AI Marketing News, 31 Aug 2026: Claudeforce, Meta's Hatch Agents, AI Referrals Up 163%
Salesforce puts CRM inside Claude with Claudeforce, Meta preps its Hatch agent platform, and Brainlabs finds AI referrals up 163% as organic falls 10.5%.
Salesforce and Anthropic launch Claudeforce, putting the CRM inside Claude. Announced August 26, the expanded partnership ships “Salesforce in Claude,” a plugin with 37 prebuilt sales skills that lets sellers reason over live revenue data, update pipeline, and take governed actions without opening Salesforce. It is live with select pilot customers now, with open beta expected in September 2026 and additional skill packs starting late 2026. Salesforce is pitching this as AI becoming the interface to enterprise software, with more Claude, Salesforce and Slack integrations planned. VentureBeat’s readout of the launch was blunter: Salesforce says you may never need its app again.
Importance for marketers: The CRM becomes a data layer that an agent reads and writes. That changes what “CRM hygiene” is for: every field standard, dedupe rule and lifecycle definition you maintain is now context an agent consumes before it acts, and bad data turns into bad autonomous actions rather than bad dashboards. If your stack runs on HubSpot instead, the same pattern is already buildable today with Claude connected over MCP, as laid out in the Claude, Clay and HubSpot stack.
Brainlabs: AI search cut organic sessions 10.5% but AI referrals convert at 1.5x. Digiday covered Brainlabs data from 54 advertisers showing organic search sessions fell 10.5% after Google AI Overviews became widespread, with 46 of 54 clients seeing declines. Referral traffic from AI platforms rose 163% over the same period, AI-driven key events rose 335%, and visitors arriving from ChatGPT, Copilot, Gemini and Perplexity completed key events at roughly 1.5 times the rate of organic search visitors. The read: AI interfaces are absorbing early-stage informational queries, and the users who still click through arrive later in their decision process.
Importance for marketers: The session count fell but the commercial value per session rose, which means any report still ranking channels by traffic volume is now scoring AI referrals as a failure while they outconvert organic. Split AI referrals into their own channel grouping, weight them by key events rather than sessions, and re-baseline your organic targets against the 10.5% structural decline instead of chasing it back. The measurement gap underneath this is the same one covered in the zero-click attribution gap.
The IAB is building a framework for attributing conversions influenced by AI agents. Digiday reports the Interactive Advertising Bureau is developing a measurement framework for crediting AI’s role in conversions as agents increasingly mediate discovery and purchasing. The work covers what evidence should count when referrers and UTM parameters disappear, including cases where an AI system reads content, compares products, and completes or influences a purchase. The framework is expected to distinguish AI creating awareness or intent from AI directly making the decision. Publishers are also pushing for credit when their content feeds AI answers, and limited platform transparency remains the obstacle to independent measurement.
Importance for marketers: UTMs and referrer headers are the substrate most B2B attribution is built on, and agent-mediated journeys carry neither. Do not wait for the standard. Log what you control now: server-side conversion capture, self-reported attribution fields, and AI-crawler hits on your content. That is your evidence base for the day platform-reported agentic conversions arrive with no way to audit them.
X opens ad campaign data to outside AI tools through an MCP server. Social Media Today reports X launched an Ads Model Context Protocol server that lets advertisers connect campaign data and analytics to ChatGPT, Claude Code, Grok or custom agents, querying performance and getting campaign guidance in natural language instead of working inside X’s own tooling. Meta, TikTok, Pinterest and Snapchat have announced similar MCP integrations, making this the fifth major social platform to expose its ads system to external agents.
Importance for marketers: The ad platforms are conceding that the interface layer is moving to whichever agent you already work in, and cross-platform reporting that used to require an ETL vendor or a stitched-together dashboard becomes a single conversation across five MCP connections. Connect the X, Meta and TikTok MCP servers to one internal agent and hand it your naming conventions. That is a working cross-channel analyst, running before any of the platforms ship their own.
OpenAI tests exclusion targeting for ChatGPT ads as off-topic placements persist. EMarketer reports OpenAI is testing exclusion controls with a limited advertiser group, letting brands block topics and conversation types near their ads. The move answers complaints about weak targeting and reporting: one study found 14.4% of ChatGPT ads were unrelated to the user’s query, with off-topic rates above 50% in some categories. OpenAI has also cut the minimum spend commitment from $200,000 to $50,000 while expanding the pilot internationally, though CPMs remain high and the buying infrastructure trails Google and Meta.
Importance for marketers: A 75% cut to the minimum commitment is OpenAI widening the funnel of test budgets before the measurement stack is ready, and the 14.4% off-topic rate is what you are buying into at high CPMs. The mechanism that matters is conversational adjacency: your ad renders inside an answer, so placement quality is a function of query relevance, not site lists. Test at the new $50k floor if the audience fits, but log placement context from day one and hold the spend flat until exclusions and reporting exit the invite-only phase.
Meta readies Hatch, a consumer AI agent platform, plus a new model in October. Meta is reportedly preparing to launch Hatch, a consumer agent platform built to act across outside services including DoorDash, Etsy, Reddit, Yelp and Outlook, followed by a new model codenamed Watermelon in October. Meta has considered premium subscriptions up to $199.99 per month and plans to let WhatsApp users connect with third-party AI agents, while continuing open-weight model releases in parallel. This is Meta’s first serious attempt to earn direct revenue from its AI spend rather than routing it through ads.
Importance for marketers: Agents transacting through WhatsApp puts an execution layer inside the messaging channel where roughly three billion people already are, which matters more for consumer brands than another chatbot launch. The mechanism to watch is service integration: if Hatch agents book, order and buy through partner services, being one of those integrated services becomes a distribution decision, like the app store placement fights of 2010 but for agent access.
The Trade Desk rebuilds its DSP around agentic buying with Kokai Zuma. MediaPost reports The Trade Desk introduced Kokai Zuma, an overhaul of its platform adding a redesigned interface, upgraded forecasting, and an expanded Koa assistant that plans, buys, measures and optimises campaigns through conversational workflows, with back-end AI predicting available inventory and modelling outcomes in real time. The company claims early results average a 32% improvement in cost per acquisition. The launch lands amid agency criticism, analyst downgrades, and disintermediation pressure from cloud and ad platforms selling cheaper automated buying.
Importance for marketers: A DSP under competitive pressure automating its own operators is the pattern to read here, not the CPA claim, which is early, self-reported and survivor-biased. As Koa absorbs the planning and optimisation work, the fee you pay The Trade Desk buys less human-verifiable decision-making, so the auditable part of your programmatic operation shifts to what you control: log-level data access, incrementality tests, and supply-path reporting. Renegotiate for those before renewing, because agentic buying without them is a black box with a chat window.
Google adds pay-as-you-go pricing and hard spend caps to Gemini Enterprise. Axios reports Google Cloud is adding pricing controls to Gemini Enterprise aimed at corporate anxiety over unpredictable AI costs: pay-as-you-go access as an alternative to per-seat subscriptions, monthly project spending limits that pause agent API calls when hit, committed-spend discounts, and off-peak task processing discounted up to 50%. Google frames it as lowering the barrier to experimentation before companies know how widely employees will use AI.
Importance for marketers: Spend caps that pause agent API calls are the first native budget circuit-breaker from a major vendor, and they change how AI projects get approved: a capped pilot is a finance conversation, an uncapped per-seat rollout is a procurement one. For GTM automation specifically, off-peak processing at half price fits batch enrichment and scoring jobs that run overnight anyway. Price your always-on agents and your batch workloads separately before committing, because they now cost different amounts.