What does it tell you when people accept an AI system’s recommendations more than 90% of the time? Amazon.com, Inc. (NASDAQ: AMZN) reported on September 23, 2026 that sellers accept Seller Assistant’s recommendations more than 90% of the time, in the same announcement that added always-on workflows and a plugin carrying Seller Assistant into Amazon Quick and, in beta, Anthropic’s Claude. Six other marketing technology and AI announcements carried September 23, 2026 dates. Phave, the AI-native marketing automation platform founded by Marketo co-founder Jon Miller, came out of stealth. Citigroup Inc. (NYSE: C) launched Citi Commerce Media. Akeneo published consumer pricing survey results, Peec AI added visibility tracking for the model behind Meta’s Muse agent, Liferay made its Content Marketing Platform generally available, and Mole Street released financial services research produced with HubSpot.
An acceptance rate counts agreement. Amazon’s figure records how often sellers approved what Seller Assistant proposed, and Amazon published no data on how the approved recommendations performed afterward. Amazon stated on September 23, 2026 that Seller Assistant has reached more than 90% of its selling partners worldwide, has hundreds of thousands of active users, and now keeps persistent memory of each seller’s pricing patterns, inventory cycles and growth goals. The new workflows run continuously against conditions a seller writes in plain language. Each seller chooses whether a workflow only surfaces recommendations or takes actions, and approves those actions before they run. Amazon stated that every action carries an audit trail.
Amazon’s approval step is human-in-the-loop (HITL) review in its most common form, and it has a known failure pattern. I’ve watched review queues drift toward approve-by-default once the first few hundred proposals look reasonable. A queue that says yes nine times out of ten trains its reviewers to say yes the tenth time. Amazon extended the same approve-and-act loop into outside tools on September 23, 2026: Amazon stated that 90% of its selling partners already use third-party AI tools, and that the Selling Partner plugin lets a seller’s own AI read Amazon business data and act on it with human approval.
Phave’s September 23, 2026 launch puts the design line in writing. Jon Miller, Phave’s co-founder and chief executive officer, said consent, frequency and quiet hours belong in rules, and that choosing which of two good emails a person receives is a job for a reasoning model. Phave stated that its Maestro engine computes a Playlist of touches for each person, ordered toward an objective the marketer sets, inside the rules the team keeps. Every marketing automation owner now faces the same split: write down which decisions are constraints no model may cross and which are judgments a model may make.
Phave also graded itself. Phave stated that it scored itself and legacy platforms against 481 requirements drawn from twelve enterprise evaluations, with Claude scoring and weighting each requirement, and reported 2.97 for Phave against 2.46 for HubSpot and 2.38 for Marketo. Phave picked the 481 questions; Claude marked the answers. Phave published the methodology at phave.com, which lets a buyer rerun it with its own requirement list, and prices the platform from $36,000 a year based on the number of people who receive something in a given month. The customer result in the release, Servion Global Solutions reporting more than double its marketing execution volume, measures output.
Akeneo’s September 23, 2026 survey shows where an automated decision meets the customer. Akeneo’s PX Pulse survey, conducted by Dynata in August 2026 among 1,000 US consumers, found that 77% had noticed the same product listed at different prices across retailers or platforms in the past year, that 24% already use tools such as ChatGPT or Google Gemini to compare prices or deals, and that 56% trust AI tools to provide accurate pricing information. Akeneo also found that 57% would trust a retailer less if they learned a product’s price had changed based on their personal information or shopping behavior.
Run a pricing workflow through those numbers. A home goods brand sells 150 SKUs on Amazon and on its own site, and writes a Seller Assistant workflow that matches a competitor’s price on any top-50 SKU and refreshes the listing. The workflow proposes 400 price changes in a month, and the team approves 90%, or 360 changes, at about 30 seconds each, three hours of review. Suppose 5% of the approved changes, 18 in all, drop the Amazon price below the brand’s own-site price by more than its parity tolerance, and each stays live for a week on a SKU selling 60 units a week with $2.40 of margin given up per unit. The brand gives away 18 × 60 × $2.40, or $2,592 a month, and the price gap sits in plain view of the 24% of Akeneo’s respondents who compare prices with AI tools. Three hours of review looks cheap until someone prices the 5%.
The repair is two lines of configuration and one named owner. A per-SKU price floor and a cross-channel parity rule go in as hard limits the workflow cannot override, the same category Miller assigns to consent and quiet hours. A person then samples approved changes every week against sell-through and margin. That sample produces the number Amazon’s release omits: the share of accepted recommendations that turned out wrong.
Citigroup brought card data into the same buying decisions. Citi’s U.S. Consumer Cards business launched Citi Commerce Media on September 23, 2026, a payments media network (PMN) that reaches Citi credit card customers across Citi.com, the Citi Mobile app and paid media properties. Citi stated that it serves more than 70 million US customers and sees 6.5 billion annual transactions across more than 700 spending categories, and that it plans to add the rewards platform Kard, which it agreed to acquire. Citi reported that initial campaigns delivered up to 5x incremental return on ad spend (iROAS) for an online retailer. One advertiser, an upper-bound qualifier and no published test design make that figure an opening question for Citi’s sales team. Brands that run offers built from card transactions can read Akeneo’s 57% finding as a caution about how visible personalization feels to the person receiving it.
Peec AI, the Berlin AI search visibility platform, added support on September 23, 2026 for Muse Spark, the model powering Meta’s personal AI agent Muse, which Meta introduced on September 8, 2026. Peec AI stated that Muse books travel, negotiates bills and completes purchases through Link, Stripe’s agent checkout, and Peec AI’s chief product and marketing officer Malte Landwehr described tracking the Muse Spark model as the closest approximation brands have to their performance inside Muse. A Generative Engine Optimization (GEO) report built on that data shows how a model answers test prompts. The agent’s actual purchase choices sit outside the measurement.
Liferay and Mole Street addressed the plumbing under those decisions. Liferay announced general availability of its Content Marketing Platform (CMP) on September 23, 2026 as an add-on that runs natively on Liferay CMS, sharing one data model, permission set and publishing infrastructure, so approval status and campaign context move with content from brief to publication. Mole Street, an Elite HubSpot Solutions Partner serving financial services firms, shared findings the same day from The State of Financial Services Growth, a 2026 survey of 270 US financial services leaders produced with HubSpot, in which 78% said organizational silos, disconnected systems and incomplete client data limit the client experience they can deliver. Mole Street reported that advanced AI adopters were 68% more likely to report significant improvement in client retention over the past year, a self-reported correlation from a survey the platform vendor co-authored.
Sort the seven September 23, 2026 announcements by who produced each number. Amazon measured agreement with its own assistant. Phave scored itself. Citi reported one campaign’s upper bound. Peec AI measures a proxy model. Mole Street and HubSpot surveyed the market HubSpot sells into. Akeneo’s consumer survey, commissioned by a company that sells product and pricing data software, is the one number in the set that measured the person on the receiving end of an automated decision.
Check what each number counted before you budget against it
Amazon’s more than 90% acceptance rate counts approvals and carries no accuracy, revenue or margin measure. Phave’s 2.97 score rests on requirements Phave selected, scored by a model Phave chose, and Servion’s doubled execution volume counts output. Citi’s up to 5x iROAS comes from one online retailer in initial campaigns, with no disclosed holdout size or measurement window. Peec AI’s Muse coverage tracks the Muse Spark model, which Peec AI itself describes as an approximation of brand performance inside the agent. Mole Street’s 68% retention finding is self-reported and correlational, from a survey HubSpot co-authored. Akeneo’s figures come from 1,000 US adults surveyed by Dynata in August 2026 on behalf of Akeneo, which sells the product and pricing data tools its findings support. Liferay published no customer or performance figures.
Every company in the set sells the product its number describes.
Strategic priorities for marketing leaders
- Report error rate next to acceptance rate. For every AI system that proposes actions, sample accepted recommendations each month and score them against the sales, margin or engagement that followed.
- Write hard limits before you switch on always-on workflows. Price floors, cross-channel price parity, consent, frequency and regulated claims belong outside any model’s discretion, configured as rules the workflow cannot override.
- Seed the approval queue. Insert a deliberately wrong proposal into human review each month and track whether reviewers catch it, so a 90% approval rate reflects judgment.
- Ask for the test design behind every incrementality claim. Before moving budget to Citi Commerce Media or any payments or retail media network, request the holdout size, the measurement window and the median result alongside the best one.
- Label proxy metrics as proxies. Report model-level visibility for agents such as Muse separately from any measured purchase or conversion data.
- Rerun vendor self-evaluations on your own requirements. Phave published its method, so score it and your incumbent platform against your team’s requirement list and weights before accepting any vendor scorecard.
Back to the opening question. A 90% acceptance rate tells a Chief Marketing Officer (CMO) how often the team agreed with the machine. Whether the machine was right shows up only in a sample someone pulls and scores. This week, pull 50 accepted AI recommendations from the system your team leans on most, check each against the sales, margin or engagement that followed, and put the error rate next to the acceptance rate in your next operating review.
Here’s The News for September 23, 2026
Amazon Adds Always-On Workflows and a Quick and Claude Plugin to Seller Assistant
Amazon.com, Inc. (NASDAQ: AMZN) announced on September 23, 2026, during Amazon Accelerate 2026, an update to Seller Assistant, its AI assistant for third-party sellers. Amazon stated that Seller Assistant now keeps persistent memory of each seller’s pricing patterns, inventory cycles and growth goals, and that new workflows monitor conditions and run routine tasks such as restocking and pricing adjustments around the clock, with seller-defined guardrails, seller approval of actions and complete audit trails. Amazon launched a Selling Partner plugin in Amazon Quick and in beta with Anthropic’s Claude for sellers in its US stores, and offered primary account holders a free 12-month Amazon Quick Plus subscription through December 31, 2026. Amazon stated that Seller Assistant has rolled out to more than 90% of its selling partners worldwide, that sellers accept its recommendations more than 90% of the time, and that 90% of selling partners already use third-party AI tools.
Read the full press release (Published: September 23, 2026 | Source: Amazon News (aboutamazon.com))
Phave Launches AI-Native Marketing Automation From Marketo Co-Founder Jon Miller
Phave came out of stealth on September 23, 2026 as an AI-native marketing automation platform built to replace Marketo, Pardot, Eloqua and HubSpot, founded by Jon Miller, co-founder of Marketo and Engagio, and Nick Bonfiglio, formerly EVP of global product at Marketo. Phave stated that its Maestro engine computes a Playlist of touches for each person, account and buying group toward an objective the marketer sets, within rules the team keeps, and that teams can run the platform through its own interface, their existing AI client, or headlessly through MCP and a versioned REST API. Phave reached general availability in August 2026, lists SambaNova, SPS Commerce, mabl, Servion and Hypha as users, and starts at $36,000 a year priced on the number of people who receive something in a given month. Phave reported a self-assessment against 481 requirements drawn from twelve enterprise evaluations, scored and weighted by Claude, in which Phave scored 2.97 against 2.46 for HubSpot and 2.38 for Marketo.
Read the full press release (Published: September 23, 2026 | Source: PR Newswire)
Akeneo PX Pulse Survey Finds 77% of US Shoppers Have Spotted Inconsistent Prices
Akeneo, the product experience (PX) company, released PX Pulse survey findings on September 23, 2026 from a Dynata survey of 1,000 US consumers aged 18 and older, conducted in August 2026. Akeneo reported that 59% of consumers say price matters more than it did six months ago, 32% completely or mostly trust retailers to offer a fair or competitive price, 79% have delayed a purchase expecting a lower price, and 77% have noticed the same product at different prices across retailers or platforms in the past year. Akeneo also reported that 24% use tools such as ChatGPT or Google Gemini to compare prices or deals, 56% trust AI tools for accurate pricing information, and 57% would trust a retailer less if a product’s price changed based on their personal information or shopping behavior.
Read the full press release (Published: September 23, 2026 | Source: PR Newswire)
Citi Launches Citi Commerce Media Built on Credit Card Transaction Insights
Citigroup Inc. (NYSE: C) announced on September 23, 2026 that its U.S. Consumer Cards business launched Citi Commerce Media, a media platform that connects brands with Citi credit card customers across Citi.com, the Citi Mobile app and paid media properties. Citi stated that it serves more than 70 million US customers and sees 6.5 billion annual transactions across more than 700 spending categories, and that the platform combines first-party transaction insights, paid media and closed-loop measurement, with the planned addition of Kard, a commerce media and rewards platform Citi agreed to acquire. Citi reported that initial campaigns across retail, payments, technology, and beauty and wellness categories delivered up to 5x incremental return on ad spend for an online retailer.
Read the full press release (Published: September 23, 2026 | Source: Citigroup)
Peec AI Adds Visibility Tracking for the Model Behind Meta’s Muse Agent
Peec AI, the Berlin AI search visibility platform, announced on September 23, 2026 support for Muse Spark, the large language model powering Meta’s personal AI agent Muse, which Meta introduced on September 8, 2026. Peec AI stated that marketing teams and agencies can now measure brand visibility, sentiment and cited sources in Muse Spark alongside ChatGPT, Google AI Mode, Google AI Overviews, Gemini, Perplexity, Microsoft Copilot, Claude, DeepSeek, Qwen, Naver and Mistral, and described Muse Spark tracking as the closest approximation brands have to their performance inside Muse. Peec AI reported more than 3,000 brand and agency customers, more than $15 million in annual recurring revenue since launching at the end of 2025, and $40 million raised from investors.
Read the full press release (Published: September 23, 2026 | Source: GlobeNewswire)
Liferay Announces General Availability of Liferay Content Marketing Platform
Liferay, a digital experience platform (DXP) provider, announced on September 23, 2026 general availability of Liferay Content Marketing Platform (CMP), an add-on to Liferay CMS for planning campaign content, coordinating work, managing content from brief to launch and measuring performance. Liferay stated that Liferay CMP runs natively on Liferay CMS with a shared data model, permissions and publishing infrastructure, includes a Content Population Matrix that maps content assets against target audiences and funnel stages before launch, and tracks tasks in calendar or Kanban views with configurable approval workflows. Liferay CMP is available with a 90-day free trial.
Read the full press release (Published: September 23, 2026 | Source: GlobeNewswire)
Mole Street Shares Financial Services Growth Research Produced With HubSpot
Mole Street, an Elite HubSpot Solutions Partner and technology consultancy for financial services firms, announced on September 23, 2026 that it is sharing findings at the Digital Marketing for Financial Services Midwest Summit in Chicago from The State of Financial Services Growth, a 2026 report co-authored with HubSpot and based on a survey of 270 US financial services leaders. Mole Street reported that 41% of leaders prioritize retaining and expanding existing relationships and 40% prioritize acquiring new clients and members, and that 78% say organizational silos, disconnected systems and incomplete client data limit the client experience they can deliver. Mole Street also reported that advanced AI adopters were 68% more likely to report significant improvement in client retention over the past year.
Read the full press release (Published: September 23, 2026 | Source: PR Newswire)
Frequently asked questions
What marketing technology news was announced on September 23, 2026?
Seven marketing technology and AI announcements carried September 23, 2026 dates. Amazon added always-on workflows and a Quick and Claude plugin to Seller Assistant, Phave launched AI-native marketing automation, Citi launched Citi Commerce Media, and Akeneo released PX Pulse consumer pricing survey results. Peec AI added tracking for the model behind Meta’s Muse agent, Liferay made its Content Marketing Platform generally available, and Mole Street shared financial services research produced with HubSpot.
What did Amazon announce for Seller Assistant on September 23, 2026?
Amazon announced at Amazon Accelerate 2026 on September 23, 2026 that Seller Assistant now keeps persistent memory of each seller’s business and runs always-on workflows for tasks such as restocking and pricing adjustments, with seller-approved actions and audit trails. Amazon also launched a Selling Partner plugin in Amazon Quick and in beta with Anthropic’s Claude for US sellers. Amazon stated that sellers accept Seller Assistant’s recommendations more than 90% of the time and published no outcome data for the accepted recommendations.
What is Phave, the marketing automation platform launched by Jon Miller?
Phave is an AI-native marketing automation platform that came out of stealth on September 23, 2026, founded by Marketo co-founder Jon Miller and former Marketo product executive Nick Bonfiglio. Phave’s Maestro engine computes a per-person sequence of touches toward an objective the marketer sets, within rules the team keeps, and pricing starts at $36,000 a year based on monthly recipients. Phave reported a Claude-scored self-assessment across 481 requirements in which it scored 2.97, against 2.46 for HubSpot and 2.38 for Marketo.
What did Akeneo’s September 2026 survey find about pricing and AI shopping tools?
Akeneo’s PX Pulse survey, conducted by Dynata in August 2026 among 1,000 US consumers and released September 23, 2026, found that 77% had noticed the same product at different prices across retailers or platforms in the past year. Akeneo reported that 24% already use tools such as ChatGPT or Google Gemini to compare prices or deals and that 56% trust AI tools for accurate pricing information. Akeneo also found that 57% would trust a retailer less if a price changed based on their personal information or shopping behavior.
What is Citi Commerce Media?
Citi Commerce Media is a media platform that Citigroup’s U.S. Consumer Cards business launched on September 23, 2026 to connect brands with Citi credit card customers across Citi.com, the Citi Mobile app and paid media properties. Citi stated that it serves more than 70 million US customers and sees 6.5 billion annual transactions across more than 700 spending categories, and that the platform uses closed-loop measurement. Citi reported up to 5x incremental return on ad spend for one online retailer in initial campaigns.
How should marketing leaders evaluate AI recommendation acceptance rates?
Marketing leaders can treat an acceptance rate, such as the more than 90% Amazon reported for Seller Assistant on September 23, 2026, as a measure of agreement and pair it with a measured error rate. A monthly sample of accepted recommendations scored against the sales, margin or engagement that followed produces that error rate. Hard limits such as price floors, cross-channel price parity and consent rules belong outside any AI workflow’s discretion.
