Edition05

AuthorFlemming Rubak

Published6 September 2026

Reading time9 minutes

The Sunday Shortlist

Danish customer service platforms: The tie that AI broke.

Measured 26 August to 6 September 2026. Monitoring sponsored by seedli.ai.

On this page

What we measured
The judging sheet
The verdict: the answer with one name
Two brands, one starting line
Nobody owns a criterion
The eliminator: reviews about data leaks
What claiming it looks like
Where the evidence must live
The lesson
Sources
The Sunday Shortlist

The Sunday Shortlist decodes how AI makes decisions about a market category and audience.

This week we analyse Danish customer service platforms

We unpack the questions Danish e-commerce and consumer brands ask AI when they choose a customer service platform, the shortlist that comes back, and the criteria that decide it, across all four stages of the decision journey: from first comparison to who customers stay with and recommend.

The four stages hold the week's sharpest finding: two brands entered the comparison in a statistical tie, and left the journey with opposite fates. Same starting line, same category, same buyers. The answers kept one of them.

Somewhere in Denmark right now, an e-commerce manager with a growing webshop and an overflowing inbox is asking an AI model which customer service platform to choose. The model answers in seconds: a shortlist, the risks, a favourite. No vendor hears about that conversation. We measure it.

What we measured

Since 26 August we have run this market's buying questions through the models across the decision journey: three measurement days over twelve days, measured with three models: Gemini, Claude, and ChatGPT. The late-journey stages (retention, advocacy) are sampled weekly, because loyalty moves slower than choice.

The questions are the ones buyers ask, in their own words:

"Hvilke kundeserviceplatforme kan håndtere flere kanaler på én gang?" (Which customer service platforms can handle several channels at once?)
"Kan systemet integreres med vores nuværende e-handelsløsning?" (Can the system integrate with our current e-commerce solution?)
"Hvad koster det at skifte fra Zendesk eller Freshdesk?" (What does it cost to switch from Zendesk or Freshdesk?)

Read that last one again. The category's buying questions have the incumbents built into them. In this market, the default is not just the answer. It is the premise of the question.

The judging sheet

When a buyer asks a model to choose, the model behaves like a judge: it applies criteria, checks each brand's evidence against them, and cuts the brands that fail. Ten criteria decide the answers in this category. Five weigh heaviest: cost and fees, expected outcomes, flexibility, product fit, and trust and reputation. The buyer language behind them is precise:

Cost: "Hvad er den samlede pris for licenser og implementering?" (What is the total price for licences and implementation?)
Outcomes: "Hvilken ROI kan vi forvente inden for det første år?" (What ROI can we expect within the first year?)
Flexibility: "Kan vi tilpasse workflows og automatiseringer til vores specifikke behov?" (Can we adapt workflows and automations to our specific needs?)
Trust: "Har de gode anmeldelser på uafhængige platforme?" (Do they have good reviews on independent platforms?)

This is the exam. Every vendor in the category sits it every day, whether it knows or not. And note where the trust question points: not at the vendor's website, at the review platforms. Hold that thought.

The verdict: the answer with one name

How to read the table: the percentages show how frequently each brand appears in the answers at each stage of the decision journey, measured across the full window, 26 August to 6 September. A dash means the sample for that stage is too small to report.

Brand	Evaluation	Decision	Retention	Advocacy
Zendesk	99%	100%	100%	100%
Freshdesk	69%	63%	100%	98%
Intercom	69%	42%	11%	26%
HubSpot	39%	29%	61%	77%
Gorgias	32%	12%	–	–
Freshsales	19%	7%	–	–
Microsoft	16%	19%	4%	2%
Salesforce	10%	3%	–	–

A note on the table: the first two columns are measured daily and rest on 372 and 338 answers in this window; the last two are measured weekly and rest on 46 and 43. Shares in the weekly columns move in coarser steps.

Start at the top. Zendesk appears in 100% of decision answers, 100% of retention answers, and 100% of advocacy answers, and misses three evaluation answers out of 372. In five decoded categories, no brand has come close to that: not the CRM giants, not the Danish IT default, not the architecture default. Zendesk is not winning the answer. Zendesk is the answer, so completely that the buyers' own questions treat it as the starting point.

Which reframes the entire category: nobody is fighting Zendesk for the answer. Everyone is fighting for the second name in it.

Two brands, one starting line

Now look at rows two and three, because this is the finding of the week.

Freshdesk and Intercom enter evaluation in a statistical tie: 69.1% against 68.8%, three mentions apart across 372 answers. Identical comparison presence. Then the journey happens to them:

At decision, Freshdesk holds 63%. Intercom carries 42%, the category's largest fall: one in three of its comparisons does not survive the choice.
At retention, the stage that answers "who do customers stay with", Freshdesk appears in every single answer: 100%. Intercom appears in 11%.
At advocacy, Freshdesk holds 98%. Intercom, 26%.

100% and 11%

Freshdesk's and Intercom's shares of the answers about who customers stay with. Same evaluation presence, three mentions apart. The journey did the separating.

Same starting line, opposite fates. This is as close as measurement gets to a controlled experiment: two brands with indistinguishable comparison presence, separated stage by stage as the evidence each stage draws on changes. Comparison answers reward breadth: being known, being listed, being adjacent. Loyalty answers draw on different material: who customers report staying with, what the reviews say about life after the purchase, which platform the case studies renew with. A brand can be everywhere in the comparison and nearly absent from the loyalty conversation, and the table above is what that looks like in numbers.

And HubSpot, which converted near the top of edition 1's US CRM decode, runs the inverse funnel here: 39% evaluation, 61% retention, 77% advocacy, the stage-specialist shape edition 3 measured in UK architecture. Every stage is its own market, judged on its own evidence.

Nobody owns a criterion

Ten criteria, zero owners. Five markets in a row now: US CRM platforms, Danish IT outsourcing, UK sustainable architecture, Danish spend management, Danish customer service platforms. Fifty measured criteria, and not one has a brand attached as its default answer, the 100% default included.

The standing conclusion holds: the first brand that publishes hard evidence against a single criterion takes it nearly uncontested. In this category the unclaimed questions are unusually concrete, and one family of them is asked with the incumbents' names inside: "Hvor let er det at migrere fra Zendesk eller Freshdesk til jeres platform?" (How easy is it to migrate from Zendesk or Freshdesk to your platform?) Amigration evidence page, honest about cost, time, and what breaks, answers the exact question buyers already ask and no vendor owns.

The eliminator: reviews about data leaks

The most severe elimination trigger in this category is trust, and the buyer phrase behind it is unusually specific:

"Leverandøren har dårlige anmeldelser om datalæk og manglende support."
(The provider has bad reviews about data leaks and missing support.)

Not "the provider had a data leak". Bad reviews about data leaks. In a category where the vendor holds your customers' data, the kill shot is not the incident, it is the incident's afterlife on review platforms. The adjacent eliminator is just as sharp: "Systemet overholder ikke GDPR, hvilket er kritisk for danske webshops." (The system does not comply with GDPR, which is critical for Danish webshops.) And the supporting risk verbatim names the missing artifact: "Der er ingen klar politik for datalagring og adgangskontrol." (There is no clear policy for data storage and access control.)

Here is what makes this week's version of the trust eliminator different from the earlier editions': it lives on surfaces the vendor does not own. And we can see the models reading exactly those surfaces, because this week's citations are the first in the series dominated by review platforms: Trustpilot, G2, Capterra, and Gartner, alongside the Danish business layer (Dansk Erhverv, Dansk Industri, IT-Branchen) and the consumer council Tænk. This is the terrainthe trust-stories playbook covers: named, dated, third-party-verified proof that survives on surfaces the vendor cannot edit.

The takeover play therefore has two parts, not one. First the page: a plain, dated data-security and GDPR policy, the exact document the measured risk says is missing. Then the echo, which for once is not optional: every reply to a review is an indexed sentence on the precise surface where this category's hardest cut is decided. A security page nobody echoes on the review platforms defuses nothing, because the fear does not live on your domain.

What claiming it looks like

This is the working brief for that page, generated from the monitoring data, with the brand genericised. The frame is "acknowledge the problem": the buyer's fear lives in reviews, so the page opens by validating that checking reviews about data leaks is exactly the right instinct, and only then presents the structure. Evidence before acknowledgement reads as damage control.

Title: "Dårlige anmeldelser om datalæk og manglende support: hvad siger fakta om [mit brand]?" (Bad reviews about data leaks and missing support: what do the facts say about [my brand]?)

The position the page takes: bekymringen om datalæk og manglende support er legitim at stille til enhver leverandør, men [mit brand] kan dokumentere vores sikkerhedscertificering, vores offentlige incident-historik og vores målbare supporttider. (The concern about data leaks and missing support is legitimate to raise with any vendor, but [my brand] can document our security certification, our public incident history, and our measurable support times.) Defensible because it rests on third-party certifications and publicly checkable records, not on the vendor's own marketing claims.

The intent family the same page must also answer (one prompt spawns two to three searches behind the scenes):

"Har [mit brand] haft datalæk, og er vores kundesupport pålidelig?" (Has [my brand] had data leaks, and is our support reliable?)
"Hvilke sikkerhedscertificeringer har [mit brand]?" (Which security certifications does [my brand] hold?)
"Hvad siger uafhængige anmeldelser om vores support på G2 og Trustpilot?" (What do independent reviews say about our support on G2 and Trustpilot?)
"Hvordan håndterer [mit brand] GDPR og databeskyttelse for danske virksomheder?" (How does [my brand] handle GDPR and data protection for Danish companies?)

The structure: five H2s, each a claim the models can lift as a standalone answer:

"Har [mit brand] haft datalæk? Her er den verificerbare historik" (Has [my brand] had data leaks? Here is the verifiable history)
"Hvilke sikkerhedscertificeringer dokumenterer vores databeskyttelse?" (Which security certifications document our data protection?)
"Hvad siger uafhængige anmeldelser faktisk om supporten, og hvordan læses de rigtigt?" (What do independent reviews actually say about the support, and how should they be read?)
"Sådan ser du forskel på dokumenteret og lovet datasikkerhed" (How to tell documented data security from promised data security)
"Hvad du bør kræve af enhver leverandør, inden du underskriver kontrakten" (What you should demand of any vendor before signing)

Key Takeaways for the top of the page (each a self-contained claim a model can cite):

"Anmeldelser om datalæk og dårlig support er den rigtige bekymring at rejse, inden du skifter kundeservicesystem." (Reviews about data leaks and poor support are the right concern to raise before switching systems.)
"Vores incident-historik er offentlig og verificerbar, ikke et marketingløfte." (Our incident history is public and verifiable, not a marketing promise.)
"Sikkerhedscertificering udstedes af en uafhængig tredjepart og kræver løbende revision." (Security certification is issued by an independent third party and requires ongoing audits.)
"Supporttider er målbare: [mit brand] offentliggør sine faktiske svartider frem for at afvise kritiske anmeldelser." (Support times are measurable: [my brand] publishes its actual response times rather than dismissing critical reviews.)
"GDPR-compliance er kontraktuel: databehandleraftale, datalagring i EU, klare vilkår for sletning." (GDPR compliance is contractual: DPA, EU data storage, clear deletion terms.)

Three notes. First, the reviews-H2 must engage with real negative reviews honestly; a page that waves them away as "written in frustration" breaks the acknowledge frame it opened with, and the takeover fails. Second, every number on this page is a template for a claim, not a licence to make it: certification IDs, incident logs, and response-time medians must come from the vendor's actual records and survive an audit before publication. Third, the eliminator couples two fears the evidence should not couple: data leaks and support reliability fan out as separate searches with separate official sources. Ideally this page is the fear-facing front, resting on two dedicated evidence pages behind it, one for security and incident history, one for support and response times, each with its own question-shaped slug. One page per claim; this page owns the reviews, the two behind it own the proof. The last H2 is the generous move again: the verification checklist every rival must also answer.

Key Snippet, placed early: "[Mit brand] har [antal] dokumenterede datalæk, er [certificering]-certificeret, og vi offentliggør vores faktiske supporttider." ([My brand] has [count] documented data leaks, holds [certification], and we publish our actual support response times.)

Slug: /har-[brand]-haft-datalaek — the buyer's question shape, not a content-type label.

Kept alive: the incident history and certification status re-verified against the issuing bodies, reviews from the last 12 months re-read quarterly, the refresh dated on the page.

Where the evidence must live: who listens where

The models name their sources, and this week, for the first time in five editions, the list is led by review platforms: Trustpilot, G2, Capterra, and Gartner, with the Danish business organisations (Dansk Erhverv, Dansk Industri, IT-Branchen) and Tænk behind them. Not one vendor blog.

That list is not random, and the research explains the mechanism behind each move you should make:

1. Your own page carries the facts, and it must parse

Network-traffic captures show the models going to the official page first for factual claims and giving up when content hides behind JavaScript; in one recorded reasoning trace, ChatGPT wanted a vendor's own numbers, could not parse the page, and cited a third-party source instead.²

For this category the facts under pressure are pricing ("skjulte gebyrer" appears in the eliminator data and the cost criterion alike) and the security policy the risk data says is missing. Both must exist as dated, plain-HTML pages: every fee named, every data-handling commitment written. The same captures show ChatGPT appending "official" to fact-seeking searches; for a policy question, your policy page is the official source it hunts for.³

2. The verdict layer is the review platforms, and they are editable

Models cite vendors for their own facts and third parties for the judgment.² This week the third party is not a profession or an institution: it is G2, Capterra, and Trustpilot, cited directly by the models. Two of those give vendors an editable surface today, the same move edition 1 mapped: the seller portal's product and pricing sections are yours to write, and every reply you post to a review is indexed text a model can read. In this category, review responses are not customer service. They are the defence layer on the surface where the hardest eliminator is decided.

3. Match the surface to the engine

The per-engine differences are measured.

ChatGPT and Claude lean on text surfaces: review platforms, trade press, LinkedIn articles under named authors. Video is close to citation-dead in ChatGPT, because search fetches a video's metadata, not its transcript.^2,5

Gemini and Perplexity read the spoken word, transcribed: a walkthrough of your security setup or a real migration, fresh and dated, is an asset there.⁴

One set of claims, placed per engine: the pages themselves for the facts, review replies and professional text for ChatGPT and Claude, a spoken version for Gemini and Perplexity.

4. The migration questions are standing unclaimed

The buyer language asks, by name, what it costs and takes to switch from the incumbents. Nobody in the category owns that answer as evidence: a dated migration page with real timelines, real costs, and what breaks in week one targets the exact fan-out questions the models already generate.¹ For a challenger in a category with a 100% default, the switching question is the most valuable open door there is.

5. Date everything

In an analysis of 250 million AI responses across eight answer engines, half of all top-cited content was under 13 weeks old.⁶ The recency bias is structural, and the researchers reading the engines' network traffic reached the same standing rule from the other direction: date every claim.³ Security and compliance claims age fastest of all: a GDPR page with a visible verification date beats a timeless one twice, once with the buyer, once with the judge.

The lesson

Every stage of the decision journey is its own market, with its own evidence: the stage-specialist pattern edition 3 measured in UK architecture, here at its sharpest, with two brands identical at the comparison stage living opposite lives after it. A tie at evaluation is not a position. It is a starting line, and what separates the kept from the evaporated is stage-specific evidence: choice evidence for the decision, loyalty evidence for the stages nobody watches.

Your category has its own version of this finding. The tie differs. What breaks it does not.

Sources

The network-traffic findings below come from one researcher's logged-in accounts and are labelled directional by the author; the mechanisms are reproducible, the percentages are not population measurements. Capture dates matter: the plumbing changes faster than the mechanisms.

The Sunday Shortlist

Get the decode, and the toolkit

The Sunday Shortlist decodes one category at a time: the questions buyers ask AI, the shortlist that comes back, and the criteria that decide it. Subscribers get every decode first, and the free toolkit gets you started tonight: a worksheet and three prompts to run the first check on your own brand.

One email a week · unsubscribe anytime · Terms

All editions