What is best to do with this data
This scan measures what is happening. It becomes strategic the moment it is joined to Search Console and SEMrush, because each dataset answers a question the other structurally cannot. GSC now isolates AI Overview and AI Mode impressions, but shows impressions only, with no clicks, CTR or query breakdown. This scan knows exactly what was asked and which URL got pulled. Neither is useful alone.
Search Console knows which of your URLs appear in AI answers. This scan knows which question triggered it and whether the answer actually recommended NiCE. Put them together and you can score a page for citability before you publish it.
The citability decoupling matrix
189 unique nice.com URLs earned citations. Pull GSC position, impressions and AI impressions for every URL ranking top 20 on the equivalent keyword, then quadrant them.
Ranks well, never cited is your diagnostic set: these pages pass Google's ranking function and fail LLM retrieval. Cited often, ranks poorly is your template set. /enlighten-autosummary converts 15 of 15 citations into a NiCE mention at average rank 1.4. If that page is not a GSC star, you have found a structure LLMs reward and classic ranking does not.
The output is not a report. It is a scoring function you run against drafts before publishing.
Does AEO demand even exist in SEMrush?
These 348 queries average 9.5 words, median 10, and exactly one is under six words. SEMrush keyword databases are short-head instruments. Run all 348 through volume lookup.
Prediction: 70 to 85% return zero or sub-10 volume, including high-value ones like "how should a company evaluate AI chatbots". If that holds, you have proved the buying conversation inside LLMs is invisible to your keyword tooling, which converts AEO from a line item competing with SEO budget into a separate demand channel with its own measurement.
One afternoon of work, and it decides the strategy.
Zero-click erosion, measured per topic
You have variance nobody else has. IVR runs 44.4% nice.com citation coverage; Conversational AI runs 9.7%. Pull GSC clicks, impressions and CTR for the URL clusters behind each topic across the same July to August window.
If CTR falls faster in high-citation topics, you can quantify the trade: each citation costs X clicks and buys Y mentions. That is the number leadership actually wants, and most organisations cannot produce it because they lack the topic-level citation variance needed to isolate it.
Does domain authority predict citation at all?
7,678 unique non-nice URLs from 2,193 domains sit in this corpus. Run those domains through SEMrush for authority and traffic.
The signal already leans one way: fin.ai, rasa.com, gumloop.com, crescendo.ai and thelevel.ai all outperform specifically where NiCE is absent, and none are authority plays. If authority correlates weakly with citation, the link-building playbook does not transfer to AEO and you can stop funding it as an AEO tactic. Contrarian, defensible, and one query away.
The 239 wasted citations, at URL level
29.7% of nice.com citations produce no NiCE mention. The models read the page and recommend someone else. Join those specific URLs to GSC and you will likely find top-of-funnel explainers that rank fine and sell nothing.
| URL | Cites | Converted | Rank |
|---|---|---|---|
| /customer-service-ai/customer-service-ai-agents | 7 | 0 | 6.6 |
| /glossary/what-is-contact-center-software-api | 8 | 1 | 1.5 |
| /agentic-ai/agentic-ai-tools | 24 | 5 | 2.8 |
| /platform/ai-for-cx | 25 | 25 | 1.4 |
/agentic-ai/agentic-ai-tools is the one that should bother you: retrieved 24 times at average rank 2.8 in your weakest topic, converting at 21%. Compare its structure against /platform/ai-for-cx, which converts perfectly, and the fix takes a week.
The version worth building: a citability predictor
Do not build a dashboard. Positive labels you already have, all 22,049 of them. SEMrush solves the hard part, negative labels: for each query take the domains ranking top 20 that were not cited in the LLM answer. That is a real candidate set, not a synthetic one.
Features from GSC (position, impressions, AI impressions), SEMrush (authority, traffic, backlinks) and a crawl (schema types, answer-first structure, tables, freshness, entity markup). Output: score any URL, including unpublished drafts, for probability of citation on a target query cluster. AEO moves from post-hoc measurement to pre-publication QA, and it compounds with every scan wave.
That LLM retrieval behaviour is stable enough over months for a model trained on July data to still predict October. This dataset argues against it: Zendesk AI went 0 to 19.1% in six weeks without gaining citation share, meaning something changed in model behaviour that no content feature would have predicted. Cheapest test: train on wave 1, predict wave 3. If it fails, the predictive framing dies and you are back to monitoring, which is fine but a far smaller project.
Google will keep extending its Search Console AI reports. Query data and clicks are the obvious next additions, and when they land, any measurement layer you build becomes a free product feature. Put the value in the prescriptive layer (what to publish, scored before publishing) rather than the observational layer (what happened). Measurement gets commoditised. Prediction does not, because it needs your specific corpus of query intents and your competitive frame, which Google will never build for you.
The five checks proposed above have been run. One claim in this report was contradicted.
Every claim in this report that outside data can test was tested against the live Google Search Console API (the nice.com and cognigy.com properties) and the SEMrush API on 14–15 Aug 2026. Six claims confirmed, one contradicted, two not testable — and the contradicted one makes its own fix cheaper. Two corrections are recorded below, and a third against the verification pass itself.
This report predicted that 70–85% of its 348 buyer questions would return zero or negligible keyword volume. The true figure is 348 of 348 — every single one.
Only 2 of the 348 appear anywhere among the 21,339 queries Search Console recorded for nice.com in the same window, and both at negligible volume. Two independent instruments, built on completely different data collection, agree that the buying conversation measured here is invisible to the tools currently used to plan content. That is the strategic finding, and it is now corroborated rather than asserted.
Claim ledgern8n note
| Claim in this report | Verdict | What the outside data returned |
|---|---|---|
| AEO demand is invisible to keyword tooling. Predicted 70–85% zero or sub-10 volume. | confirmed, stronger | 348 of 348 return no US volume. Verified one by one, not only in batch: what is contact center software (5 words) returns 70/mo, best ai chatbots for enterprise customer service (7 words) returns nothing. A long-tail effect, not an API limit. |
| The LLM buying conversation is a separate channel. | confirmed | 2 of 348 query strings appear in 21,339 Search Console queries — 10 and 54 impressions, zero clicks. |
| Domain authority does not predict citation. | confirmed | Spearman 0.20 across the 80 most-cited domains. The contrast carries it: amplifai.com at authority 35 earns 259 citations; linkedin.com, youtube.com and microsoft.com at authority 100 earn 135, 130 and 126. |
| Zendesk is not buying its rise with citation share. | confirmed, sharper | zendesk.com holds authority 79, 1,176,823 organic keywords and roughly 12× nice.com's organic traffic — and half its citations (411 against 823). The largest search estate in the category does not convert into answer-engine sourcing. |
| Citation and classic ranking are decoupled. | confirmed | Quadrant built from live data on the 125 cited www.nice.com pages — see below. |
| Product pages carry the citation load; glossary and FAQ punch above their footprint. | confirmed | /products/ earns 332 citations and 3,592 Google clicks. /faq/ earns 42 citations on 14 clicks — real answer-engine value, no measurable search value. |
| A NiCE vs Genesys comparison page is the cheapest P0. | confirmed as AEO, not SEO | genesys vs nice = 50/mo, nice vs genesys = 70/mo. Search Console shows genesys vs at 821 impressions, position 7.7, zero clicks. No keyword-led process would ever fund this page. That is the point, not an objection. |
| “Search Console now isolates AI Overview and AI Mode impressions.” | not available to us | The searchAppearance dimension on this property returns only REVIEW_SNIPPET and TRANSLATED_RESULT. Whatever Google shows in its interface, AI-surface impressions cannot be retrieved through the API, so they cannot be a feature in the citability model proposed in §01. A blocker on that play, not a flaw in this report. |
| Zendesk's surge, the Cognigy entity gap, the sentiment trajectory. | not testable | Properties of model behaviour. Neither instrument observes them. They stand or fall on the next scan wave — which is why action #9 matters more than its position suggests. |
The first-party number that reframes the budget conversationn8n note
Search Console queries in the same window, banded by word count, with the consumer telephony cluster (no caller id and relatives) removed so the comparison is about buying language. The queries measured in this report average 9.5 words.
| Query length | Queries | Impressions | Clicks | CTR | Avg position |
|---|---|---|---|---|---|
| 1–3 words | 9,967 | 3,856,595 | 36,655 | 0.950% | 12.0 |
| 4–5 words | 6,587 | 1,161,856 | 2,582 | 0.222% | 14.1 |
| 6–7 words | 1,947 | 137,314 | 375 | 0.273% | 8.9 |
| 8–9 words | 798 | 51,005 | 86 | 0.169% | 6.3 |
| 10+ words | 1,293 | 117,891 | 35 | 0.030% | 5.9 |
n8n note Google Search Console, nice.com, 05 Jul – 12 Aug 2026. Query-grain data, which recovers 41,025 of the property's 72,339 clicks — shares are within that sample, not site totals.
Conversational queries rank better than any other shape — position 5.9 against 12.0 for head terms — and convert 32× worse. Individually: which platforms integrate custom ai chatbots with existing customer communication software? earns 11,966 impressions at position 4.7 and zero clicks.
Of the 41,025 clicks visible at query grain, 27,497 — 67% — come from queries containing “nice”. Non-brand queries using buyer vocabulary account for 42% of query-grain impressions and 15.5% of query-grain clicks, at 0.30% CTR. Classic organic is now functioning as a brand-defence channel; the vendor-selection conversation is the one measured in this report.
The citability quadrant from §01, executedn8n note
The 125 cited www.nice.com pages joined to Search Console, split at the median citation count (2) and median in-window impressions (6,094).
| Quadrant | Pages | Citations | Google clicks | Impressions | What to do with it |
|---|---|---|---|---|---|
| Fortress — high cite, high SEO | 32 | 411 | 1,771 | 699,026 | Defend. Mostly /products/. Freshness and depth, not new pages. |
| Template set — high cite, low SEO | 30 | 232 | 209 | 68,163 | Study these. 32% of attributable citations on 4% of the clicks — the structure LLMs reward and Google does not. |
| Diagnostic set — low cite, high SEO | 30 | 43 | 2,615 | 1,767,200 | Pages Google serves that LLMs ignore. Retrofit the template structure here first — already crawled and trusted. |
| Neither | 33 | 46 | 168 | 73,453 | Leave alone. |
n8n note The five analyst reprints on resources.nice.com are excluded — that host is not a verified Search Console property, so no data exists to place them with.
Two findings this scan could not have produced alonen8n note
Google has already linked NiCE and Cognigy. The models have not.n8n note
On the cognigy.com Search Console property, nice cognigy earns 524 clicks at position 2.2, and 572 distinct queries contain “nice” (8,480 impressions), including nice cognigy acquisition and nice cognigy careers. The search index treats the acquisition as established fact; the answer engines co-mention at 45.0%.
This decides where the work ships. cognigy.com holds 2,299 organic keywords at authority 36; nice.com holds 45,284 at authority 52, and 85.5% of Cognigy's clicks are its own brand name. Cognigy has no non-brand estate to leverage, so action #3 must be built on nice.com — schema, attribution and joint pages on the stronger domain.
NiCE is cited at rank 1–2 for compliance recording, and the page has been dead since March.n8n note
/products/recording/compliance-center returns 404. It carries 6 citations at average rank 1.83 — and on 10 July, the last date all three retrieval-enabled platforms ran, it was 6 of the 11 nice.com citations for those queries.
Search Console dates the break exactly: the 19 Feb 2026 rename of /products/recording to /products/recording-management built a redirect for the parent and none for its children. Last impression 6 Mar 2026.
Carry this number out of the report: the models cited a URL in July that Google stopped serving in early March. LLM citation memory outlives a dead URL by months — so redirect hygiene is an answer-engine concern with a long tail, not only a search concern.
Correction to §06 — the comparison page exists, ranks third, and is citedn8n note
This report states that “there is no comparison page on nice.com earning citations for these queries” and calls it the highest-severity single defect. That is not accurate. https://www.nice.com/info/nice-cxone-vs-genesys-cloud is live, self-canonical, titled NiCE vs Genesys Cloud CX, and performs well in Google.
| Measure | Value | Read |
|---|---|---|
| Search Console, 16 months | 80,549 imp · 561 clicks | Position 16.1 average across 286 queries. |
Position on nice vs genesys | 3.2 | 76 clicks. On genesys vs nice it ranks 2.4; on nice cxone vs genesys, 1.7. |
| LLM citations earned | 8 | All eight on Perplexity, at ranks 0, 1, 1, 2, 2, 8, 8 — and present in all three waves including 14 Aug. |
| Citations on OpenAI / Anthropic / Google AI | 0 | Those engines retrieve a comparison — just not this one. |
| Structured markup | none | No comparison table, no FAQPage schema, no Product or ItemList markup. 2,178 words of prose that names Genesys 105 times. |
n8n note Verified live and against the Search Console API, 15 Aug 2026.
Across the head-to-head and “compare” queries the citation slot is owned by third parties: cxtoday.com takes 30 citations to nice.com’s 10, with infotech.com, trustradius.com and genesys.com behind it. A single CX Today article — “NiCE CXone vs Genesys Cloud: the ultimate CCaaS battle” — is cited on OpenAI and Perplexity across every wave, at ranks 1, 2, 4 and 6.
Action #1 in §11 reads “publish a comparison page” at medium effort. It should read “add a structured comparison table and FAQPage schema to the page that already ranks third” — small effort, on an asset that is already indexed, already trusted and already converting on Perplexity. The gap is not the page. It is that the page carries none of the structure the retrieval layer rewards, so three of four engines reach for CX Today instead. This is also the clearest worked example in the dataset of the “ranks well, under-cited” diagnostic the Playbook proposes in §01.
Correctionsn8n note
1. The “root / homepage — 59 citations, 7.2%” row in §05 is not the homepage. Those 59 are Google Vertex AI grounding-redirect URLs (vertexaisearch.cloud.google.com/grounding-api-redirect/…) whose destination is opaque in the export. Refetching returns 404 for all 59 because they are short-lived signed URLs — expiry, not a defect. Of 823 nice.com citations, 764 are attributable to a specific page and 59 are not. No work should be scheduled against the homepage on this basis.
2. AI Overview and AI Mode impressions are not retrievable through the Search Console API on this property, so the citability matrix in §01 cannot use them as a feature. The rest of that play is executable and is executed above.
3. A first draft of this verification reported five analyst-report PDFs as deleted. It was wrong. It reconstructed each cited page from its path and refetched it against www.nice.com, discarding the subdomain — the PDFs live on resources.nice.com and are all live. Re-verified against the exported URL strings: 129 of 130 own-domain cited URLs return 200. It is recorded here because a verification pass that hides its own error is not a verification pass.
What this means for the decision to continuen8n note
The diagnosis is finishedn8n note
Three instruments now agree, and no external data contradicted this report. Another scan wave re-establishes what is already established. The measurement question is answered.
The causal question is notn8n note
The 2.55× citation lift is a correlation measured inside one 41-day corpus. Nobody has yet shown that publishing moves citation coverage. That is the only open question that matters.
And measurement is currently brokenn8n note
Three of four platforms have web retrieval disabled after 16 July. Every trend claim here is single-engine. Nothing new should be scanned until that setting is restored.
Continue — but stop expanding measurement and run one causal test. Re-enable retrieval for OpenAI, Anthropic and Google AI. Take roughly ten pages from the diagnostic set above, apply the template-set structure, and hold ten matched pages as a control. Re-scan the same query cohort in four to six weeks. If citation coverage moves on the treated set and not the control, the lever is demonstrated and every action in §11 funds itself. If it does not move, a year of content spend has been saved. Either outcome is worth more than another wave of measurement, and neither needs new tooling.
The claim that non-brand acquisition has largely left classic search for this category rests on one 39-day window of nice.com data, compared against a summer period that carries its own seasonality. It is the assumption everything else leans on. Before committing a year of budget, run the same cut across two more windows — it costs nothing and it is the cheapest way to be wrong early rather than late.
n8n note Verification performed 14–15 Aug 2026 against the Google Search Console API (nice.com and cognigy.com properties) and the SEMrush API. Full working, including the material this section summarises, is in .apsolut/01-reporting/020-aeo-scan-vs-gsc-semrush.html. This section was added to the original export; every figure above it is unchanged.
The lead is real, thin, and built on the past
NiCE is the most-mentioned brand in the category at 34.1% presence, ahead of Genesys at 32.1%. That two-point margin is the entire lead, and it is earned almost entirely in legacy contact-center topics. In the two fastest-growing topics on the board, AI Agents and Conversational AI, NiCE appears in 12.4% and 9.7% of answers respectively.
Those two topics account for 1,283 of 3,649 measured answers. NiCE is dominant across 35% of the market it is losing and marginal in the 35% that is growing.
The five findings that matter
| # | Finding | Evidence | Type |
|---|---|---|---|
| 1 | Owning the citation is the mechanism, not a side effect. When nice.com is cited as a source, NiCE is named in 66.0% of answers. When it is not, 25.9%. On ChatGPT the gap is 18.7% versus 81.4%, a 4.35× lift. | 3,581 response units | Lever |
| 2 | NiCE is close to absent in the largest topic. AI Agents is 871 answers, 24% of the entire measured market. NiCE presence there is 12.4%. Genesys is barely better at 11.3%. Nobody owns it yet. | 871 answers | Opportunity |
| 3 | Zendesk AI went from zero to 19.1% in six weeks. On a like-for-like cohort it was absent in early July and has already overtaken NiCE in Customer Service Automation, 45.0% to 31.0%. | 41-query cohort | Threat |
| 4 | Genesys wins the branded head-to-head query. On "Genesys vs NICE for CCaaS" Genesys appears in 100% of answers; NiCE in 75%. On the rephrased variant NiCE drops to 56%. NiCE is losing a comparison carrying its own name. | 32 answers | Defect |
| 5 | The models do not know Cognigy is NiCE. Cognigy leads Conversational AI at 27.9% while NiCE sits at 9.7%. In 144 of 262 answers naming Cognigy, NiCE is never mentioned. Resolving the entity link would put the combined brand at 37.6%, more than double Genesys. | 262 answers | Structural |
A two-horse race with a very different second horse
Presence rate is the share of the 3,649 measured answers naming a brand. Share of voice weights by how many times a brand is named within an answer. NiCE and Genesys are statistically inseparable on share of voice; NiCE separates on citation rate, where it is more than twice as strong.
| Brand | Presence | Mentions | SOV | Depth | Cite rate | Cites / mention | Positive | Negative | Net |
|---|---|---|---|---|---|---|---|---|---|
| NiCE | 34.1% | 2,654 | 34.6% | 2.13 | 19.5% | 0.57 | 34.4% | 0.53% | +33.8 |
| Genesys | 32.1% | 2,642 | 34.5% | 2.25 | 9.0% | 0.28 | 29.4% | 0.19% | +29.2 |
| Five9 | 17.5% | 1,263 | 16.5% | 1.98 | 5.7% | 0.33 | 27.0% | 0.00% | +27.0 |
| Cognigy | 7.5% | 494 | 6.4% | 1.81 | 2.0% | 0.27 | 22.1% | 0.81% | +21.3 |
| Zendesk AI | 4.1% | 205 | 2.7% | 1.39 | 9.5% | 2.34 | 39.0% | 0.00% | +39.0 |
| Yellow.ai | 3.6% | 192 | 2.5% | 1.45 | 0.4% | 0.10 | 25.5% | 0.00% | +25.5 |
| Boost.ai | 1.3% | 65 | 0.8% | 1.38 | 0.1% | 0.09 | 27.7% | 0.00% | +27.7 |
| Parloa | 1.2% | 145 | 1.9% | 3.22 | 0.8% | 0.62 | 13.1% | 3.45% | +9.7 |
Depth = mentions per appearance. Cite rate = share of answers citing the brand's own domain. Net = positive minus negative as % of mentions.
Genesys is discussed more deeply
Depth of 2.25 against NiCE's 2.13. When Genesys shows up it gets slightly more airtime inside the answer. NiCE wins on frequency of appearance, not prominence within it.
Zendesk AI is running a pure content play
Cited 2.34 times for every mention earned, four times NiCE's 0.57. The models read Zendesk constantly and recommend it rarely. That is a machine mid-conversion, and it is converting fast.
Parloa is the only reputation problem
Depth 3.22 with 13.1% positive and 3.45% negative: discussed at length and unfavourably. Five negative mentions cluster on a single IVR features query on Google AI.
The lead is not evenly distributed across engines
| Platform | Answers | NiCE | Genesys | Five9 | Cognigy | Zendesk AI | Margin |
|---|---|---|---|---|---|---|---|
| Google AI · Gemini 2.5 Flash | 938 | 40.7% | 39.9% | 14.6% | 10.0% | 7.5% | +0.8 |
| Perplexity · Sonar | 828 | 33.7% | 31.4% | 19.3% | 8.5% | 4.1% | +2.3 |
| Anthropic · Claude Haiku 4.5 | 922 | 33.1% | 33.5% | 20.2% | 6.8% | 3.6% | −0.4 |
| OpenAI · GPT-4o mini | 961 | 28.9% | 23.9% | 16.1% | 4.8% | 1.1% | +5.0 |
OpenAI is simultaneously NiCE's weakest platform in absolute terms and its strongest relative to Genesys. Every brand is suppressed there because the model names fewer vendors per answer. That makes ChatGPT the highest-leverage surface in the set, since the citation lift there is 4.35× against 1.81× on Perplexity. Anthropic is the only platform where NiCE trails.
Nothing is wrong, and that is its own problem
Across 2,654 NiCE mentions the engines recorded 912 positive, 1,728 neutral and 14 negative. A negative rate of 0.53% is effectively zero. There is no reputation fire to fight. The real finding is the 65.1% neutral block: two thirds of the time NiCE is named in a list with no reason attached.
The spread across platforms is the actionable part. Perplexity qualifies NiCE positively in 54.3% of mentions. Claude does so in 18.9%. That is not a difference of opinion about NiCE; it is a difference in retrieval behaviour. Perplexity pulls sourced comparative content and inherits its evaluative language. Claude answers more from prior knowledge, and prior knowledge is flat and descriptive.
Positive sentiment tracks citation density, not brand affinity. Perplexity has both the highest NiCE citation rate at 26.8% of answers and the highest positive rate at 54.3%. Anthropic has 18.4% and 18.9%. Feeding the retrieval layer is what converts neutral list-mentions into qualified recommendations.
Every negative mention, in full
All 14 negative NiCE mentions occurred in a four-day window at the start of the measurement period and have not recurred in the five weeks since.
| Date | Query | Topic | Platform | Neg |
|---|---|---|---|---|
| 2026-07-09 | Which vendors provide enterprise proactive customer engagement? | Proactive Engagement | openai | 4 |
| 2026-07-09 | What should enterprises look for in AI agents that automate complex service journeys? | AI Agents | openai | 4 |
| 2026-07-06 | Which interaction analytics tools identify churn risk? | Voice of Customer | googleai | 4 |
| 2026-07-09 | Which AI customer service platforms help enterprises improve customer satisfaction scores? | Customer Service Automation | googleai | 2 |
Two observations. The Proactive Engagement query produced negative sentiment for Genesys and Cognigy simultaneously on the same platform and date, which points to critical framing in the retrieved sources rather than a NiCE-specific issue. And three of the four sit in topics where NiCE presence depends heavily on citations, reinforcing that thin sourcing invites unflattering third-party framing.
Controlling for platform by isolating Perplexity, the only engine instrumented across all three waves, NiCE positive sentiment moved 44.3% to 58.2% to 82.4%. Genesys over the same waves on the same query cohort moved 30.6% to 49.0% to 48.3%. NiCE is not just ahead on sentiment, it is separating.
Citation is the causal lever, and it is measurable
Joining 22,049 citation records to answer-level mention data at the query, platform and date level produces the strongest single relationship in this dataset. It turns AEO from a content-volume guess into a targetable metric.
Two consequences follow immediately. nice.com is currently cited in only 21.1% of the answers where it could be, so the lever is mostly unpulled. And position inside the citation list matters up to a point then falls off a cliff: ranks one through three all convert at 71 to 73%, rank four to five at 63.0%, rank six or lower at 47.0%. Being cited eighth is worth barely more than not being cited at all.
The lift is wildly uneven across engines
| Platform | Named, not cited | Named, cited | Lift | Interpretation |
|---|---|---|---|---|
| OpenAI | 18.7% | 81.4% | 4.35× | Almost pure retrieval dependence. Win the source, win the answer. |
| Anthropic | 24.7% | 67.9% | 2.75× | Strong retrieval weighting, and currently NiCE's only losing platform. |
| Google AI | 33.3% | 69.2% | 2.08× | Highest baseline; brand priors already strong. |
| Perplexity | 27.8% | 50.2% | 1.81× | Cites many sources per answer, so each counts for less. |
A citation earned on ChatGPT is worth roughly 2.4 times a citation earned on Perplexity in mention terms. Schema, crawlability and freshness work should be sequenced against OpenAI retrieval behaviour first, not against the platform where NiCE already looks healthy.
What nice.com is actually cited for
nice.com is the single most cited domain in the corpus with 823 citations across 199 queries, ahead of zendesk.com at 411 and genesys.com at 336. Average rank tells a different story: genesys.com averages position 2.2 and five9.com 1.8, against nice.com at 3.7. NiCE is cited more often and less prominently.
Top pages earning citations
| URL path | Cites | Rank |
|---|---|---|
| /products/interactive-voice-response-ivr | 59 | 1.3 |
| /products/omnichannel-routing | 43 | 2.0 |
| /products/interaction-analytics | 77 | 3.5 |
| /products/automated-summary | 27 | 2.7 |
| /platform/ai-for-cx | 25 | 1.4 |
| /agentic-ai/agentic-ai-tools | 24 | 2.8 |
| /call-center-ai/ai-call-routing | 20 | 2.7 |
Citations by site section
| Section | Cites | Share |
|---|---|---|
| /products/ | 332 | 40.3% |
| /info/ | 111 | 13.5% |
| root / homepage | 59 | 7.2% |
| /glossary/ | 53 | 6.4% |
| /faq/ | 42 | 5.1% |
| /agentic-ai/ | 33 | 4.0% |
| resources. PDF assets | 32 | 3.9% |
| /blog/ | 27 | 3.3% |
Product pages carry the citation load. Glossary and FAQ together contribute 95 citations at 11.5%, a strong return for their footprint and a validation of the structured-content approach. The /agentic-ai/ cluster earns 33 citations, evidence that models will retrieve NiCE agentic content when it exists. There is simply not enough of it.
Analytics, quality and workforce are held outright
The lead is not distributed. It concentrates in the measurement-and-optimisation half of the contact centre, the categories NiCE has owned for two decades, and it is often overwhelming there.
| Topic | Answers | NiCE | Genesys | Five9 | Cognigy | Zendesk AI | Margin |
|---|---|---|---|---|---|---|---|
| Quality Management | 28 | 67.9% | 28.6% | 3.6% | 0.0% | 3.6% | +39.3 |
| Voice of Customer | 68 | 23.5% | 1.5% | 0.0% | 0.0% | 0.0% | +22.0 |
| Customer Service Automation | 380 | 33.9% | 19.7% | 11.3% | 8.9% | 11.8% | +14.2 |
| Interaction Analytics | 421 | 61.3% | 49.2% | 24.5% | 1.4% | 2.1% | +12.1 |
| Omnichannel Routing | 184 | 50.5% | 40.8% | 36.4% | 1.1% | 0.0% | +9.7 |
| Contact Center | 104 | 53.8% | 47.1% | 25.0% | 4.8% | 5.8% | +6.7 |
| Recording Compliance | 194 | 32.0% | 29.4% | 21.1% | 0.0% | 1.0% | +2.6 |
| Integrations | 126 | 24.6% | 23.0% | 15.9% | 0.0% | 0.0% | +1.6 |
| AI Agents | 871 | 12.4% | 11.3% | 4.4% | 7.8% | 6.9% | +1.1 |
| Workforce Management | 185 | 68.6% | 67.6% | 40.0% | 0.5% | 5.9% | +1.0 |
| Performance Management | 179 | 63.1% | 67.6% | 46.9% | 0.0% | 2.8% | −4.5 |
| Conversational AI | 412 | 9.7% | 16.7% | 2.9% | 27.9% | 1.2% | −7.0 |
| IVR | 187 | 44.4% | 57.8% | 36.4% | 13.9% | 0.0% | −13.4 |
| Proactive Engagement | 131 | 43.5% | 58.8% | 32.8% | 2.3% | 0.0% | −15.3 |
| CCaaS | 32 | 65.6% | 100.0% | 0.0% | 18.8% | 3.1% | −34.4 |
| Knowledge Management | 47 | 0.0% | 0.0% | 0.0% | 0.0% | 6.4% | 0.0 |
Topics under 40 answers carry wide confidence intervals. Workforce Engagement, Customer Experience and Customer Service are omitted at n=4 each.
The strongest single position: Interaction Analytics
421 answers, 61.3% NiCE presence, and nice.com is the number one cited domain in the topic at 4.27% share, ahead of amplifai.com and replicant.com. This is the only topic where NiCE wins the answer and owns the source. Specific queries run far higher: "What are the best AI interaction analytics platforms for contact centers?" returns NiCE in 90% of answers against Genesys at 30%.
Queries where NiCE is close to unbeatable
| Query | NiCE | Genesys |
|---|---|---|
| Best AI interaction analytics platforms | 90% | 30% |
| Best interaction analytics platforms | 88% | 62% |
| Best conversation analytics platforms | 86% | 38% |
| Best customer feedback analytics platforms | 85% | 25% |
| Automate after-call work | 80% | 33% |
| Agent performance management vendors | 81% | 62% |
| Best IVR software for enterprise | 92% | 100% |
The branded query NiCE is losing
| Query | NiCE | Genesys |
|---|---|---|
| Genesys vs NICE for CCaaS | 75% | 100% |
| Which platform is better, Genesys vs NICE for CCaaS? | 56% | 100% |
| Verint vs NICE for workforce engagement management | 100% | 50% |
Only 4 of 16 answers cite nice.com at all on the first query, and 41 of 47 NiCE mentions across it are neutral.
A prospect typing NiCE's own name into ChatGPT gets an answer that mentions Genesys every time and NiCE around two thirds of the time. There is no comparison page on nice.com earning citations for these queries. This is the cheapest fix in the entire report and the most embarrassing gap.
Half the lead is brand memory, not retrieved evidence
Splitting each topic by whether nice.com was cited exposes two entirely different mechanisms holding up NiCE's presence, carrying very different risk profiles.
| Topic | Answers | Coverage | Presence when cited | Presence when not | Ratio | Mechanism |
|---|---|---|---|---|---|---|
| Workforce Management | 185 | 9.7% | 66.7% | 68.9% | 0.97× | Prior-driven |
| Contact Center | 104 | 10.6% | 54.5% | 53.8% | 1.01× | Prior-driven |
| Performance Management | 179 | 14.5% | 76.9% | 60.8% | 1.26× | Prior-driven |
| Interaction Analytics | 421 | 24.5% | 86.4% | 53.1% | 1.63× | Mixed |
| Recording Compliance | 194 | 20.6% | 50.0% | 27.3% | 1.83× | Mixed |
| Omnichannel Routing | 184 | 33.2% | 85.2% | 33.3% | 2.56× | Retrieval-driven |
| Proactive Engagement | 131 | 30.5% | 90.0% | 23.1% | 3.90× | Retrieval-driven |
| Conversational AI | 412 | 9.7% | 30.0% | 7.5% | 4.00× | Retrieval-driven |
| Customer Service Automation | 380 | 42.1% | 61.2% | 14.1% | 4.34× | Retrieval-driven |
| IVR | 187 | 44.4% | 77.1% | 18.3% | 4.21× | Retrieval-driven |
| AI Agents | 871 | 12.4% | 38.9% | 8.7% | 4.47× | Retrieval-driven |
| Integrations | 126 | 23.8% | 63.3% | 12.5% | 5.06× | Retrieval-driven |
| Voice of Customer | 68 | 22.1% | 93.3% | 3.8% | 24.6× | Fully retrieval-driven |
Prior-driven topics
Where NiCE is baked into the model weights. Workforce Management returns NiCE 68.9% of the time with no source at all. That is twenty years of category ownership showing up in training data. It is free presence, and the most fragile asset on this page: it does not respond to content work, it decays as models refresh, and it is invisible to any competitor content strategy until it collapses.
Retrieval-driven topics
Where presence is bought entirely with citations. Voice of Customer swings from 3.8% to 93.3% depending on a single citation. AI Agents runs 8.7% to 38.9%. These respond directly to content and schema work, and NiCE currently covers only 12.4% of AI Agents answers with a citation.
The bottom-left quadrant contains 1,398 answers, 38% of the measured market, where NiCE averages 11.7% presence. The top-left quadrant contains 662 answers of high presence resting on almost no owned citations. Between them, that is the strategic picture: a large exposed flank and a large undefended asset.
The buyer-education layer is unclaimed by everyone
Classifying all 348 queries by linguistic archetype rather than topic exposes a cleaner structure than the topic taxonomy does. The market splits into a fiercely contested recommendation layer and a completely empty education layer.
| Query archetype | Answers | Queries | NiCE | Genesys | Five9 | Cognigy | Cite rate | Positive |
|---|---|---|---|---|---|---|---|---|
| Vendor list — "Which vendors offer…" | 1,407 | 138 | 44.1% | 44.0% | 23.2% | 9.7% | 22.0% | 31.4% |
| Superlative — "What are the best…" | 1,175 | 100 | 45.8% | 40.3% | 24.3% | 9.9% | 21.2% | 39.1% |
| Other / narrative | 382 | 38 | 11.5% | 13.4% | 6.3% | 4.5% | 12.6% | 42.6% |
| Buying criteria — "What should enterprises look for…" | 177 | 6 | 2.3% | 2.3% | 0.0% | 0.0% | 16.4% | 0.0% |
| Capability — "How can AI agents…" | 160 | 23 | 8.8% | 2.5% | 0.0% | 0.0% | 19.4% | 13.0% |
| Feature criteria — "What features matter most…" | 154 | 23 | 5.2% | 2.6% | 0.0% | 0.0% | 18.8% | 31.6% |
| Evaluation framework — "How should a company evaluate…" | 134 | 8 | 0.0% | 0.0% | 0.0% | 0.0% | 6.0% | — |
| Business case — "What is the business case for…" | 40 | 10 | 0.0% | 0.0% | 0.0% | 0.0% | 10.0% | — |
| Head-to-head — "X vs Y" | 20 | 2 | 80.0% | 90.0% | 10.0% | 20.0% | 25.0% | 10.9% |
Two archetypes return zero vendor mentions for any brand in the category. When a buyer asks how to evaluate customer service chatbots, or what the business case for AI agents is, the models answer with generic frameworks assembled from consultancies and content marketers. No vendor is in the room. That is 174 answers of pure whitespace, and adding buying criteria and feature criteria brings the near-empty education layer to 505 answers, 13.8% of the measured market, where combined vendor presence is under 3%.
Winning one point of presence in the recommendation layer requires taking it from Genesys, who is fighting back. Winning the education layer requires only publishing evaluation frameworks and ROI models the retrieval layer can find. There is no incumbent to displace. NiCE's citation rate in the evaluation-framework archetype is already the lowest in the set at 6.0%, so even the sourcing job is undone.
62 queries return NiCE zero times
Of 215 queries with a meaningful sample, 62 have never once produced a NiCE mention across any platform on any date. Those represent 799 answers, and they cluster with almost comic consistency.
| Query with zero NiCE presence | Topic | Importance | Answers | Genesys |
|---|---|---|---|---|
| What AI agent software is best for teams that need to resolve customer issues without human handoffs? | AI Agents | 5 | 34 | 5.9% |
| What are the best enterprise AI agents to handle account changes across backend systems? | AI Agents | 5 | 30 | 0.0% |
| What should enterprises look for in AI agents that recommend next best actions? | AI Agents | 5 | 30 | 0.0% |
| What should enterprises look for in AI agents that support agents during live interactions? | AI Agents | 5 | 30 | 3.3% |
| What should enterprises look for in AI agents that handle account changes across backend systems? | AI Agents | 5 | 29 | 0.0% |
| What AI agent software is best for teams that need to handle account changes across backend systems? | AI Agents | 5 | 29 | 0.0% |
| What should enterprises look for in AI agents that summarize customer conversations? | AI Agents | 5 | 24 | 8.3% |
| What are the best customer service chatbots for enterprise customer service? | Conversational AI | 5 | 22 | 0.0% |
| What are the best AI chatbots for enterprise customer service? | Conversational AI | 5 | 19 | 10.5% |
| How should a company evaluate AI chatbots? | Conversational AI | 5 | 19 | 0.0% |
| How should a company evaluate customer service chatbots? | Conversational AI | 5 | 19 | 0.0% |
| How should a company evaluate virtual agents? | Conversational AI | 5 | 19 | 0.0% |
| How should a company evaluate conversational self-service? | Conversational AI | 5 | 18 | 0.0% |
| What is the best AI software to improve customer satisfaction scores? | Customer Service Automation | 5 | 15 | 6.7% |
| What is the best AI software to deflect repetitive support requests? | Customer Service Automation | 5 | 15 | 0.0% |
| How should a company evaluate voice bots? | Conversational AI | 5 | 15 | 0.0% |
| How should a company evaluate customer support automation? | Conversational AI | 5 | 15 | 0.0% |
| How should a company evaluate digital assistants? | Conversational AI | 5 | 15 | 0.0% |
| Which vendors provide enterprise-grade digital assistants? | Conversational AI | 5 | 15 | 6.7% |
| Which AI customer service platforms help enterprises automate knowledge retrieval? | Customer Service Automation | 5 | 15 | 40.0% |
| What are the best digital assistants for enterprise customer service? | Conversational AI | 5 | 15 | 46.7% |
| How should a company evaluate AI virtual assistants? | Conversational AI | 5 | 14 | 0.0% |
| Which vendors lead in real-time analytics? | Interaction Analytics | 5 | 12 | 0.0% |
| Which vendors lead in text analytics? | Interaction Analytics | 5 | 12 | 0.0% |
| What features matter most in skills-based routing? | Omnichannel Routing | 3 | 11 | 0.0% |
Top 25 of 62 zero-presence queries by answer volume. Queries with fewer than 8 answers excluded.
The two Interaction Analytics entries deserve attention because they contradict the topic-level picture. NiCE holds 61.3% presence in Interaction Analytics overall, yet "Which vendors lead in real-time analytics?" and "Which vendors lead in text analytics?" both return zero. The topic is won; those two phrasings are not. Presence is query-shaped, not topic-shaped, and topic averages hide total losses.
Modelled opportunity: what 40% citation coverage buys
Applying each topic's own measured conversion rates and raising nice.com citation coverage to 40% per topic, a level already exceeded in Customer Service Automation and IVR:
| Topic | Answers | Current coverage | Current presence | Modelled | Gain | Extra appearances |
|---|---|---|---|---|---|---|
| AI Agents | 871 | 12.4% | 12.4% | 20.8% | +8.4 | 73 |
| Conversational AI | 412 | 9.7% | 9.7% | 16.5% | +6.8 | 28 |
| Interaction Analytics | 421 | 24.5% | 61.3% | 66.4% | +5.1 | 21 |
| Voice of Customer | 68 | 22.1% | 23.5% | 39.6% | +16.1 | 11 |
| Integrations | 126 | 23.8% | 24.6% | 32.8% | +8.2 | 10 |
| Recording Compliance | 194 | 20.6% | 32.0% | 36.4% | +4.4 | 9 |
| Proactive Engagement | 131 | 30.5% | 43.5% | 49.9% | +6.4 | 8 |
| Omnichannel Routing | 184 | 33.2% | 50.5% | 54.1% | +3.6 | 7 |
| Performance Management | 179 | 14.5% | 63.1% | 67.2% | +4.1 | 7 |
| Total | 3,489 | — | — | — | +5.0 pts | 174 |
A five-point overall presence gain would move NiCE from 34.1% to roughly 39%, opening a seven-point gap on Genesys where today there is two. Two thirds of that gain comes from AI Agents and Conversational AI alone, the same two topics identified as the exposed flank.
Two competitors are moving, and one of them is inside the house
Threat 1 — Zendesk AI, absent to third place in six weeks
Isolating a like-for-like cohort of 41 queries measured in all three waves removes query-mix noise. On that cohort Zendesk AI went from 0.0% presence in early July to 16.7% in late July to 19.1% in mid-August. No other brand moved like that.
It has already overtaken NiCE where NiCE was winning
In the 14 August wave, Customer Service Automation returned Zendesk AI in 45.0% of answers against NiCE's 31.0%. In AI Agents it returned 28.6% against NiCE's 13.3%. Both were topics where NiCE led on pooled full-period figures.
It is not buying this with citation share
zendesk.com's share of Perplexity citations was 2.49% in wave one and 2.62% in wave three, essentially flat. The surge is not more retrieved pages. It is Zendesk being newly treated as a category-appropriate answer, which is harder to counter than a content push and will not show up in a citation-share dashboard.
Threat 2 — Cognigy is competing with its own parent
Cognigy leads the Conversational AI topic at 27.9% presence, ahead of Genesys at 16.7% and nearly three times NiCE's 9.7%. Across the full period, 262 answers named Cognigy. In 144 of them, NiCE was never mentioned.
The entity gap, quantified
| Answers naming Cognigy | 262 |
| Of those, also naming NiCE | 118 · 45.0% |
| Baseline NiCE presence | 35.5% |
| Co-mention lift from the acquisition | +9.5 pts |
A genuine parent-subsidiary link in the model's knowledge would drive co-mention far above 45%. The engines treat Cognigy as an independent vendor.
What resolving it is worth
| Conversational AI, NiCE alone | 9.7% |
| Conversational AI, Cognigy alone | 27.9% |
| Combined entity | 37.6% |
| Genesys in the same topic | 16.7% |
The same arithmetic applies in IVR (44.4 + 13.9 = 58.3) and Customer Service Automation (33.9 + 8.9 = 42.8). The asset is already earned; it is not attributed.
Cognigy's own presence is declining on the like-for-like cohort, from 8.7% in wave one to 4.7% then 5.6%. Whatever entity consolidation is happening in the models is currently subtracting from Cognigy without adding to NiCE. That is the worst of both outcomes and argues for treating the entity link as urgent rather than a slow-burn brand exercise.
Threat 3 — the borrowed lead in Workforce Management
NiCE holds 68.6% presence in Workforce Management on 9.7% citation coverage, and presence when cited (66.7%) is actually lower than presence when not cited (68.9%). The position is entirely a model prior. The topic's citation layer is led by g2.com at 61 citations and cxtoday.com at 44, with nice.com eighth at 22. Every model refresh shifting weight from parametric knowledge toward retrieval erodes this position, and there is no owned content underneath to catch the fall.
What the next ninety days look like
This dataset contains three measurement waves with uneven query mixes and a 12-day gap between waves two and three. Three points do not support confident extrapolation, and a naive linear fit produces absurd outputs: it projects Genesys at 80.4% and Zendesk AI at 73.6% presence by mid-November, impossible in a market where answers name three to five vendors. The projections below are bounded scenarios anchored on wave-two-to-wave-three movement, not trend-line extensions.
Base case, no intervention
| Brand | W1 | W2 | W3 | W2→W3 | Mid-Nov range | Reasoning |
|---|---|---|---|---|---|---|
| NiCE | 31.4% | 38.2% | 39.0% | +0.8 | 38–42% | Decelerating after a strong W1 to W2 jump. Gains now marginal without new content. |
| Genesys | 30.9% | 32.1% | 33.0% | +0.9 | 33–37% | Steady, slower climb. Tracks NiCE closely; no sign of breaking away or falling behind. |
| Zendesk AI | 0.0% | 6.7% | 23.0% | +16.3 | 28–40% | Widest band in the table. Cohort data shows deceleration (16.7→19.1) while Perplexity shows acceleration. Both reads defensible. |
| Five9 | 19.2% | 18.8% | 21.0% | +2.2 | 20–24% | Stable mid-tier, no strategic movement visible in the data. |
| Cognigy | 8.2% | 8.5% | 10.0% | +1.5 | 6–11% | Perplexity shows mild growth, the like-for-like cohort shows decline from 8.7% to 5.6%. Direction genuinely unclear. |
Perplexity-only figures, the single engine instrumented across all three waves. See caveats in Method.
Four calls, with the evidence each rests on
| # | Prediction | Confidence | Rests on / would be falsified by |
|---|---|---|---|
| 1 | Zendesk AI passes Five9 into third place on pooled presence within one quarter. Already ahead in the two AI-native topics and climbing on both trend measures. | High | 0 to 19.1% on the fixed cohort; 45.0% in W3 Customer Service Automation. Falsified if W4 shows Zendesk flat or declining there. |
| 2 | NiCE's Workforce and Performance Management presence declines before its AI Agents presence rises. Prior-driven positions decay with model refreshes; retrieval-driven positions only move when content moves. | Medium | 9.7% and 14.5% citation coverage against 68.6% and 63.1% presence. Falsified if either holds above 65% with coverage under 15% for two more waves. |
| 3 | Evaluation-framework and business-case archetypes stay at 0% vendor presence unless a vendor deliberately publishes into them. Nothing suggests organic drift. | High | 174 answers, 18 queries, zero mentions for any of eight brands across four platforms over six weeks. Falsified by any vendor appearing there in W4. |
| 4 | NiCE and Genesys stay within four points of each other on pooled presence through Q4 absent a deliberate content programme. | Medium | W2→W3 deltas of +0.8 and +0.9; pooled SOV of 34.6% and 34.5%. Falsified by either brand gaining more than four points in one wave. |
If citation coverage reaches 40% across the nine addressable topics, presence moves from 34.1% to roughly 39%, a seven-point gap over Genesys instead of two. The dominant uncertainty is not whether the lever works, since the conversion rates are measured directly rather than assumed, but the lag between publishing and re-indexing, which this dataset cannot observe. The one usable signal: NiCE's citation rate on Perplexity rose from 25.8% to 39.0% across the measurement period while positive sentiment rose from 44.3% to 82.4%, indicating the retrieval layer responds within weeks rather than quarters.
Ranked by measured return, not by effort
| # | Action | Target | At stake | Effort | Priority |
|---|---|---|---|---|---|
| 1 | Publish a NiCE vs Genesys CCaaS comparison page and make it retrievable. Structured comparison table, explicit head-to-head framing, FAQ schema. Currently only 4 of 16 answers on this query cite nice.com at all. | Branded head-to-head | 32 | S | P0 |
| 2 | Build the evaluation-framework content layer. One page per "How should a company evaluate X" pattern: chatbots, virtual agents, voice bots, digital assistants, conversational self-service, support automation. Vendor-neutral criteria with NiCE as the worked example. | Evaluation + business case | 174 | M | P0 |
| 3 | Resolve the Cognigy entity link. Consistent "Cognigy, a NiCE company" attribution across both domains, cross-linked sameAs and parentOrganization in Organization schema, joint product pages. Target co-mention above 70%. | Conversational AI, IVR, CSA | 262 | M | P0 |
| 4 | Expand the /agentic-ai/ cluster to full topic coverage. It already earns 33 citations at average rank 2.8, proving retrievability. Target the seven zero-presence AI Agents queries: handoff-free resolution, backend account changes, next-best-action, live agent assist, conversation summarisation. | AI Agents | 871 | L | P1 |
| 5 | Underwrite the borrowed positions with owned content. Workforce and Performance Management hold 68.6% and 63.1% presence on 9.7% and 14.5% coverage. Add comparison, buyer-guide and benchmark content so the position survives a model refresh. | WFM, Perf Mgmt, Contact Center | 468 | M | P1 |
| 6 | Sequence technical AEO work against OpenAI retrieval first. Citation lift is 4.35× on ChatGPT versus 1.81× on Perplexity, and OpenAI is NiCE's lowest-presence platform at 28.9%. Schema, llms.txt, crawlability and freshness should be validated there before elsewhere. | All topics, OpenAI surface | 961 | M | P1 |
| 7 | Fix citation rank, not just citation presence. nice.com averages rank 3.7 against genesys.com at 2.2 and five9.com at 1.8. Conversion holds at 71–73% through rank 3 and collapses to 47.0% at rank 6+. Prioritise depth, freshness and specificity on already-cited pages over publishing new ones. | Existing product pages | 823 | S | P2 |
| 8 | Attack query phrasings, not topics. "Which vendors lead in real-time analytics" and "text analytics" both return zero despite NiCE holding 61.3% of the parent topic. Map exact-phrase coverage across the 62 zero-presence queries. | Zero-presence long tail | 799 | M | P2 |
| 9 | Close the measurement gap before the next wave. 551 of 1,000 defined queries have never run, including 89 at importance 5, and citation instrumentation only captured Perplexity in waves two and three. Trend conclusions are currently single-engine. | Measurement programme | — | S | P0 |
Presence rate is the scoreboard but it is not steerable. Citation coverage, the share of answers where nice.com appears in the top three cited sources, is the leading indicator. It converts at a measured 2.55× and it is the only number in this report that content and engineering work moves directly. Current value is 21.1% overall against a 40% target.
Download the source datasets
Every figure in this report is recomputed from these five exports. Filenames are unchanged from the original export so they can be verified against the source system. Place the files alongside this HTML file for the links to resolve.
497a0bb7…. Excluded from analysis to avoid double counting. Included here for completeness.What this dataset can and cannot support
Caveats that materially affect the conclusions
| Issue | Impact | How it was handled |
|---|---|---|
Post-publication reconciliation against the production database (14 Aug 2026). Stored totals: 3,743 answers and 22,049 citations; this report's reconstructed denominator of 3,649 undercounts by 94 rows (2.5%) because it was derived from the daily analytics rollup — a dedicated Answers export (one row per answer, joins citations on response_id) now exists and should feed the next revision. Additionally, 310 of 1,322 scans (23%) FAILED before storing any answer and are invisible to every number here. | Medium Percentages built on n=3,649 are directionally sound but carry ±≈0.1–0.3pt denominator noise; nothing here counts the failed runs. | Citation count verified exact. Answer percentages left as computed; next revision should rebuild from the Answers export. |
Root cause of the Perplexity-only citation window (verified). Every ChatGPT, Claude and Gemini answer after 16 July carries search_status = not_triggered (290 answers each, uniformly) — web retrieval was not requested for those platforms, coinciding with the mid-July scan-execution/citations-toggle changes. This is a workspace setting, not an instrumentation drop. | High Waves 2–3 measure parametric model memory for 3 of 4 platforms, not the live web. Perplexity-only trend claims in this report remain valid. | Re-enable citations/search for OpenAI, Anthropic and Google AI under Settings → Scan execution before the next wave; the search_status column in the exports is the per-answer guard. |
| Citation instrumentation is Perplexity-only after 16 July. Waves 2 and 3 contain zero citation records for Anthropic, Google AI and OpenAI, against 4,079 / 4,296 / 2,354 in wave 1. | High A naive read shows nice.com citation share falling from 4.15% to 2.47%. That is a measurement artifact, not a decline. | All trend claims computed Perplexity-only. Controlled, nice.com citation share is flat (2.57 / 2.08 / 2.47) and NiCE's citation rate rose from 25.8% to 39.0%. |
| Competitor rows are mention-only. The analytics export records every scan for NiCE but only mentioned scans for competitors. | Medium Competitor mention rate cannot be computed as mentioned / scans directly from the export. | NiCE's scan rows reconstruct the shared denominator of 3,649 answers; all competitor rates computed against it. |
| Uneven query mix across waves. Wave 1 covers 269 queries and 2,583 answers; wave 3 covers 102 queries and 406 answers. | Medium Pooled trend lines conflate real movement with mix change. | A fixed cohort of 41 queries measured in all three waves is used for every like-for-like claim. |
| Only 348 of 1,000 defined queries have run. 551 never executed, including 89 at importance 5, concentrated in Contact Center (175) and Customer Service Automation (74). | Medium Topic rates reflect the queries that ran, not the topic as defined. | Answer counts shown on every topic row. Topics under 40 answers flagged or excluded. |
| Three waves is not a time series. Waves separated by an 11-day and a 16-day gap. | Medium Linear extrapolation is not defensible. | Forecasts are bounded scenarios anchored on the W2→W3 interval, with falsification criteria stated. |
| Sentiment is engine-assigned, not human-validated. | Low Directional confidence good; absolute values should not be quoted externally. | Negative mentions enumerated individually rather than summarised, so the sample can be audited. |
| Two citation exports supplied were byte-identical. | Low No impact once deduplicated. | Only one copy analysed. No double counting. |
Definitions
| Term | Definition |
|---|---|
| Answer / response unit | One model response to one query on one platform on one date. 3,649 in scope. |
| Presence rate | Share of answers naming the brand at least once. The primary scoreboard metric. |
| Share of voice | Brand's total mentions divided by all brands' mentions (7,660). Weights repeated mentions within one answer. |
| Depth | Mentions divided by appearances. Above 2.0 means the brand is discussed rather than listed. |
| Citation coverage | Share of answers where the brand's own domain appears in the retrieved source list. 21.1% for nice.com. |
| Citation-to-mention lift | Presence rate when own domain is cited, divided by presence rate when it is not. 2.55× overall. |
| Prior-driven topic | Lift ratio below roughly 1.5×. Presence comes from model parametric knowledge, not retrieval. |
| Retrieval-driven topic | Lift ratio above roughly 2.5×. Presence is bought with citations and responds to content work. |
Corpus, top cited domains and the displacement set
Top 15 cited domains, whole corpus
| # | Domain | Citations | Queries | Avg rank | Share | Type |
|---|---|---|---|---|---|---|
| 1 | nice.com | 823 | 199 | 3.7 | 3.73% | Owned |
| 2 | zendesk.com | 411 | 123 | 6.5 | 1.86% | Competitor |
| 3 | g2.com | 405 | 110 | 7.2 | 1.84% | Review aggregator |
| 4 | genesys.com | 336 | 91 | 2.2 | 1.52% | Competitor |
| 5 | nextiva.com | 335 | 114 | 6.1 | 1.52% | Adjacent vendor |
| 6 | gartner.com | 279 | 89 | 4.8 | 1.27% | Analyst |
| 7 | cloudtalk.io | 260 | 92 | 6.3 | 1.18% | Adjacent vendor |
| 8 | amplifai.com | 259 | 84 | 4.2 | 1.17% | Adjacent vendor |
| 9 | thelevel.ai | 250 | 89 | 5.7 | 1.13% | Adjacent vendor |
| 10 | ringcentral.com | 244 | 75 | 5.2 | 1.11% | Adjacent vendor |
| 11 | kore.ai | 239 | 57 | 5.1 | 1.08% | AI-native |
| 12 | crescendo.ai | 236 | 57 | 4.8 | 1.07% | AI-native |
| 13 | cresta.com | 222 | 71 | 5.2 | 1.01% | AI-native |
| 14 | verint.com | 220 | 68 | 4.2 | 1.00% | Competitor |
| 15 | talkdesk.com | 218 | 63 | 3.3 | 0.99% | Competitor |
Domains that gain specifically where NiCE is absent
Comparing citation share across 13,841 citations on NiCE-absent answers against 7,249 on NiCE-present answers isolates the ecosystem that displaces NiCE.
| Domain | Share, NiCE absent | Share, NiCE present | Skew | Category |
|---|---|---|---|---|
| fin.ai | 1.37% | 0.32% | +1.05 | AI-native support agent |
| rasa.com | 1.29% | 0.37% | +0.92 | Conversational AI framework |
| ibm.com | 1.01% | 0.25% | +0.76 | Enterprise AI platform |
| zendesk.com | 2.08% | 1.57% | +0.51 | Competitor |
| gumloop.com | 0.80% | 0.30% | +0.50 | Agent automation |
| youtube.com | 0.72% | 0.30% | +0.42 | Video |
| botpress.com | 0.77% | 0.37% | +0.40 | Bot framework |
| microsoft.com | 0.72% | 0.32% | +0.40 | Enterprise AI platform |
| kore.ai | 1.25% | 0.86% | +0.39 | AI-native |
| nice.com | 1.73% | 7.79% | −6.06 | Owned, the mechanism in one row |
Every positively skewed domain is an AI-native or platform-AI property. NiCE is not being displaced by traditional CCaaS competitors; it is being displaced by the content ecosystem of the agentic AI category, precisely the ecosystem it does not publish into.
Snapshot cross-check: the platform 30-day snapshot reports a 30% NiCE mention rate against 34.1% computed here. The difference is window scope, 30 days with 25 days of data depth versus the full 41-day export.