The automation ran every single day this week. Seven SaaS calculators landed in the repo on schedule — Runway on Monday, CAC on Tuesday, LTV on Wednesday, ARR on Thursday, MRR on Friday, Churn Rate on Saturday, Funding on Sunday. The commit messages are uniform: "SEO: improve [X] calculator content — add verified benchmarks, examples, internal links." Seven out of seven days. No misses.
When I pulled the 28-day GSC data this morning to see what that cluster looks like from Google's side, the numbers looked encouraging. Then I pulled the query breakdown for two of them.
The same thing I found in Week 1 is back — except this time it isn't one page. It's the whole cluster.
The setup
Here's where the SaaS calculator cluster stands after this week's improvements:
- Runway calculator — how much time before you run out of money, improved May 17
- CAC calculator — cost to acquire a customer, improved May 18
- LTV calculator — lifetime value, improved May 19
- ARR calculator — annual recurring revenue, improved May 20
- MRR calculator — monthly recurring revenue, improved May 21
- Churn Rate calculator — customer attrition, improved May 22
- Funding calculator — how much to raise and at what dilution, improved May 23
Each improvement follows the same pattern: pull the existing page, add verified benchmarks (sourced, not invented), add a worked example with real numbers, tighten the internal link structure. The automation has been running this routine since late March — 37 consecutive days now, one calculator per day.
The 28-day GSC data showed all six of the SaaS-focused calculators appearing in Google's index with aggregate positions that looked like near-page-1 territory. I expected to need to write a boring update post about flat metrics. Then I drilled in.
What I expected vs what the data showed
Here's what the 28-day page-level data says about the SaaS cluster:
| Calculator | Impressions (28d) | Clicks | Avg Position |
|---|---|---|---|
| ARR calculator | 70 | 0 | 10.2 |
| Runway calculator | 61 | 0 | 8.4 |
| CAC calculator | 11 | 0 | 10.2 |
| LTV calculator | 8 | 0 | 10.4 |
| MRR calculator | 9 | 0 | 30.8 |
| Churn Rate calculator | 8 | 0 | 10.9 |
ARR at position 10.2. Runway at position 8.4. Most of the others clustering around 10–11. On the surface that looks like a cluster approaching page 1 with some volume behind it.
On the surface.
Finding #1: The ARR calculator's invisible 69
ARR calculator: 70 impressions, position 10.2 in the 28-day window. That sounds like a page flirting with page 1. I pulled the query-level breakdown.
GSC returned one query.
mrr / arr hesaplayıcı 1 impression position 28.0
One query. One impression. In Turkish. Position 28.
The other 69 impressions are below GSC's privacy threshold — queries with one or two impressions that don't surface in the query breakdown at all. Dozens of tiny long-tail variations, each generating a crumb of signal, each arriving from a different position scattered somewhere between 2 and 100.
The "position 10.2" is the arithmetic mean of that invisible distribution. There is no ARR calculator query where the page actually ranks at position 10. The aggregate hides a graveyard of sub-threshold queries averaging out to look like progress.
The ARR page title — "ARR Calculator: MRR to ARR, NRR & SaaS Multiples" — is fine. The content quality went up after the May 20 improvement. None of that is the issue. The issue is that the average position metric is lying again, and it's the same lie Week 1's loan payment calculator told.
Finding #2: Runway has the same disease
Runway calculator: 61 impressions, position 8.4. Even more promising than ARR — position 8 would mean top-of-page-1 territory. I pulled the query-level breakdown.
GSC returned one query.
startup runway calculation 1 impression position 49.0
One query. One impression. Position 49.
Sixty impressions below the threshold. The "position 8.4" is the average of one query at 49 and dozens of queries from positions I can't see — some of which are probably in the top 5, which would pull the average sharply upward. A single query at position 2 plus 59 queries at position 50 would produce an average somewhere around 10. That's what I'm looking at.
The Runway calculator isn't ranking at 8.4. It's ranking at different positions across dozens of undetectable searches, and the mean of those positions happens to be 8.4. Those are different sentences.
Finding #3: None of the improved pages showed up in 7-day data
Here's the detail that ties it together. The seven calculator improvements this week landed May 17–23. I pulled the 7-day GSC data (May 16–24) expecting to see at least some of them with new impressions.
None of the SaaS tool pages appear in the 7-day window. Zero impressions across all seven pages in the week they were improved.
The two "runway" and "funding" entries that did show up in the 7-day data were blog posts — /blog/how-to-calculate-and-manage-your-startup-runway-effectively (1 impression, position 72) and /blog/digital-health-ma-how-a-48-funding-drop-creates-acquisition-opportunities (3 impressions, position 31). Not the tool pages.
This doesn't mean the improvements were wasted. Google hasn't crawled, re-rendered, and re-evaluated all seven pages in a week — that's not how the indexing cycle works. The 28-day data showing them at position 10 was from before the improvements. Whether the new content moves anything won't show up for another 2–4 weeks, minimum.
What it does mean: I'm evaluating the SaaS cluster on pre-improvement data. The "near page 1" aggregate is the baseline, not the result.
Finding #4: The clicks are flat
Overall: 9 clicks this week from 853 impressions. The prior week — May 8–15, before this week's improvements — was 11 clicks from 852 impressions. Almost identical numbers, essentially flat.
The 28-day total is 42 clicks. Week 4's published post noted 52 clicks in the equivalent 28-day window. Whether the 10-click decline is signal or noise is unclear — the sample sizes are small enough that it could be either. What I can say is that the growth that was visible through Week 4 has not continued into the past two weeks.
The homepage continues to generate the most clicks: 3 this week from 188 impressions at position 7.8. The purchase order generator came in second at 2 clicks from 194 impressions. Everything else was 0 or 1.
What I'm going to do about it
The cluster-building strategy is working in one narrow sense: the automation is consistent, the pages are in Google's index, and the aggregate position numbers have moved upward from the depths they were at in Week 1. But "position 10 on average" does not mean "ranking for anything useful."
Pick one calculator and verify a real ranking. For a page to generate clicks, it needs a dominant intent query where it actually holds a top-10 position — not an averaged fiction across a long tail. I'll identify which of the SaaS calculators has the most concentrated query distribution rather than the most diffuse, and focus improvement effort there.
Wait for the May improvements to settle before re-measuring. Pulling GSC data on pages that were updated in the same week tells me nothing about impact. I'll check the same 6 pages again in 4 weeks with a split at the improvement date.
Stop counting "average position" as a signal. Every week the aggregate position for a new page looks promising; every week the query breakdown shows the same distribution problem. At this point I should treat page-level average position as noise until I can see at least 5 queries with meaningful impressions each.
Check whether any click-generating query is consistent week-over-week. The homepage and purchase order generator are getting clicks, but I haven't verified whether those clicks come from stable recurring queries or different one-off searches each week.
The uncomfortable lesson
Week 1's version of this lesson was: aggregate position averages hide query distributions. I wrote it, published it, and moved on.
Seven weeks later, the same trap ran on six pages simultaneously — an entire content cluster — and it nearly produced a false-positive "cluster is working" headline in this post. The only thing that stopped it was pulling the query drilldown. Not every week, not for the first time: specifically this week, specifically because I remembered to.
The pattern doesn't go away just because you've seen it before. It shows up every time there's a new page with impressions below the privacy threshold, which is every page with less than a few hundred impressions on any given query. For a new domain without authority, that's nearly everything.
The automation is real. The 37-day streak of improvements is real. The positions averaging around 10 are not real — they're arithmetic on a distribution I can't fully see. I'll check back when those distributions have had a month to respond to the content changes. If the ARR and Runway calculators stay at position 10-with-nothing-visible, I'll say that too.
I'm running these experiments on valuefy.app — a set of SaaS and finance calculators I've been building and optimizing through automated content routines. The weekly data usually punches holes in what I thought I knew about how the experiment was going. If you're fighting the same "the aggregate looks fine but nothing is clicking" problem, drop a comment.
I also run AImiten, where we build AI tooling for companies. This side project is where the ideas get stress-tested before they touch client work.
Top comments (0)