AI Marketing News, 5 Oct 2026: OpenAI Dots, AI Overviews on 80% of Branded Queries, Reddit Ends Public API
OpenAI ships always-on dots agents, AI Overviews hit 80% of branded queries, Reddit ends RSS and its public API, Sonnet 5.5 matches Opus at half the price.
OpenAI launches dots, always-on agents that work across ChatGPT, Slack and Teams. Announced at DevDay on September 29, dots run on GPT-6 Astra with their own cloud computer and browser, connect to more than 4,000 apps through plugins, and keep working toward goals when nobody is prompting them. In the background a dot runs “proactive research” across your connected apps using read-only tools, and Custom Rules let you allow an action, require approval for it or block it outright, with all work logged in an Activity View. Dots are rolling out to Pro and Business Premium plans in eligible markets, while Enterprise workspaces get a beta that admins must switch on. OpenAI is also piloting specialist dots with their own identity and credentials for set responsibilities, naming email marketing among its internal tests, and is integrating them with Microsoft Agent 365. The same event added @ChatGPT in Slack and Teams, team tasks that fire on a new email or Slack message, and an OpenAI Marketplace whose first 32 partners include HubSpot and Salesforce.
Importance for marketers: A dot connected to HubSpot can read every deal, note and email thread in the objects it reaches, and the Custom Rules someone on your team writes are the only control between that context and an outbound send. Write them before the first seat connects a CRM: block email sends and sequence enrolment, require approval for any field update, and allow read access only to the objects the person already owns. Treat a specialist dot like a service account, with its own HubSpot user, field-level permissions and a line in the agent inventory RevOps keeps.
Google AI Overviews jump from 26% to 80% of tracked branded queries in a week. DemandSphere data reported by Search Engine Land on October 1 shows AI Overviews on roughly 27% to 35% of its tracked branded queries through most of September, then 69.21% on September 26 and a peak of 90.48% on September 27. Presence held at 80.23% on September 28 and 82.06% on September 29. Separately, Nectiv co-founder Chris Long tested 100 enterprise brands with SerpApi and found AI Overviews for 93 of them, including Salesforce, Mailchimp and Reddit. Google has not announced a change, and neither dataset measured clicks, so the effect on branded click-through rate is still unknown.
Importance for marketers: Branded search is where your outbound, paid social and word of mouth land, so a prospect who searches your name after a cold email now reads Google’s summary of you before your own homepage, assembled from review sites, forums and competitor comparisons. Pull your top 50 branded queries from Search Console (brand plus pricing, reviews, integrations and “vs” terms), record the AI Overview text and cited sources weekly, and fix or create the pages behind any wrong answer. Compare October’s branded CTR against September before you blame outbound for a softer reply-to-meeting rate. The audit process is in What AI assistants say about your product, and how to audit it.
Reddit ends RSS feeds on November 13 and closes its public API by March 2027. Reddit announced on September 30 that RSS stops working on November 13, new public API requests end October 31, and apps that have not registered start losing access on January 12, 2027, with the remaining public endpoints closing in March. For RSS use outside a community you moderate, Reddit says “there is no replacement.” Developers are being pushed to its Devvit platform with a $1 million App Migration Program paying $1,000 per eligible app, applications due November 30. Reddit cites scraping and automated abuse. Its other revenue line, which carries AI licensing deals with Google and OpenAI, grew 24% year over year to $43 million in Q2.
Importance for marketers: Plenty of signal-based workflows run on a subreddit RSS feed piped into n8n: competitor mentions, “alternatives to” threads, category complaints that seed outbound angles and content briefs. Every one of those breaks on November 13 with no error in your dashboard, just an empty feed. Audit your n8n, Zapier and Clay workflows for reddit.com/.rss and API calls this month, and ask your social listening vendor in writing whether it collects through licensed access or scraping. If Reddit is a primary source of customer language for you, budget for a licensed listening tool now rather than finding the gap in January.
Claude Sonnet 5.5 and GPT-6.1 Sol both land at $2 and $10 per million tokens with near-flagship results. Anthropic released Sonnet 5.5 on September 28 at Sonnet 5’s pricing of $2 input and $10 output, against $4 and $20 for Opus 5.5. On GDPval-AA, a benchmark built from real occupational tasks, Sonnet 5.5 scores 1844 to Opus 5.5’s 1846, and it beats Opus on Terminal-Bench 4.0 at 70.6% to 66.4%. Anthropic says it runs more than 30% faster than Sonnet 5 and costs up to 30% less per task, with Haiku 5.5 due in the coming weeks. A day later OpenAI shipped GPT-6.1 Sol at the same $2 and $10, with cached input at $0.10, describing it as near-Astra intelligence at a fifth of Astra’s token prices.
Importance for marketers: Last week’s split was a cheap model for bulk classification and a frontier model for judgement. Sonnet 5.5 sits two points below Opus 5.5 on occupational work, so account research briefs, call summaries and first-draft proposals can run at half the Opus rate. Anthropic itself says Opus stays stronger on open-ended work that needs sustained judgment, so keep it on messaging strategy and exception handling. Re-run your 20-case test set on Sonnet 5.5 and GPT-6.1 Sol this week, and switch every Clay and n8n step where the score gap falls inside your tolerance.
ChatGPT Ads adds bulk product campaigns and conversion reporting beyond the optimisation goal. Search Engine Land reported on September 30 that advertisers can now create multiple product feed campaigns and ad groups by spreadsheet upload in Ads Manager, with ad templates generated automatically. The Products tab now shows review status and the reason a product was rejected. The public Insights API can return attributed conversion events beyond a campaign’s chosen optimisation goal, so a campaign optimising toward one event can still report on others. Advertisers can now manage account, brand and legal names separately, though a brand-name change pauses serving until review.
Importance for marketers: The Insights API change is the one B2B teams should act on. You can optimise toward a demo request while pulling attributed pipeline-stage events into your reporting, which means a scheduled n8n job can join ChatGPT Ads conversions to HubSpot deal stages without changing how the campaign bids. Set that join up before you scale spend, so the channel review in Q1 runs on opportunity data instead of form fills. The pipeline test to hold it to is in ChatGPT Ads as a B2B pipeline channel.
Google now says it is “critical” to manually fact-check all AI-generated content before publishing. Google updated its Search guidance on generative AI content on October 1, adding that generative models “don’t retrieve facts, but predict a likely sequence of words” and that outputs may contain hallucinations. The new line reads: “It is critical to manually factcheck and review all AI-generated content for accuracy and trustworthiness before publishing.” Google says the change brings its documentation in line with what it presents at developer events. The update lands while the September 2026 spam update is still rolling out.
Importance for marketers: If you run programmatic pages, integration pages or comparison content generated from a Clay table or an n8n pipeline, the review step now needs to exist in the workflow and leave a record. Add a human approval node before publish, log who approved each page and when, and pull any template that ships numbers, pricing or competitor claims without a source field. Pages that went live unreviewed during the spam update window are the ones to check first.
Salesforce agrees to buy AI customer research platform Listen Labs, reportedly for about $2 billion. Announced September 29, Listen Labs automates research design, participant recruitment, interviews and analysis, with access to more than 50 million research participants and interviews in more than 120 languages. It also builds “digital twins,” AI simulations meant to predict how customers respond to products and messages before testing with people. Salesforce says Listen Labs will complement Marketing Cloud and Service Cloud and give its agents more customer context. The deal is expected to close in Salesforce’s fiscal fourth quarter of 2027, and integration details have not been announced.
Importance for marketers: Once interview transcripts and simulated customer responses both sit inside the CRM, an agent drafting messaging can cite either, and the output will not tell you which. Tag every research input in your CRM by evidence type (recorded interview, survey, simulation) before it feeds any agent, and require that positioning and ICP changes trace back to recorded interviews. Use simulations to cut the list of message variants you test, then let real reply and meeting rates decide the winner.