Skip to content
How we review
ai-agent.review — AI Agent Reviews, home

Changelog

Every change to a published fact, with the source behind it and the evidence states it moved. Each release is a patch against the records; replaying them all reproduces the records exactly, and the site cannot be published with a record that differs from this log.

Atom feedMarkdown recordHow the changelog works

releases
27
claims added
1,352
claims edited
122
claims removed
2
evidence states moved
453

October 2026

Release 26Evidence

Customer stories from YouTube for eleven agents

We added 34 claims from customer interviews on YouTube to Agentforce, Decagon, Fin, Forethought, Gorgias AI Agent, Kore.ai, LivePerson, Lyro, NiCE Cognigy, PolyAI and Zendesk AI Agents, each linked to the moment in the video it comes from. Thirty-three come from videos the vendor published and are labelled vendor-reported; Cloudbeds describing its use of Forethought, on Gainsight's channel, is the only independent report. They add named-customer results alongside the limits those customers describe: knowledge articles that had to be rewritten, daily review of answers, and wrong answers until guidance changed. No evidence state moves.

11 agents · 34 claims added

Read the release: Customer stories from YouTube for eleven agents

Release 25EvidenceRe-check

Zendesk AI Agents: channel limits, industry stories and new customer reports

We re-checked all 28 sourced claims on this record against Zendesk's help center, pricing page, product page, Babbel story and two community threads, and every one still matches its source. Zendesk's documentation adds WhatsApp as a channel and states several limitations: new AI agents serve one channel each, email formatting control is limited, social channels do not pass the customer's locale for automatic translation, generative procedures ignore knowledge search rules, and voice agents can occasionally produce distorted speech and are not meant for emergency healthcare triage. Voice AI agents became generally available on 1 September 2026. Vendor-published stories from NOBULL, Papier, Best Egg, Phonero and Serko, plus Zendesk's government page, place the agents in retail, financial services, telecommunications, travel and the public sector. New independent reports come from G2 reviews of the Zendesk suite, Trustpilot reviews of Zendesk the company, Zendesk Community threads and one tutorial on YouTube, and they cover both sides: answers to basic questions and a working setup on one side; on the other, a subscription conversation that went in circles, intelligent triage (a Copilot feature) misreading sarcasm, resolution billing, a WhatsApp routing failure and SLA gaps after escalation. Billing and product questions move to Caveat documented.

1 agent · 36 claims added · 28 claims re-checked · 14 evidence states moved

  • Zendesk AI Agents · BillingVendor-documented→Caveat documented
  • Zendesk AI Agents · Product questionsVendor-documented→Caveat documented
  • Zendesk AI Agents · Industry · Retail and e-commerceNot established→Vendor-documented
  • Zendesk AI Agents · Industry · Financial servicesNot established→Vendor-documented
  • Zendesk AI Agents · Industry · TelecommunicationsNot established→Vendor-documented
  • Zendesk AI Agents · Industry · Travel and hospitalityNot established→Vendor-documented
  • and 8 more

Read the release: Zendesk AI Agents: channel limits, industry stories and new customer reports

Release 24EvidenceRe-check

Yellow.ai: five more workflows backed, G2 and SoftwareReviews reports, Gartner placement and a security flaw in its website chatbot

Yellow.ai's documentation and case studies now back billing, orders, returns, account issues and complaints: Shopify order lookup, Razorpay refunds, utility bill payments, card blocking for banks and 311 complaint tickets for a city. Eighteen G2 reviews of the yellow.ai vendor listing are now cited, from both ends of the rating scale. Reviewers describe the agent answering loan and general questions, sending order updates to patients over WhatsApp and handing unanswered queries to human agents, which moves orders, product questions and human handoff to External report. Others report a stalled five-month implementation, falling service after onboarding, bot cost and a steep learning curve. SoftwareReviews adds an enterprise reviewer and a note that the platform is pricey for smaller businesses. CX Today reports Gartner's 2025 Challenger placement, with its strengths and cautions, and a cross-site scripting flaw found in the chatbot on Yellow.ai's own website, which the vendor says was a demo widget. An independent YouTube creator testing the free tier with placeholder inputs could not get past its knowledge-base setup. The knowledge base documentation now lists supported file types and a 50 MB limit. Existing sources were re-checked: the pricing page no longer shows a Basic tier, HubSpot is documented only for customising lead details, the 135-language figure now refers to voice agents, and on-premises deployment is documented elsewhere. Capterra blocked every request.

1 agent · 48 claims added · 5 claims edited · 9 claims re-checked · 20 evidence states moved

  • Yellow.ai · BillingNot established→Vendor-documented
  • Yellow.ai · OrdersNot established→External report
  • Yellow.ai · ReturnsNot established→Vendor-documented
  • Yellow.ai · Product questionsVendor-documented→External report
  • Yellow.ai · Account issuesNot established→Vendor-documented
  • Yellow.ai · ComplaintsNot established→Vendor-documented
  • and 14 more

Read the release: Yellow.ai: five more workflows backed, G2 and SoftwareReviews reports, Gartner placement and a security flaw in its website chatbot

Release 23EvidenceRe-check

Sprinklr: languages, order, billing and account tasks, Service suite help-desk integrations and re-checked sources

Sprinklr's help center states its conversational AI supports more than 100 languages, subject to each brand's contract, which fills the languages row. The same documentation and Sprinklr's virtual-agent page describe order status, returns, cancellations, account maintenance, password resets, authentication and payment tasks, and list WhatsApp and SMS among the channels, so billing, orders, returns, account issues, WhatsApp and SMS move to Vendor-documented. Sprinklr's own Aramex story and healthcare blog add logistics and healthcare as Vendor-documented, and a G2 review of the Sprinklr Service suite that calls it great for e-commerce customer service moves retail and e-commerce to External report. Sprinklr lists Zendesk, Salesforce, Freshdesk, Dynamics 365 and ServiceNow as integrations of its Sprinklr Service suite, not as places its AI agents run; Salesforce Group and Professional editions are not supported, so the Salesforce row reads Caveat documented. New weaknesses include Sprinklr's drop to Niche Player in Gartner's 2025 Magic Quadrant for Conversational AI Platforms, with AI Agent Studio only at the top of its pricing, G2 reviewers of Sprinklr Service describing complex configuration and high costs, its CEO's enterprise-first sales stance and the end of its self-serve products. A BT Group release on EE's assistant is filed as vendor-reported because it was issued with Sprinklr. On re-check, the CX Today article behind seven claims no longer loads, and an archived copy says the Umniah and Telefónica figures were published by Sprinklr itself, so those claims are flagged for re-typing.

1 agent · 30 claims added · 1 claim edited · 12 claims re-checked · 14 evidence states moved

  • Sprinklr · BillingNot established→Vendor-documented
  • Sprinklr · OrdersNot established→Vendor-documented
  • Sprinklr · ReturnsNot established→Vendor-documented
  • Sprinklr · Account issuesNot established→Vendor-documented
  • Sprinklr · Industry · Retail and e-commerceVendor-documented→External report
  • Sprinklr · Industry · HealthcareNot established→Vendor-documented
  • and 8 more

Read the release: Sprinklr: languages, order, billing and account tasks, Service suite help-desk integrations and re-checked sources

Release 22EvidenceRe-check

Talkdesk: channels, help desks, industries, documented limits and G2 reviews of the AI agent

Talkdesk's help center states that its virtual agent (now Autopilot) works on chat, SMS and WhatsApp, and that Facebook Messenger does not support it yet. That fills the chat, SMS and WhatsApp rows, and gives social messaging a sourced limitation beside Talkdesk's marketing claim. Autopilot's voice documentation names Salesforce and Zendesk as CRMs it looks data up in, which adds the first two help-desk rows. Vendor sources add healthcare, travel and hospitality, and government. An Opus Research note written after Talkdesk's own analyst summit is added as a strength; its healthcare and Epic remarks relay Talkdesk's strategy, so healthcare stays Vendor-documented. Talkdesk's documentation sets out the AI agent's limits: 256-character customer messages, no attachments on digital channels, 5-second backend calls, caps on agent hierarchies, extra usage costs, and telco charges for the whole call when the agent sits in front of another contact centre. Its voice language list is narrower than its digital one. From G2, a reviewer describes the website chatbot answering product and subscription questions, which moves product questions to External report. The same reviewer reports a 1,500-character instruction limit and no after-hours escalation, and other reviewers mention Autopilot on email and a complex AI billing model. G2's suite-level rating of 4.4 out of 5 is now shown as a strength. Small and enterprise businesses move to Vendor-documented, through Talkdesk Express and the Quadient story. Rechecks: most sources still say what we cite. CX Today, Vendr and Talkdesk's AI agents page moved to new addresses. The Vendr figure now rests on 67 purchases. The homepage wording on deployment and the customer figures on the Autopilot page changed, and both are flagged for review, as is the trade-press claim behind human handoff, whose source lists routing to in-store specialists but not escalation from the AI to a human.

1 agent · 27 claims added · 12 claims edited · 9 claims re-checked · 13 evidence states moved

  • Talkdesk · Product questionsNot established→External report
  • Talkdesk · Small businessesNot established→Vendor-documented
  • Talkdesk · EnterpriseNot established→Vendor-documented
  • Talkdesk · Industry · Travel and hospitalityNot established→Vendor-documented
  • Talkdesk · Industry · HealthcareNot established→Vendor-documented
  • Talkdesk · Industry · Public sector and educationNot established→Vendor-documented
  • and 7 more

Read the release: Talkdesk: channels, help desks, industries, documented limits and G2 reviews of the AI agent

Release 21EvidenceRe-check

Sierra: G2 reviews, industry pages, Gladly integration and a recheck of every vendor source

We re-read all 23 sourced claims. 21 still match their sources. Sierra's homepage no longer names AOL, so the two claims that cited it are flagged for review; their source is unchanged for now. Eighteen claims from thirteen G2 reviews were added, from both ends. Some reviewers describe the agent changing orders, return labels and damaged-order photos, others its use by a two-person tour operator and on WhatsApp and email. Others report high prices, setup that needs engineering skills, and answers that miss context on complex questions. Orders and returns move to External report, and so does small business on the basis of that one operator. Enterprise stays Vendor-documented: an analyst report's customer list repeats Sierra's own case studies, so it does not count as an external report. Vendor pages now support financial services, healthcare, telecommunications, travel, public sector (FedRAMP High) and logistics (RunBuggy). A Chubbies case study names a Gladly integration. Weaknesses add Salesforce Ben's point that outcome-based pricing depends on how an outcome is defined, plus a Reddit report that SiriusXM's Sierra-built Harmony agent cancelled a subscription without offering a human.

1 agent · 36 claims added · 5 claims edited · 16 claims re-checked · 13 evidence states moved

  • Sierra · OrdersVendor-documented→External report
  • Sierra · ReturnsVendor-documented→External report
  • Sierra · Small businessesNot established→External report
  • Sierra · Industry · Retail and e-commerceVendor-documented→External report
  • Sierra · Industry · Financial servicesNot established→Vendor-documented
  • Sierra · Industry · TelecommunicationsNot established→Vendor-documented
  • and 7 more

Read the release: Sierra: G2 reviews, industry pages, Gladly integration and a recheck of every vendor source

Release 20EvidenceRe-check

PolyAI: UK government records, Forbes reporting and a source recheck

Two UK government algorithmic-transparency records now document PolyAI in production. NS&I says PolyAI answers all of its inbound calls and contains about 22% of them, and the Ministry of Justice says it answers process questions about five prisons and passes calls to a human when needed. Both records support product questions, handoff to a human and the public-sector row. Forbes reported PolyAI handling card registration and management for UniCredit, front-desk queries and restaurant bookings for Las Vegas casinos, and power-outage calls for PG&E, which moves account issues to External report and adds to orders and enterprise. Product questions also move to External report, on these records and a G2 reviewer. PolyAI's documentation says agents process refunds, but the email channel always passes refunds, complaints and attachments to the team, so returns moves to Caveat documented. A complaint call at Bloomberg Tech was handled by a demo agent PolyAI built for the stage, so it is filed as vendor-documented. WhatsApp, HubSpot, Microsoft Dynamics 365 and ServiceNow are added as vendor-documented, along with a PG&E vendor metric and G2 reviews, including one critical point; the G2 aggregate now notes that four of the twelve reviews are incentivized. The existing sources were re-read: the languages page now says "75+ languages across 25+ countries, with native-quality speech", the Gartner rundown moved to a new address, and several cells are flagged for an editor.

1 agent · 31 claims added · 6 claims edited · 15 claims re-checked · 10 evidence states moved

  • PolyAI · ReturnsNot established→Caveat documented
  • PolyAI · Product questionsVendor-documented→External report
  • PolyAI · Account issuesNot established→External report
  • PolyAI · Industry · Financial servicesVendor-documented→External report
  • PolyAI · Channel · EmailVendor-documented→Caveat documented
  • PolyAI · Channel · Voice and phoneVendor-documented→External report
  • and 4 more

Read the release: PolyAI: UK government records, Forbes reporting and a source recheck

Release 19EvidenceRe-check

NiCE CXone: workflow, channel and industry documentation, pricing and deployment cautions

NiCE's help centre and product pages now document four workflows that had no evidence (billing payments, account updates, answers from a knowledge base and returns) and both small and mid-sized businesses, all as vendor documentation; order status gains vendor documentation too. The help centre also states that CXone bots work on SMS, WhatsApp and social channels, the AI voice agents page claims more than 100 languages through NiCE Cognigy, and NiCE names Salesforce as a handoff target and Zendesk as a knowledge source its AI can draw on. Vendor case studies from ECSI and Money Management International, an Openreach press release and a courts brochure add a financial-services row, which reads External report because CX Today's existing report on Citi's complaints automation also places CXone there, a vendor-documented telecommunications row and vendor documentation for the existing public-sector row, and NiCE's industry packages and retail page add vendor documentation to the existing healthcare and retail rows. CX Today reports that Gartner named advanced AI and analytical capabilities a NiCE strength. The weaknesses now record that CXone offers four separate virtual agent engines, with Amelia white-labelled inside Autopilot, that integration needs custom scripting and extra licences, that self-service is billed per session, and that CMSWire and NiCE's own CEO say deployments depend on the customer's data readiness. G2's suite-level rating was added. Re-checking the existing sources confirmed the vendor pages and found that CX Today has moved its articles; four claims built on NiCE's own earnings call are flagged for review of their source type.

1 agent · 25 claims added · 9 claims edited · 9 claims re-checked · 13 evidence states moved

  • NiCE CXone · BillingNot established→Vendor-documented
  • NiCE CXone · ReturnsNot established→Vendor-documented
  • NiCE CXone · Product questionsNot established→Vendor-documented
  • NiCE CXone · Account issuesNot established→Vendor-documented
  • NiCE CXone · Small businessesNot established→Vendor-documented
  • NiCE CXone · Mid-marketNot established→Vendor-documented
  • and 7 more

Read the release: NiCE CXone: workflow, channel and industry documentation, pricing and deployment cautions

Release 18EvidenceRe-check

NiCE Cognigy: workflows, industries and help desks from case studies, plus independent reviews

Cognigy's own case studies and documentation now back billing (Frontier Airlines refunds), orders and returns (Mister Spex), account changes (Humm Group) and complaint handling (Mobily), along with six industries, the email channel and the Freshdesk, Microsoft Dynamics 365 and ServiceNow help desks. A PeerSpot reviewer who builds a Swiss airline's chat assistant on Cognigy describes refund and rebooking flows, so billing and travel move to External report. CX Today's rundown of the 2025 Gartner Magic Quadrant adds independent reporting of Cognigy's large enterprise presence. Further reviews on G2, PeerSpot and Trustpilot report limited analytics and chat flows, scarce documentation, a skills barrier beyond the basics and chat bots that do not work. The documentation now says on-premises installations are no longer offered to new customers and lists default data retention limits. In the re-check, the endpoint count rose to 30, real-time queue notifications now also cover Salesforce MIAW and Genesys, and CX Today's articles moved to new addresses.

1 agent · 32 claims added · 6 claims edited · 15 claims re-checked · 17 evidence states moved

  • NiCE Cognigy · BillingNot established→External report
  • NiCE Cognigy · OrdersNot established→Vendor-documented
  • NiCE Cognigy · ReturnsNot established→Vendor-documented
  • NiCE Cognigy · Account issuesNot established→Vendor-documented
  • NiCE Cognigy · ComplaintsNot established→Vendor-documented
  • NiCE Cognigy · Industry · Retail and e-commerceNot established→Vendor-documented
  • and 11 more

Read the release: NiCE Cognigy: workflows, industries and help desks from case studies, plus independent reviews

Release 17EvidenceRe-check

LivePerson: language limits, case studies, G2 and Trustpilot reviews, and a dead CX Today source

LivePerson's developer documentation now fills the languages row: Conversation Builder lists 39 language options for messaging bots. The same documentation states that its LLM-based Routing AI agents are supported only in American or British English, and that generatively enriched KnowledgeAI answers are supported only for English or Spanish, with other languages experimental. Both limits are recorded as weaknesses, along with documented hallucination liability for consumer-facing bots and an auto-escalation dialog that never triggers when a KnowledgeAI search handles fallback. Vendor case studies and documentation now cover billing (payments inside messaging conversations), orders (an Order Status bot template), returns, account data, and the email, web chat and social channels, along with healthcare, travel, telecoms and payments deployments and LivePerson's retail page. Reviews of LivePerson on G2's vendor-level listing were added: one describes AI bots answering common customers' questions, and one describes bots handling simple questions with humans stepping in only when needed, so product questions and human handoff move to External report, while billing, orders, returns and account issues move to Vendor-documented. Other G2 reviews note complex configuration, the need for dedicated specialists, high cost for small companies, reliability problems and AI features rolling out slowly. Two 2020 Trustpilot reviews of LivePerson the company report bots that stall and need babysitting, and bot licences charged monthly. LivePerson's Q4 2025 results attribute a 19% revenue fall to customer cancellations and downsells. Recheck: the CX Today article behind four claims now returns 404, the SoundHound deal article moved to a new address, and the KnowledgeAI overview no longer describes internal and external knowledge bases.

1 agent · 43 claims added · 4 claims edited · 12 claims re-checked · 16 evidence states moved

  • LivePerson · BillingNot established→Vendor-documented
  • LivePerson · OrdersNot established→Vendor-documented
  • LivePerson · ReturnsNot established→Vendor-documented
  • LivePerson · Product questionsVendor-documented→External report
  • LivePerson · Account issuesNot established→Vendor-documented
  • LivePerson · Human handoffVendor-documented→External report
  • and 10 more

Read the release: LivePerson: language limits, case studies, G2 and Trustpilot reviews, and a dead CX Today source

Release 16EvidenceRe-check

Kore.ai: workflows, integrations, languages and new user reports

Kore.ai's own pages and documentation now back billing, orders, returns, account issues and complaints: a healthcare page on billing and payments, a named retailer (Autodoc) whose AI agents handle order tracking and refund checks, retail returns handling, banking account servicing, and complaints listed as a reason to escalate to a person. The documentation states support for more than 100 languages, which fills the languages row, and names live-agent handoff to Salesforce, ServiceNow, Genesys, Five9, NICE inContact and Intercom, plus connectors for Freshdesk, HubSpot, Dynamics 365, Front and Help Scout. A PeerSpot reviewer who builds Kore.ai customer-service bots in-house reports a 50 to 60 percent drop in calls to agents and live-agent handoff in use. G2 reviewers of Kore.AI, a listing that covers the whole company, report both grounded answers and occasional hallucinations, occasional latency in voice interactions, and limited templates for WhatsApp and other Meta channels. Product questions move to Caveat documented. On re-check, Kore.ai's customer-outcome figures have moved to another page and one of them is gone, and CX Today has moved two articles; four items are flagged for a person to review.

1 agent · 37 claims added · 3 claims edited · 13 claims re-checked · 21 evidence states moved

  • Kore.ai · BillingNot established→Vendor-documented
  • Kore.ai · OrdersNot established→Vendor-documented
  • Kore.ai · ReturnsNot established→Vendor-documented
  • Kore.ai · Product questionsVendor-documented→Caveat documented
  • Kore.ai · Account issuesNot established→Vendor-documented
  • Kore.ai · ComplaintsNot established→Vendor-documented
  • and 15 more

Read the release: Kore.ai: workflows, integrations, languages and new user reports

Release 15EvidenceRe-check

Breeze Customer Agent: workflow documentation, voice limits, community and G2 reports

HubSpot's documentation now gives Customer Agent examples for order status, returns-policy questions, password resets and, since its August 2026 release notes, looking up invoices, so billing, orders, returns, account issues and enterprise move from Not established to Vendor-documented, and HubSpot's Shopify actions and ISSA case study give retail and e-commerce and public sector and education their first vendor documentation. HubSpot documents its voice channel as a public beta that answers inbound calls only, which moves voice to Caveat documented. A HubSpot Community user reported in June 2025 that the agent offered a transfer button before trying to answer, and a G2 reviewer of HubSpot Agent Hub reported a website agent making up answers outside its knowledge base until boundaries were added, so human handoff and product questions move to Caveat documented. A healthcare user's February 2025 report that they could not take the agent live for accuracy reasons is the first evidence on healthcare, which therefore reads Caveat documented. New weaknesses cover CRM access limited to ten contact properties, weekly re-sync of non-HubSpot knowledge, required onboarding fees and credit auto-upgrades. All existing sources were re-checked: CX Today and Vendr moved to new addresses, and two comparison cells were reworded to match what HubSpot's pages now say.

1 agent · 43 claims added · 6 claims edited · 16 claims re-checked · 12 evidence states moved

  • Breeze Customer Agent · BillingNot established→Vendor-documented
  • Breeze Customer Agent · OrdersNot established→Vendor-documented
  • Breeze Customer Agent · ReturnsNot established→Vendor-documented
  • Breeze Customer Agent · Product questionsVendor-documented→Caveat documented
  • Breeze Customer Agent · Account issuesNot established→Vendor-documented
  • Breeze Customer Agent · Human handoffVendor-documented→Caveat documented
  • and 6 more

Read the release: Breeze Customer Agent: workflow documentation, voice limits, community and G2 reports

Release 14EvidenceRe-check

Genesys Cloud CX: token rates, channels, help desks and first user reports

Genesys documentation now fills the deployment and help-desk rows: Virtual Agent runs inside voice and digital bot flows, and data actions connect to Salesforce, Zendesk and Microsoft Dynamics 365. ServiceNow is added from Genesys's announced agent-to-agent integration with ServiceNow's AI agents. Digital bot flows are documented for email, and Genesys states SMS can be automated with virtual agents. Billing and account issues move to Vendor-documented: an announced ServiceNow scenario explains bill spikes, and AI Guides can look up account number and status in a CRM. An independent YouTube tutorial, built in the creator's own test lab, shows the same bot working by phone and in web messaging and then transferring the chat to a live agent, so human handoff moves to External report. Genesys's own product manager acknowledged gaps in how voice bots handle background noise, so voice moves to Caveat documented. Weaknesses add the per-interaction token rates, knowledge-answer search limited to digital channels and digital bot flows not yet PCI DSS-compliant, plus user reports calling Virtual Agent pricing steep and its conversational AI in need of improvement. G2 reviews of the Genesys Cloud CX suite (4.4 out of 5 across 1558 reviews) and one Trustpilot review were added. Vendr's guide and one Genesys AI page moved to new addresses; the other sources were re-checked and still support their claims.

1 agent · 29 claims added · 5 claims edited · 13 claims re-checked · 14 evidence states moved

  • Genesys Cloud CX · BillingNot established→Vendor-documented
  • Genesys Cloud CX · Account issuesNot established→Vendor-documented
  • Genesys Cloud CX · Human handoffVendor-documented→External report
  • Genesys Cloud CX · Industry · Retail and e-commerceNot established→Vendor-documented
  • Genesys Cloud CX · Industry · Financial servicesNot established→External report
  • Genesys Cloud CX · Industry · HealthcareNot established→External report
  • and 8 more

Read the release: Genesys Cloud CX: token rates, channels, help desks and first user reports

Release 13EvidenceRe-check

Freddy AI Agent: channels, workflows, session limits and first marketplace reviews

Freshworks' support documentation now names the channels the Chat AI Agent answers on (web widget, WhatsApp, Instagram, Messenger, LINE and SMS) and states 60+ languages for its AI agents, which fills the languages row; it also lists 24 languages for the Email AI Agent and documents handover settings, which fill the handoff row. The documentation describes workflows for billing, order tracking, exchanges, login problems and complaints, plus a handoff block, so those workflows move to Vendor-documented. It also states several limits: the 500 free sessions are a one-time allowance, unused sessions expire, chat agents stop taking new queries when sessions run out, and an agent deployed in the self-service widget is then not handed to a live agent. That last limit moves human handoff to Caveat documented, and Vendr's purchase data moves enterprise to External report. Standalone Freshdesk accounts get only the Email AI Agent. Vertical agents for e-commerce, travel, fintech and logistics are documented. New independent reports: Freshchat reviews on AWS Marketplace describe the chatbot answering product questions and passing customers to live agents, and raise AI-unit limits and the need for training. Two G2 reviews of Freshdesk calling its AI features limited, a Trustpilot review saying Freshworks' own Freddy support bot makes things up, a YouTube tutorial showing Freddy's channel bot templates, and the G2 aggregate for Freshdesk were also added. Existing sources were re-checked: Vendr's guide moved and its $15,411 figure is an average rather than a median, the 96% satisfaction figure on Freshworks' AI agent page is not attributed to Hobbycraft, and Freshworks' community roundup does not call the self-evaluation feature a public beta.

1 agent · 39 claims added · 6 claims edited · 13 claims re-checked · 14 evidence states moved

  • Freddy AI Agent · BillingNot established→Vendor-documented
  • Freddy AI Agent · OrdersNot established→Vendor-documented
  • Freddy AI Agent · ReturnsNot established→Vendor-documented
  • Freddy AI Agent · Account issuesNot established→Vendor-documented
  • Freddy AI Agent · ComplaintsNot established→Vendor-documented
  • Freddy AI Agent · Human handoffNot established→Caveat documented
  • and 8 more

Read the release: Freddy AI Agent: channels, workflows, session limits and first marketplace reviews

Release 12EvidenceRe-check

Forethought: G2 and Trustpilot reports, Zendesk availability, industries and a fresh recheck

Every sourced Forethought claim was re-checked: the Solve page moved to /platform/solve, and the pricing page no longer carries the 'Enterprise AI Agents for Any CX Platform' line, which now sits on the homepage. Zendesk's help centre states that Forethought has been on sale since 4 June 2026, as an add-on for Zendesk customers and on its own for everyone else, and lists Assist, Triage, Discover and QA as add-ons. Vendor pages now document returns, billing (payment disputes and invoice lookups), product questions and all three company sizes, plus retail, fintech, healthcare and education. The vendor's FAQ also says the AI needs at least 2,000 email or chat tickets a month. G2 reviewers describe answers built from the knowledge base and handoff only when a person is needed, and also report a steep learning curve, slow onboarding, limited reporting and high prices. G2's average of 4.3 out of 5 across 166 reviews is added. On Trustpilot, two people using chatbots built on Forethought report a missing-invoice request that went in circles without reaching a person and an account problem where the bot repeated itself and blocked live support; a third reviewer, who trains the bot, finds its answers too brief. Account issues and human handoff move to Caveat documented; product questions moves to External report; billing, returns and the three company sizes move to Vendor-documented.

1 agent · 43 claims added · 9 claims edited · 13 claims re-checked · 17 evidence states moved

  • Forethought · BillingNot established→Vendor-documented
  • Forethought · ReturnsNot established→Vendor-documented
  • Forethought · Product questionsNot established→External report
  • Forethought · Account issuesExternal report→Caveat documented
  • Forethought · Human handoffExternal report→Caveat documented
  • Forethought · Small businessesNot established→Vendor-documented
  • and 11 more

Read the release: Forethought: G2 and Trustpilot reports, Zendesk availability, industries and a fresh recheck

Release 11EvidenceRe-check

Five9: payments limitation, languages, case-study results and help-desk links

Five9's AI Agents documentation states that AI Agents is currently not PCI-compliant and that card payments must be handled outside it, for example by a PCI-compliant IVR or its older Studio IVA, so billing moves to Caveat documented. The same documentation fills the languages row: an agent can answer in several selected languages and switch mid-call. Five9's Wyndham case study backs account issues (its AI agents handle about 40,000 password resets a month), and Five9's own case studies and statements place the agent in financial services, travel and hospitality, telecommunications, healthcare and logistics; all of these are the vendor's own reports. Vendor sources add the WhatsApp channel and Salesforce and ServiceNow integrations. CX Today's coverage of Gartner's 2025 contact-centre report adds an external report of a high take-up of AI agents among Five9's enterprise customers. The record now carries G2's 4.1 out of 5 rating for the whole Five9 suite, a G2 reviewer and a Trustpilot reviewer on Five9's older Intelligent Virtual Agent, a G2 complaint about professional-services costs, and documented editing limitations. Existing sources were re-checked; several moved to new addresses and two quotes were corrected to match their sources word for word.

1 agent · 25 claims added · 9 claims edited · 10 claims re-checked · 10 evidence states moved

  • Five9 · BillingNot established→Caveat documented
  • Five9 · Account issuesNot established→Vendor-documented
  • Five9 · Industry · Financial servicesNot established→Vendor-documented
  • Five9 · Industry · TelecommunicationsNot established→Vendor-documented
  • Five9 · Industry · Travel and hospitalityNot established→Vendor-documented
  • Five9 · Industry · HealthcareNot established→Vendor-documented
  • and 4 more

Read the release: Five9: payments limitation, languages, case-study results and help-desk links

Release 10EvidenceRe-check

Fin: G2 and community reports, Fin Voice, help-desk coverage and a re-check of every source

All 28 sourced claims on the Fin record were re-read on 2026-10-04 and still say what we quoted. New evidence comes from G2 reviews, the Intercom Community forum, a Trustpilot review of Intercom (Trustpilot lists the company, not Fin alone) and the Shopify App Store listing for Fin AI Agent & Intercom, alongside new Intercom and Fin documentation. Small businesses move to External report: one reviewer writes that they are a small team and that Fin takes a big part of the workload off their shoulders. Product questions and human handoff move to Caveat documented: one reviewer reports Fin looping when a customer asks which volume-based plan suits them, and another says that by the time Fin hands a conversation to the human queue there can be a lot of chat history to read. Orders stays Vendor-documented: a community post about a tracking question does not say the data concerned orders, and a YouTube demo was not used because it is a pre-sales pitch built on test data. Weaknesses now include reviewer complaints about cost and billed resolutions, community posts saying Fin keeps replying until an agent answers, Intercom's statement that Fin Voice is limited to select customers working with its Sales team, and the help-center note that every help desk other than Intercom needs sales-assisted onboarding. Voice, financial services and healthcare rows were added, as were the Front and Gorgias help-desk rows.

1 agent · 41 claims added · 28 claims re-checked · 13 evidence states moved

  • Fin · Product questionsVendor-documented→Caveat documented
  • Fin · Human handoffVendor-documented→Caveat documented
  • Fin · Small businessesVendor-documented→External report
  • Fin · Industry · Financial servicesNot established→Vendor-documented
  • Fin · Industry · HealthcareNot established→Vendor-documented
  • Fin · Channel · Web and in-app chatVendor-documented→External report
  • and 7 more

Read the release: Fin: G2 and community reports, Fin Voice, help-desk coverage and a re-check of every source

Release 9EvidenceRe-check

Decagon: G2 and Reddit reports, industry pages, WhatsApp and Front, and a corrected reviewer name

Every sourced Decagon claim was re-read on 4 October 2026. All but one still match their sources: the review we credited to a reviewer named LOKESH G. does not appear on the AWS Marketplace page that syndicates Decagon’s G2 reviews, and the claim now cites CA Rahul B., whose review of 20 August 2026 says the setup is quite complex. From G2, a reviewer says the team uses Agent Operating Procedures to update refunds, and an enterprise reviewer reports more than 4,000 customers a day using the agent; G2 rates Decagon 4.7 out of 5 across 31 reviews. Other reviews describe answers that miss the question, mistakes when policies are wrong, setup that takes longer than expected, a need for a dedicated owner and requests for smoother handoffs. A Reddit user who runs Decagon on top of Zendesk reports about 20,000 conversations a month with roughly 55% deflected, and says Decagon has no human agent tool for escalations as far as they know. A Backbone customer on Reddit says the AI stopped replying during a warranty claim; the post names Decagon, but it does not show whether a handover failed, so it is listed as a weakness rather than a handoff failure. TechCrunch reports more than 100 large customers. Billing, product questions and enterprise move to External report. Decagon’s own pages add telecommunications, travel, health and financial-services offerings, WhatsApp on chat, Decagon Assist inside Front, and a private-offer AWS Marketplace contract.

1 agent · 40 claims added · 1 claim edited · 25 claims re-checked · 13 evidence states moved

  • Decagon · BillingVendor-documented→External report
  • Decagon · Product questionsVendor-documented→External report
  • Decagon · EnterpriseVendor-documented→External report
  • Decagon · Industry · Financial servicesNot established→Vendor-documented
  • Decagon · Industry · TelecommunicationsNot established→Vendor-documented
  • Decagon · Industry · Travel and hospitalityNot established→Vendor-documented
  • and 7 more

Read the release: Decagon: G2 and Reddit reports, industry pages, WhatsApp and Front, and a corrected reviewer name

Release 8EvidenceRe-check

Crescendo: languages, billing and account workflows, Gartner Peer Insights reviews and re-checked sources

Crescendo's Shopify App Store listing states 24/7 availability in more than 50 languages, which fills the languages row. Customer stories on Crescendo's site now document billing questions (SimpleSUB Water), password resets and account questions (RealVNC), sensitive-case escalation and a six-person team (Kiki Milk), a healthcare deployment (NIMA) and returns handled as exchanges, so billing, returns, account issues, complaints and small businesses move to Vendor-documented. Crescendo has 2 ratings on Gartner Peer Insights (4.5 out of 5); both reviews were added: one praises smooth escalation to human agents but reports limited customization, the other praises instant answers to routine questions but reports scaling challenges. Further additions are ServiceNow as a handoff target, the vendor's own statements on extra oversight for custom back-end integrations and on Shopify and HubSpot integration limits, a vendor-entered Enterprise Connect 2026 product award and an analyst's note on staffing risk. Every previously cited source was re-checked: one CX Today article moved to a new address, the pricing page now redirects to the homepage, the outcome-pricing and enterprise statements are re-quoted from where they now appear, and the Sacra profile no longer shows the customer and revenue figures we quoted. Two CX Today articles quoting Alorica, which resells Crescendo, are left for an editor to review.

1 agent · 23 claims added · 6 claims edited · 13 claims re-checked · 9 evidence states moved

  • Crescendo · BillingNot established→Vendor-documented
  • Crescendo · ReturnsNot established→Vendor-documented
  • Crescendo · Account issuesNot established→Vendor-documented
  • Crescendo · ComplaintsNot established→Vendor-documented
  • Crescendo · Small businessesNot established→Vendor-documented
  • Crescendo · Industry · Retail and e-commerceNot established→Vendor-documented
  • and 3 more

Read the release: Crescendo: languages, billing and account workflows, Gartner Peer Insights reviews and re-checked sources

Release 7EvidenceRe-check

Agentforce: channel limits, industry deployments, Help-site complaints and a re-check of every source

Salesforce's Service Agent documentation now names email, SMS, WhatsApp, Facebook Messenger, Apple Messages and LINE as channels. The same pages state that the agent handles only conversations the customer starts, that a message longer than a channel's character limit gets no reply, that some standard actions such as updating record fields are unsupported, and that Trust Layer data masking is switched off. Vendor case material adds deployments in public sector (TSA), travel (Engine), retail (Falabella), banking (Agibank), telecoms (One NZ) and a healthcare feature announcement, along with the June 2026 prepackaged Help Agent with pay-per-resolution pricing. On the independent side, Diginomica's reports on Wiley and Secret Escapes describe account-access fixes, handoff to staff and useful knowledge-base answers. CX Today reports Salesforce Help users saying the agent is slow and rarely gives the specific answer first time, and Gizmodo, citing Bloomberg, reports advertised UChicago Medicine and Williams-Sonoma features that were not yet live, with anonymous sources blaming the UChicago delay partly on product glitches. G2 reviews of the Service Cloud suite add praise for handoff with a summary and complaints about token pricing, consumption tracking, latency and setup effort. A Trustpilot reviewer describes being sent to Agentforce and then disconnected. Account issues and human handoff move to External report; product questions moves to Caveat documented. All 24 previously sourced claims were re-checked: 22 still say what we quote, two of them now carrying the date their source was published, and the Data Library article has moved to a new address.

1 agent · 41 claims added · 4 claims edited · 20 claims re-checked · 13 evidence states moved

  • Agentforce · Product questionsVendor-documented→Caveat documented
  • Agentforce · Account issuesVendor-documented→External report
  • Agentforce · Human handoffVendor-documented→External report
  • Agentforce · Industry · Retail and e-commerceNot established→Vendor-documented
  • Agentforce · Industry · Financial servicesNot established→Vendor-documented
  • Agentforce · Industry · TelecommunicationsNot established→Vendor-documented
  • and 7 more

Read the release: Agentforce: channel limits, industry deployments, Help-site complaints and a re-check of every source

Release 6EvidenceRe-check

Ada: reviewer reports, help-desk handoff limits and new industries

Twenty-two G2 reviews of Ada and G2’s aggregate rating are now cited, along with The Logic’s report on Trust & Will and Vendr’s pricing data. Orders gains a reviewer’s account of Ada offering to cancel orders it could not cancel, and moves to Caveat documented. Product questions, account issues and human handoff move to External report. Mid-market companies gain their first report: Trust & Will, whose support staff are just over a tenth of its employees, says Ada’s agent helps its team manage the workload. Reviewers describe workarounds needed for the Zendesk integration and, in 2023, images that could not be passed to Kustomer. Ada’s documentation now backs WhatsApp and SMS, handoffs to Gorgias and Gladly, and email integrations with Kustomer and Microsoft Dynamics. It also records the limits Ada states for the Gorgias, Gladly and Help Scout handoffs, for email (no List or Capture blocks; not HIPAA eligible), for social channels and for voice languages. Financial services, travel and health insurance are added from Ada’s industry pages. Twenty-four existing claims were re-checked against their sources. Ada’s demo page no longer addresses enterprise teams directly, so that claim now quotes what it does say.

1 agent · 50 claims added · 1 claim edited · 23 claims re-checked · 21 evidence states moved

  • Ada · OrdersVendor-documented→Caveat documented
  • Ada · Product questionsVendor-documented→External report
  • Ada · Account issuesVendor-documented→External report
  • Ada · Human handoffVendor-documented→External report
  • Ada · Mid-marketNot established→External report
  • Ada · Industry · Financial servicesNot established→Vendor-documented
  • and 15 more

Read the release: Ada: reviewer reports, help-desk handoff limits and new industries

Release 5EvidenceRe-check

Lyro: Premium-only resolution guarantee, help-desk integrations and both sides of merchant feedback

Tidio's pricing page lists a guaranteed 50% Lyro AI resolution rate as a feature of its Premium plan only, priced on request and starting from 3,000 Lyro conversations; the Free, Starter, Growth and Plus plans do not include it. We re-checked every vendor and Shopify source in the record: Tidio's Lyro page moved from /lyro/ to /ai-agent/ with the same content, the Shopify app's rating now covers 1,349 reviews, and one merchant review moved page. Tidio documents Actions for subscription and payment changes and for account updates and password resets, and escalation of angry customers to a person. Its standalone Lyro site, getlyro.ai, documents an Enterprise plan with guaranteed SLA and resolution rate, a phone channel through ElevenLabs and integrations with HubSpot, Freshdesk, Gorgias, Help Scout and ServiceNow. Billing, account issues, complaints, mid-market and enterprise move to Vendor-documented. So do financial services, on Tidio's Borrowell case study, the voice channel, and the ServiceNow, Freshdesk, HubSpot, Gorgias and Help Scout help-desk rows. The chat channel moves to External report, on a merchant review saying Lyro answers its website live chats. Returns move to External report, on a merchant review about answering returns questions. Human handoff moves to Caveat documented, after a Trustpilot reviewer of Tidio reported Lyro failing to transfer conversations or create tickets. New G2, Trustpilot and Shopify reviews cover both praise and complaints: wrong answers, cost, quotas and rigid automation. The record is fact-checked by Gosia Madej.

1 agent · 46 claims added · 13 claims edited · 16 claims re-checked · 15 evidence states moved

  • Lyro · BillingNot established→Vendor-documented
  • Lyro · ReturnsVendor-documented→External report
  • Lyro · Account issuesNot established→Vendor-documented
  • Lyro · ComplaintsNot established→Vendor-documented
  • Lyro · Human handoffExternal report→Caveat documented
  • Lyro · Mid-marketNot established→Vendor-documented
  • and 9 more

Read the release: Lyro: Premium-only resolution guarantee, help-desk integrations and both sides of merchant feedback

Release 4EvidenceRe-check

Gorgias AI Agent: billing, complaints and company-size evidence, Trustpilot reviews, and a full re-check

Gorgias's own documentation now establishes four more rows on the profile: subscription billing through Recharge, complaints about damaged or missing items read from customer photos, and two case studies placing the agent at a mid-sized brand and at a brand handling about 15,000 tickets a month. Each is vendor-documented, not independently confirmed. Every sourced claim was re-checked against its page: the returns, actions, channels, languages and benchmark claims were rewritten to say only what their sources say today, the review count and two review quotes were corrected, and a claim that Gorgias publishes no price for the agent was removed because it is no longer true. Three documented limits were added: the agent works only with Shopify stores, and its WhatsApp, Instagram and Messenger channels are in open beta. No source found establishes a voice channel or account-issue handling, so both stay unknown. Gorgias's Trustpilot reviews were read in full for the first time, through the correct listing: an independent report of the agent acting on a removed order-cancellation action for about 12 hours moves orders to a documented caveat, two more critical reviews of the AI agent and the 2.1 out of 5 company rating join the weaknesses, and both earlier Trustpilot quotes were confirmed word for word.

1 agent · 12 claims added · 10 claims edited · 1 claim removed · 13 claims re-checked · 5 evidence states moved

  • Gorgias AI Agent · BillingNot established→Vendor-documented
  • Gorgias AI Agent · OrdersExternal report→Caveat documented
  • Gorgias AI Agent · ComplaintsNot established→Vendor-documented
  • Gorgias AI Agent · Mid-marketNot established→Vendor-documented
  • Gorgias AI Agent · EnterpriseNot established→Vendor-documented

Read the release: Gorgias AI Agent: billing, complaints and company-size evidence, Trustpilot reviews, and a full re-check

Release 3Evidence

Lyro: languages, channels, refund procedures and new merchant reports

Tidio's help center now documents 48 languages for Lyro, which fills the languages row and retires our note that no language list could be found. The same documentation names email, WhatsApp, Messenger and Instagram as channels Lyro can answer on, gives refund requests as an example of a Procedure, and states that on email Lyro answers only the first message in a thread. Five merchant reviews on the Shopify App Store were added: two describe Lyro answering order and delivery-status questions, one describes it answering email and social messages, one mentions multiple languages, and one from 2024 reports answers that matched none of the content it was given. Orders, email and social messaging move to External report. The record is now fact-checked by a named person.

1 agent · 8 claims added · 1 claim edited · 1 claim removed · 5 evidence states moved

  • Lyro · OrdersVendor-documented→External report
  • Lyro · ReturnsNot established→Vendor-documented
  • Lyro · Channel · EmailNot established→External report
  • Lyro · Channel · WhatsAppNot established→Vendor-documented
  • Lyro · Channel · Social messagingNot established→External report

Read the release: Lyro: languages, channels, refund procedures and new merchant reports

Release 2Structure

Channels and help desks become evidence rows, and records gain bylines

Every profile now says, row by row, which customer channels and which help desks the cited sources document each AI agent working with, in the same four evidence states as workflows and industries. The rows cite claims the records already carried: no claim was added, edited or removed. Records also gained the fields that name who researched them and who fact-checked them; no byline is published until named people hold those roles. Notes the profiles carried before launch now live in this changelog.

22 agents · 141 evidence states moved

  • Ada · Channel · Web and in-app chatNot established→Vendor-documented
  • Ada · Channel · EmailNot established→Vendor-documented
  • Ada · Channel · Voice and phoneNot established→Vendor-documented
  • Ada · Channel · Social messagingNot established→Vendor-documented
  • Ada · Help desk · ZendeskNot established→Vendor-documented
  • Ada · Help desk · SalesforceNot established→Vendor-documented
  • and 135 more

Read the release: Channels and help desks become evidence rows, and records gain bylines

Release 1Catalog

Launch: 22 AI customer service agents and 532 claims, every one sourced and dated

ai-agent.review went live with the 22 agents the published inclusion rule admits and 532 claims about them. Every claim names its source, labels its source type and carries the date it was checked. No agent has been tested hands-on, and no record carries a score. Notes the profiles recorded during research before launch are kept with each agent below.

22 agents · 532 claims added

Read the release: Launch: 22 AI customer service agents and 532 claims, every one sourced and dated