On June 16, the automation improved the P/E Ratio calculator for the second time. The first time was May 13. In the 34 days between those two commits, the page collected 2 clicks.
That's what the first pass left for the second to work with.
The setup
The automation has been running since May 11 — one calculator page per evening, 21:00 UTC, without missing a day. Forty-three commits total. Each one follows the same pattern: pull a page from a queue, write in verified benchmarks, add worked examples, insert internal links, push. The message is always the same: "SEO: improve [X] calculator content — add verified benchmarks, examples, internal links."
This week something different happened. Three of the seven pages in this week's queue had already been through the process:
- P/E Ratio calculator: first improved May 13, second pass June 16 — 34 days apart
- SaaS Valuation calculator: first improved May 14, second pass June 18 — 35 days apart
- Book Value calculator: first improved May 14, second pass June 20 — 37 days apart
The other four (EBITDA, CPM, Google Ads, Occupancy Rate) were genuine first passes. But for these three, this week's commits were the second lap.
I've been writing about this automation for several weeks now — the equity cluster that collected almost nothing, the real estate cluster with 3 combined impressions across 7 pages, the pre/post split that showed the rankings predated the commits. What I hadn't done yet is ask a different question: what does 5 weeks of "first pass result" actually look like when you go back and look at the same page again?
What I expected vs what the data showed
My expectation was that P/E Ratio — the oldest of the three, with the most time since the first pass — would show at least some movement. The content improvements are real: verified benchmarks, a reference to S&P 500 forward P/E (~21x as of 2026), sector comparisons. It's better content than it had before. The meta description is specific. The H1 matches the title tag. I curled the page and it renders cleanly.
What I got back from GSC:
| Page | 28-day impressions | 28-day clicks | Avg position |
|---|---|---|---|
| P/E Ratio calculator | 711 | 2 | 18.8 |
| SaaS Valuation calculator | 17 (visible) | 0 | — |
| Book Value calculator | 0 | 0 | — |
Three pages. Two content improvements each. About two clicks between them over 28 days.
The P/E Ratio number is the most interesting because it looks almost respectable — 711 impressions, position 18.8 — until you pull the query breakdown. When I did, 10 queries accounted for only 62 of those 711 impressions. The remaining 649 impressions are in GSC's privacy graveyard: queries with one or two impressions each that never surface in the report. The 18.8 average position is the arithmetic mean of that invisible distribution.
The one query I could see that matters is "forward pe calculator" — 4 impressions, position 6.5, sitting near the bottom of page 1. The content improvement didn't move it to a click. The second improvement probably won't either.
Finding #1: The second pass inherits whatever the first pass left
This is not a surprise, but it's worth stating precisely. The automation's job is to improve content. It does that well. The commits are clean, the data is verified, the examples are real. But what the content improvements can't do is change Google's prior assessment of the page's authority level.
When P/E Ratio got its first pass in May, the page already existed in Google's index with a certain trust level. The content got better. The trust level didn't change. So 5 weeks later, when the second pass runs, it's improving content on a page that Google has already filed away at position 18 to 22 — and the new content lands on that same authority baseline.
The SaaS Valuation and Book Value data are more direct about it. Zero and near-zero impressions after the first improvement means Google either hasn't indexed the updated content yet, or has indexed it and decided it belongs somewhere below page 5 across essentially all relevant queries. Either way, the second improvement goes into the same hole.
Finding #2: The pages that were never in the queue keep climbing
I checked the current position data against the 28-day averages for the pages that actually generate traffic.
The purchase order generator sits at position 21.3 in the last 7 days — down from an average of 29.2 across the full 28-day window. That's an 8-position improvement inside a single week. It has 20 clicks in 28 days, which makes it the single best-performing page on the site by a wide margin. It has never been in the automation queue.
The impression calculator has 10 clicks and 1,041 impressions in 28 days, sitting at position 34.7 in the most recent week (down from 37.2 over the full month). Also never touched by the automation.
I'm not claiming the automation is hurting these pages by not touching them. I'm noting that neither page needed the automation to move in the right direction. Whatever is driving the improvement — organic crawl budget, external links I'm not aware of, query intent alignment, something else — it is not the content-improvement routine.
The automation made 43 commits. The two pages that are actually delivering clicks didn't receive any of them.
Finding #3: The query match for the CPM calculator deserves a mention
The CPM calculator got its first improvement this week, on June 17. That's too recent to draw any conclusions from — only four days of post-improvement data are in the window. But while I was pulling the query breakdown, one entry stood out:
| Query | Impressions | Position |
|---|---|---|
| 2000000x1000 | 12 | 7.1 |
"2000000x1000" — 2 million times 1,000. The CPM calculator happens to do arithmetic involving millions and thousands (budget divided by CPM gives impressions in thousands). Google apparently decided that a query asking to multiply 2,000,000 by 1,000 is answered by the CPM calculator. Position 7.1. Twelve impressions. Zero clicks.
The content improvement will add better explanations of CPM math. It will not fix the fact that someone searching "2000000x1000" is looking for a calculator, not a CPM marketing tool. The query intent and the page intent are orthogonal.
What I'm going to do
- Let the second passes run. The automation will keep improving pages on schedule. I'm not going to interrupt it.
- Set a checkpoint for July 21. In four weeks, I'll pull the 28-day GSC data for P/E Ratio, SaaS Valuation, and Book Value specifically. If the second pass moved anything, I'll report exactly what moved. If it didn't, I'll report that too.
- Investigate the purchase order generator more carefully. Position moving 8 points in a week without any content change is worth understanding. I want to know what queries drove it and whether the pattern shows up in external link data.
- Check whether the automation queue has a stop condition. If it cycled back to three pages after 40+ days, it may be working through an inventory that's smaller than I thought. I'd rather it skip pages it's already done than run indefinitely on diminishing returns.
The uncomfortable question
Here's the thing that keeps surfacing in these weekly checks.
The automation selects pages by cluster logic — it works through a group, moves to the next group, apparently loops back when it runs out of new pages. Google selects pages to rank by signals that are almost entirely orthogonal to that logic: authority, backlinks, domain trust, query intent match, user engagement signals.
Forty-three commits in, the pages that are performing are exactly the pages the automation never touched. The pages the automation improved most — three of them twice now — are sitting at zero to two clicks and still drifting sideways.
I don't have a clean answer for why the untouched pages are climbing while the improved pages aren't. Possibly the content improvements are fine but the pages are in categories (P/E ratio, SaaS valuation, book value) where established financial-media sites dominate and a new domain can't break through regardless of content quality. Possibly the purchase order generator happens to target a more fragmented query landscape where a new domain can compete. I'll test the checkpoint in July before drawing conclusions.
For now: the second lap has started. The data will tell us whether doing something twice is better than doing it once.
I'm running these experiments on valuefy.app and writing up what I find. If you're building programmatic SEO, fighting the "good content, bad rankings" wall, or running automation that might be improving the wrong pages — I'd be glad to compare notes in the comments.
I also work on AImiten, where we build AI tooling for companies. This side project is where the ideas get stress-tested first.
Top comments (0)