01
Position is not what is limiting us
This is the single most important number on the page, and it reframes most of the work.
Two 28-day windows appear on this page, two days apart, and the difference is deliberate:
- 7 Jul – 3 Aug 2026 — our own curve, the opportunity list (§2) and the anomalies (§5), from our stored data.
- 5 Jul – 1 Aug 2026 — the property totals (§6), two days earlier so the window sits entirely inside Google’s settled period. Comparing a still-moving figure against a settled one is how a reconciliation check goes wrong.
- Week of 27 Jul – 2 Aug 2026 — the cannibalization in §3. That measurement is weekly, not 28-day.
- Current 28 days vs the prior 28 — the decay in §4.
We hold 501 days of history (22 Mar 2025 – 4 Aug 2026). 28 days is the reporting period, not the limit of what we can see — though see the note below the curve.
Our own click-through curve, 7 Jul – 3 Aug 2026, non-brand only:
| Position | Queries | Impressions | Clicks | CTR |
|---|---|---|---|---|
| 5 | 2,983 | 184,919 | 247 | 0.134% |
So the constraint is not where we rank. Moving a query from position 11 to position 5 multiplies a very small number by a very small number. The constraint is that our results are not being clicked when they are seen — a title, snippet, intent-match and SERP-feature problem, not a ranking problem.
A 28-day figure could be a bad month, so we widened the window. cognigy.com at position 5: 2.38 per 1,000 over 28 days, 3.27 over 90 — its advantage is larger over the longer window, not an artefact. The same 90-day query for nice.com does not complete: it aggregates 8.8 million rows and exceeds the database time limit past roughly 28 days.
So the precise position: the 28-day figures are solid and dated, the longer-window check supports the direction on the property where it runs, and the nice.com equivalent 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 upside figures in §2 are modelled at our own position-5 CTR, deliberately. They are therefore an opportunity ranking, not a click forecast — and they are conservative, because they assume we stay as un-clickable as we are today.
02
Act now — ranked by clicks left on the table
Non-brand, relevance-filtered, ≥300 impressions, positions 8–20. 786 queries qualify; these are the top twelve by modelled upside.
| Query | Pos | Impressions | Clicks | Upside | Read |
|---|---|---|---|---|---|
| ai call center | 10.9 | 54,342 | 5 | +67.6 | 6 pages, none on page 1 |
| customer service | 10.8 | 58,610 | 18 | +60.3 | split across 7 pages |
| call recording | 11.0 | 52,853 | 45 | +25.6 | 2 pages |
| rpa | 8.6 | 15,341 | 6 | +14.5 | 5 pages |
| customer care | 11.3 | 11,980 | 3 | +13.0 | single page |
| performance management systems | 10.5 | 9,675 | 0 | +12.9 | single page |
| natural language generation | 10.3 | 9,017 | 0 | +12.0 | single page |
| roster | 9.5 | 7,103 | 1 | +8.5 | check intent |
| contact center ai | 13.7 | 5,489 | 0 | +7.3 | single page |
| detractors meaning | 8.1 | 6,140 | 1 | +7.2 | definitional intent |
| ai customer service | 18.2 | 10,175 | 7 | +6.6 | single page |
| ai workflow automation | 15.5 | 5,213 | 1 | +6.0 | single page |
ai call center, customer service, call recording — are also the three worst cannibalization cases in §3. That is not a coincidence: when six of our pages split the impressions for one query, none of them accumulates the engagement signal that would lift any of them.
But §3 checked the live results page for ai call center, and it changes the plan. None of our six pages 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. Read §3 before scoping this work.
03
Where we compete with ourselves
Week of 27 Jul 2026, non-brand, ≥500 impressions. 180 queries qualify. Locale variants (/de/, /ja/) are excluded — they are hreflang, not conflict — and jump-to #anchor rows are merged rather than summed, because Google re-reports the parent page’s impressions on them.
The worst case, in full
ai call center — six of our pages, 40,557 impressions in one week, one click. Every page, with its average position:
| Our page | Avg position | Type |
|---|---|---|
| /glossary/what-is-call-center-ai | 11.0 | glossary |
| /call-center-ai | 23.9 | product |
| /glossary/what-is-an-ai-call-center | 44.4 | glossary |
| /info/guide-to-the-ai-call-center-how-to-revamp-your-customer-service | 45.1 | guide |
| /glossary/what-is-contact-center-ai-artificial-intelligence | 69.6 | glossary |
| / (homepage) | 88.5 | home |
Three near-identical glossary entries — what is call center ai, what is an ai call center, what is contact center ai — plus a guide, all outranking the actual product page at 23.9.
nextlevel.ai, 8x8, Genesys, Dialpad, retellai, Salesforce, IBM and a Reddit thread.
So this is not six pages splitting page-one visibility. Our best is position 11 — the top of page two. Competitors own page one outright and we are not on it. Those 40,557 impressions are page-two impressions, which is precisely why they convert at one click.
That reframes the fix. Consolidation alone will not win this: it removes the self-competition, but the surviving page still has to beat Genesys and 8x8. SEMrush prices the term at 2,900/mo US, CPC $75.12, difficulty 56 — and that CPC is why it is worth doing anyway.
The rest of the top of the list
| Query | Pages | Impr | Clicks | What the pages actually are |
|---|---|---|---|---|
| customer service | 7 | 14,542 | 5 | /info/transforming-customer-experience · /glossary/live-chat-support · /info/a-guide-to-it-ticketing-systems… · 4 more glossary/FAQ |
| cx | 8 | 12,812 | 16 | six glossary entries (what-is-cx-ai, -cx-platform, -digital-cx, -customer-experience, -cx-solutions, -cx-software) plus /info and /blog |
| call recording | 2 | 14,957 | 12 | /products/recording-management (9.9) vs /glossary/what-is-call-center-call-recording (10.8) |
| tts | 2 | 5,435 | 24 | /glossary/what-is-contact-center-tts (8.5) vs /glossary/text-to-speech (9.5) |
| rpa | 5 | 3,380 | 0 | five competing glossary and product pages |
cx, six of them glossary. Three glossary variants for ai call center. Two for tts, two for call recording — each time a glossary entry competing with the product page that should own the term.
This is one content-architecture decision, not a long list of incidents: the glossary was built term by term without checking what already ranked, so it now competes with itself and with the commercial pages. Resolving ai call center, cx and customer service alone addresses 21 of the competing pages.
How these numbers are derived, and what they are not
Source. A weekly query × page pull from Search Console — which of our pages were shown for which query. That grain has to be requested deliberately: the daily date×query and date×page tables are separate marginals, and summing them cannot reconstruct which page served which query.
Two normalisations, both load-bearing. Locale variants are collapsed, because /de/x and /ja/x are the same page for different markets and are supposed to coexist — counting them as conflict is the easiest way to make this look alarming and be wrong. Fragments are stripped and merged by MAX rather than summed: Google re-reports the parent page’s impressions on #anchor rows, and summing them inflated one query from a true 14,500 to 40,599.
What “position 1.0” means here, and what it does not. Positions are averaged across every query variant folded into that row. Three pages showing 1.0 for customer service does not mean three number-one rankings for the head term — the live check found only one nice.com result in the top 8. It means those pages rank first for narrow long-tail variants. Read position here as a relative signal between our own pages, never as an absolute ranking claim.
Verification status. Live results via Tavily, and volume/CPC/difficulty from SEMrush, both on 5 Aug 2026. ai call center was checked in full. The other rows have not yet had a live SERP check — and since that check overturned the reading of the headline case, it should be done for each before the consolidation work is scoped.
04
Pages losing ground
Page grain — the only grain whose totals reconcile against Google. Current 28 days vs the prior 28, minimum 100 clicks in the prior period. 67 pages qualify; these are the eight steepest.
| Page | Was | Now | Change |
|---|---|---|---|
| /glossary/live-chat | 153 | 14 | −90.8% |
| /websites/nice-world-london | 2,036 | 219 | −89.2% |
| /websites/nice-world | 1,842 | 323 | −82.5% |
| /company/investors/upcoming-event | 104 | 34 | −67.3% |
| /glossary/live-chat-support | 1,091 | 401 | −63.2% |
| /workforce-management | 998 | 392 | −60.7% |
| /resources/teleperformance | 317 | 144 | −54.6% |
| /nice-labs | 133 | 66 | −50.4% |
/websites/nice-world-london and /websites/nice-world are event pages after the event. A ~85% fall is what a finished conference looks like. /company/investors/upcoming-event is the same shape.
/workforce-management is a core commercial page down 61% (998 → 392 clicks). That is the one on this list that should not be here. /glossary/live-chat and /glossary/live-chat-support falling together suggests a category-level change rather than two coincidences.
05
Verify in a browser before treating these as opportunities
Top-5 position, thousands of impressions, zero clicks. That combination looks like the best opportunity on the site and is sometimes not an opportunity at all.
seamless customer service sat at position 2.3 with 11,775 impressions and zero clicks. Three theories were written up, including one about non-human traffic. The actual answer took under a minute in a browser: that results page belongs to Seamless, the food-delivery company. We rank second for a navigational query about ordering dinner.
No query of our own data can detect that. Google reports the query and never the competition. Only a person looking at the page can tell.
| Query | Pos | Impressions | Clicks | Status |
|---|---|---|---|---|
| seamless customer service | 2.3 | 11,775 | 0 | confirmed another brand |
| call center handle time | 1.6 | 3,571 | 0 | resolved — not an opportunity |
| which platforms integrate custom ai chatbots with existing customer communication software? | 4.1 | 7,584 | 0 | next to check — likely AI-Overview answered |
| unified | 8.0 | 14,994 | 7 | ambiguous single word |
| mape | 9.1 | 18,357 | 3 | forecasting statistic — not ours |
| rtt | 7.8 | 10,692 | 5 | ambiguous acronym |
| how to do no caller id | 8.8 | 12,288 | 2 | consumer telephony |
call center handle time at position 1.6 with zero clicks is the one to check first.
- The results page is not about contact centres at all. Top results are
kall8.com, Wikipedia's Telephone call,call.com, the Google Dialer app, Cambridge and Merriam-Webster definitions of "call", anddonotcall.gov. Google reads it as generic telephony vocabulary. - SEMrush prices it at 30 searches/month, CPC $0.00, difficulty 0. Against
ai call centerat 2,900/mo and $75.12 CPC, that is not a rounding error — it is a different category of term.
The 3,571 impressions in Search Console against 30/mo of measured demand is itself the tell: those are long-tail variants aggregated under one label, not people searching this phrase. Both flagged queries have now turned out to be non-opportunities — which is the argument for the check, not against it.
Why the last three rows are marked off-topic rather than opportunities
Roughly 37% of our impressions come from queries outside our category — consumer telephony (no caller id, unknown caller), directory lookup, and statistics terms like mape. Measured, that traffic earns more clicks than the entire CCaaS/CX footprint combined.
That is deliberately not filtered away silently. It is an open question for you: is the off-topic footprint an intentional top-funnel play, or noise we should stop optimising for? The include/exclude list lives in a Supabase table (gsc_icp_terms) and is editable without any code change — so the answer is yours to set, not the pipeline's.
06
Cognigy is now measured — and it converts far better than we do
Access was granted on 4 August. www.cognigy.com now has 500 days of history loaded, back to 23 Mar 2025, on the same daily cadence as www.nice.com.
| 28 days · web | 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 |
Ranking worse and earning more is the clearest available evidence for §1's argument: the constraint is clickability, not position. Whatever Cognigy's titles, snippets and intent-match are doing, it is working roughly twice as well as ours — and that is a transferable, page-level thing rather than a domain-authority thing.
Suggested next step: take the ten highest-CTR Cognigy pages and compare their titles and meta descriptions against ours for equivalent queries. That is a concrete, cheap experiment with a measurable outcome.
One thing already visible: cognigy company (526 impressions) and cognigy official website (41) currently return a nice.com press release with zero clicks. People looking for Cognigy are being intercepted by our own acquisition announcement. Worth a redirect or a canonical decision.
07
How to read this, and what it is not
Page grain is the only one whose totals reconcile against Google — currently 102.8% of Google's own click total, which is why §4 uses it. Query grain holds only 47.9% of clicks, because Google anonymises roughly 43% of clicks out of query-dimension data. It answers which query and nothing else.
Never compute site CTR from query grain. It will look plausible and be about half right.
Impressions × our own non-brand position-5 CTR, minus clicks we already have. It is an opportunity ranking, not a forecast. Because it uses our own poor CTR as the target, it systematically understates the prize — a query fixed properly should beat it.
What is checked automatically, every day
A monitoring workflow now runs after each ingest and writes its findings to ops_errors. It checks that our stored clicks reconcile against Google’s own totals (95–108%), that each table is fresh, that every one of these queries still answers, that the database has headroom, and it maintains the §5 verification queue.
Fourteen checks, all passing as of the last run — with one caveat worth stating. The query behind §2 runs in about 7.4 seconds against a database cut-off of roughly 10. That is comfortable when the system is idle and not comfortable when it is busy: it timed out twice on 5 August, once while Cognigy’s history was loading. The first of two fixes is applied and took it off the hard failure; the second is written and pending. The figures in §2 are unaffected either way — they come from successful runs and reconcile against Google independently.