
Fix the og:title bug — for real this time. The change is one file, 15 minutes. The only reason it hasn't happened is that higher-priority work kept appearing. It should be the first commit next week.
Watch the impression calculator — if the position continues degrading past 50, I'll pull a full query-level audit for the specific weeks around the drop and look for a common theme in the queries I can see. The 91% invisible majority means I can't diagnose from inside GSC alone.
Don't claim automation credit for book-value yet — I need two more weeks of data. If the position improves or clicks continue growing, that's evidence the automation changes helped. If it stabilizes and plateaus, it was probably a routine Google volatility event.
Stop updating sitemap lastmod in bulk — the August 1 batch update touched every page at once. If Google recrawled all 160+ pages in a short window, that could explain unexpected position changes. I'll test updating lastmod only on pages that actually changed content.
The uncomfortable lesson
42 commits. One page per day. Verified benchmarks, updated worked examples, fresh external sources. The automation is doing what it's supposed to do.
The impression calculator peaked without me and fell without me. The book-value calculator appeared without warning. Both happened on Google's timetable, not mine.
What the data keeps showing is that on-page automation controls the content signal — and nothing else. Domain authority, competitive dynamics, the invisible query distribution — those are Google's domain. You can improve every sentence on a page and not move positions. You can update a date field and have the page appear in GSC for the first time.
I don't have a clean theory that predicts either of those outcomes. And I'm becoming more skeptical of anyone who claims they do.
Next week I want to know whether the impression calculator stabilizes or keeps sliding. If it's still at position 43 or deeper with one click per week, something structural changed and I'll need to look harder at what the page is actually competing against. If it recovers on its own, then it was volatility — and the lesson is to watch for two weeks before calling anything a trend.
I'll report whatever the data says.
I'm running these experiments on valuefy.app and writing up the findings as I go. If you're building a site in a competitive vertical, debugging ranking drops you didn't cause, or trying to measure whether a daily automation is actually doing anything — I'd like to compare notes. Drop a comment or reach out.
I also run AImiten, where we build AI tooling for companies. This side project is where the ideas get tested first.
Top comments (1)
ngl its wild how gsc just randomly spikes stuff sometimes. makes it so hard to tell if a change actually worked or if it was just a glitch in the matrix