01
One finding worth the whole month
Our problem is not where we rank. It is that our results are not clicked when they are seen.
Everything in this section is a 28-day measurement, not an all-time one. Two windows appear, and they are two days apart for a reason worth knowing:
- 7 Jul – 3 Aug 2026 — our own click-through curve and the opportunity list. Computed from our stored data.
- 5 Jul – 1 Aug 2026 — the property totals. Two days earlier because Google finalises its figures with a lag, and this window is entirely inside the settled period. Using a later window would compare a still-moving number against a settled one.
We hold 501 days of history for nice.com (22 Mar 2025 – 4 Aug 2026) and 500 for cognigy.com. The 28-day window is the reporting period, not the limit of what we can see.
The measurement
Over 7 Jul – 3 Aug 2026, on non-brand queries, www.nice.com earned 1.30 clicks per thousand impressions at position 5. A healthy commercial result at that position earns 50–100 per thousand. We are roughly forty times below what the ranking should deliver.
The strongest corroboration arrived this week, when cognigy.com came into measurement for the first time. Both rows below are 5 Jul – 1 Aug 2026, web search, taken from Google’s own reported property totals:
| 5 Jul – 1 Aug 2026 · web search | Clicks | Impressions | CTR | Avg position |
|---|---|---|---|---|
| www.nice.com | 51,616 | 10,174,708 | 0.51% | 10.2 |
| www.cognigy.com | 7,101 | 556,369 | 1.28% | 14.0 |
That rules out the expensive explanation. This is not a domain-authority problem that takes years and a link budget to move. It is titles, snippets and intent-match — page-level work, measurable in weeks. And we now have a working example inside the same company to copy from.
A 28-day figure could be a bad month. So we widened the window — and got a straight answer for one property and a wall for the other:
| Position-5 CTR, non-brand | 28 days | 90 days | 365 days |
|---|---|---|---|
| cognigy.com | 2.38 / 1,000 | 3.27 / 1,000 | query times out |
| www.nice.com | 1.30 / 1,000 | query times out | query times out |
Cognigy’s advantage is not a 28-day artefact — over 90 days it is larger, not smaller. For nice.com the equivalent comparison does not compute yet: the query behind it aggregates 8.8 million rows and exceeds the database time limit past about 28 days.
So the precise claim is: the 28-day figures are solid and dated, the 90-day check supports the direction on the property where it runs, and the long-window comparison is pending a performance fix that is already written. We hold the 501 days needed to answer it; we cannot yet query them in one pass.
The concrete opportunity
Over 7 Jul – 3 Aug 2026: 786 queries where we rank on page two with real volume and earn nearly nothing. The three largest, all 28-day figures:
| Query | Impressions (28d) | Clicks (28d) | Avg position |
|---|---|---|---|
ai call center | 54,342 | 5 | 10.9 |
customer service | 58,610 | 18 | 10.8 |
call recording | 52,853 | 45 | 11.0 |
ai call center: six of our own pages compete for that query. In the single week of 27 Jul – 2 Aug 2026 — a week, not the 28-day window, because this measurement is taken weekly — those six pages drew 40,557 impressions and one click. 180 queries show the same pattern that week.
When six pages split the impressions, none accumulates the signal that would lift any of them. But the live results page, checked by hand on 5 Aug, shows none of our six is on page one at all — competitors hold it outright and our best sits at position 11. Consolidation removes the self-competition; it does not by itself win the term. The companion document sets out the full page list.
02
Why these numbers can be trusted, when the last set could not
This is the substance of the past two weeks, and it is worth being blunt about.
An audit of ten real defects from the preceding month found something that determined everything built since:
Not one of them threw an error.
Every one either produced a wrong answer successfully, or produced nothing while reporting success. Our alerting caught exceptions — so it caught none of them. Examples: a published site-CTR figure whose denominator was 163% of Google's own total; four days in which nothing ran at all while every workflow showed as active; two database functions that silently stopped answering.
The first of two fixes is applied and all fourteen checks now pass. The second is written and pending, and takes it further from the edge. The figures in this report are unaffected either way: they come from successful runs and reconcile against Google independently.
It is recorded here rather than quietly dropped now that it passes, because a status page that only mentions problems while they are red teaches people to read it as marketing.
Data freshness per table · every analysis query still answers · database headroom (now 50% of 8 GB, after loading Cognigy’s history) · a standing queue of claims that need a human to look at a live results page before anyone acts on them.
The AI scoring step ran with sampling unpinned, so the same input could produce different opportunity groupings and therefore different scores. That is now fixed, and running the pipeline twice on identical input provably produces no duplicate rows — verified, not assumed.
03
What was built, and what it means for the move to the shared instance
A constraint we had missed: the company n8n instance syncs workflows, not files. Three pieces of recurring operational work lived as scripts on one laptop — meaning nobody but the author could run them. All three are now workflows and will travel.
| Capability | What it does | Status |
|---|---|---|
| Daily ingest | Both properties, every day, reconciled | live |
| Monitoring | 14 checks after each ingest, durable log | live |
| History backfill | Resumable, ledger-driven; 500 days per property | live |
| Weekly structure snapshot | Feeds the cannibalization analysis | live |
| Opportunity scoring | Reproducible; duplicate-proof | live |
| Content briefs | Newest at the top of the shared doc, colour-coded by urgency | live |
| Cross-property competition | Do nice.com and cognigy.com fight for the same queries? | next |
On history: Google deletes a day of Search Console history every day it is not collected. The backfill is deliberately resumable — it records each completed day in the database rather than holding state, so it can be stopped and restarted without losing or repeating work. Cognigy's 500 days were loaded this way in about twenty minutes.
04
Cognigy: what access gave us
Access granted 4 August. Fully measured within a day.
500 days of history back to 23 March 2025, ingested daily on the same cadence as nice.com. Every analysis we run for nice.com now answers for Cognigy on request.
Every report defaults to nice.com only. Cognigy is opt-in per query, and the two can be combined when that is the question — but nothing changes silently.
What that report got wrong was the remedy: it described the fix as "adding a term to a list." It was three changes, and the third was not code at all — it was requesting access to a property we could not see. That access is what turned this from a classification problem into a measurable one.
One immediately actionable item: cognigy company and cognigy official website currently return a nice.com press release and earn zero clicks. People looking for Cognigy are being intercepted by our own acquisition announcement.
05
What we got wrong, and how it was caught
Recorded because the mechanism matters more than the mistakes.
Three published figures were wrong. All three were caught by a person checking the source directly — opening Search Console, or opening the actual Google results page — not by any automated check.
| Published | Actually | Caught by |
|---|---|---|
| Site CTR 0.31% | 0.51% — the denominator double-counted page anchors | recomputation |
| "We rank #1 and nobody clicks" | 88% of impressions sit below the top 3 | recomputation |
| Two keywords "show the identical shape" | One is a growth curve, the other a step — not comparable | reading SEMrush |
| A zero-click query treated as an opportunity | The results page belongs to Seamless, the food-delivery company | opening the SERP |
So the monitoring now maintains a standing verification queue: any query ranking top-5 with 2,000+ impressions and zero clicks is flagged for a human to search by hand before it is treated as work. Two are currently queued. It takes under a minute each, and it has already prevented one wrong roadmap decision.
One number in this report was itself corrected during the week
A comparison of nice.com against cognigy.com initially put nice.com at 24,750 clicks. That was our stored query-level data, which holds only about half the clicks because Google withholds roughly 43% of clicks from query-level reporting. It was compared against Cognigy's property total — two different measurements in one table.
The error was caught by comparing against the Search Console dashboard. The corrected figures are the ones used throughout this report, and the pipeline now checks that ratio on every run so the same mistake becomes visible rather than plausible.
06
What happens next
1. Verify the two flagged zero-click queries in a browser.
2. Consolidate ai call center — six pages to one.
3. Compare the ten highest-CTR Cognigy pages against ours for equivalent queries.
Cross-property competition analysis · a small performance fix so the heaviest query runs in half the time · finish the remaining reproducibility items found in the audit.
Open questions — these need a decision, not more analysis
- Is our off-topic traffic deliberate? Roughly 37% of impressions come from outside our category — consumer telephony, directory lookups, statistics terms — and it earns more clicks than our entire CCaaS/CX footprint combined. Is that an intentional top-funnel play we should keep, or noise we should stop optimising for? This changes what the opportunity list contains.
- Should Cognigy be reported standalone, combined with NiCE, or both? The data supports all three. Reporting cadence and ownership follow from the answer.
- Do we want the remaining NiCE properties measured?
livevox.com,playvox.comandinfo.nice.comare ours and are currently invisible. Each needs the same read-only access grant that Cognigy took a day to produce.