AI Agents, Shrinking Search Shelves, and Governance Gaps: The Week That Was

Alex: Welcome to The Authority Report. I'm Alex, joined as always by Sam. This week we're covering the period September twenty-eighth to October fourth, twenty twenty-six — and it was a big one. AI agents went commercial, organic search rankings stopped meaning what they used to, India introduced advertising disclosure rules, and a governance crisis that's been building quietly got a lot harder to ignore.

Sam: Let's start at the top. OpenAI had a major launch this week — what happened?

Alex: On September twenty-ninth, OpenAI introduced something called Dots at its DevDay event. These are persistent AI agents — each one is assigned an ongoing goal and keeps working toward it even after you step away. They get their own cloud computer and browser, connect to more than four thousand apps, and can surface updates through Slack or Microsoft Teams.

Sam: So it's not just a chatbot you prompt and wait on — it keeps going on its own. Who can actually use it right now?

Alex: The initial rollout covers Pro, Business Premium, and Enterprise plans, one Dot per user. Reuters reported that business data is not used for training by default, and that sensitive actions — like changing passwords or permanently deleting data — require explicit consent from the user.

Sam: That's a meaningful privacy assurance. But does the evidence back up the performance claims?

Alex: Not yet. OpenAI has not published independent performance data, and the live DevDay demonstration included glitches — that was noted by Axios. For marketing teams the clearest near-term use is recurring operational work: campaign monitoring, content pipelines, recurring reports. But as one analysis put it, an agent that works continuously will reproduce weak instructions continuously too.

Sam: Good flag. Now let's talk search, because the data that came out this week was genuinely striking.

Alex: Really striking. Fayfo analysed Google Search Console's new Generative AI report across one major travel site covering June and July twenty twenty-six. They looked at eight hundred and ninety-one URLs that appeared in both organic and AI datasets. Here's the headline number: nearly sixty-nine percent of AI impressions came from pages ranked between positions four and ten — while just zero-point-three-five percent came from pages in the top three positions.

Sam: So being number one, two, or three in Google almost doesn't help you in AI search. That completely inverts the old logic.

Alex: Exactly. And there's more. Pew Research findings, summarised by The Guardian, show that when an AI Overview appears in Google Search, users click traditional results only eight percent of the time, compared with fifteen percent when no summary is present. Separately, Ahrefs data cited by Voltage president Corey Morris shows the top-ranked page loses roughly fifty-eight percent of its clicks when an AI Overview is present.

Sam: So what actually drives AI visibility if rank doesn't?

Alex: The Fayfo analysis identified two main factors: brand-topic association — where a company already holds strong recognition and third-party references across the web — and content usefulness, meaning clear answers, practical guidance, specific details. High-traffic navigational or transactional pages tended to underperform because AI answers simply don't add much value there.

Sam: That's a fundamental rethink for any team still reporting SEO success purely through ranking positions. Let's move on — there was a lot happening in AI creative production this week too.

Alex: Yes, and this one matters because it shows scaled AI creative is no longer just for the giants. Wyndham Hotels and Resorts, Opella, and BetMGM are all building in-house AI creative teams. Wyndham reported a fifteen-times increase in asset output and a seventy-five percent reduction in production time, working with AI firm Adora. Opella said it has now produced more than twenty thousand pieces of content using generative tools.

Sam: That's a dramatic shift in production economics. And regulation is starting to catch up, right?

Alex: India's Advertising Standards Council, ASCI, released new guidelines this week requiring disclosure when synthetically generated content could materially influence consumer decisions. The rules take effect three months after publication. Critically, ASCI specified that adding an AI label cannot make an otherwise misleading advertisement compliant.

Sam: So brands can't just slap an AI badge on something deceptive and consider themselves covered. That's an important line to draw. Let's talk platforms — there were two notable moves in marketing automation.

Alex: Jon Miller — co-founder of Marketo — launched a new platform called Phave on September twenty-third with co-founder Nick Bonfiglio. Its AI engine, Maestro, sequences marketing touches dynamically for individuals, accounts, and buying groups rather than following fixed rule-based journeys. It's priced at thirty-six thousand dollars a year.

Sam: The Marketo co-founder building a direct Marketo challenger — that's quite a statement. Is there independent evidence it works?

Alex: Not yet. The benchmark Phave published is company-produced and not independently audited. Worth watching, but worth scrutinising. Meanwhile, ZoomInfo acquired an agentic orchestration platform called DoubleO.ai on September thirtieth, and made its Agent Teams product available inside GTM Studio the very next day — with no new contract required for existing customers.

Sam: No new contract is a smart activation play. What can Agent Teams actually do?

Alex: The example ZoomInfo highlighted is tracking former deal champions who move to new companies — agents confirm fit, reconstruct what the champion did in the original deal, identify the new buying committee, and draft outreach. ZoomInfo founder Henry Schuck framed it simply: revenue doesn't follow a flowchart. But the reliability evidence right now rests on a single named customer reference, so caution is warranted.

Sam: Good to flag that. Now there was some sobering research out this week about the people being asked to deploy all this AI — the CX leaders. What did the data show?

Alex: A Gartner survey of three hundred and twenty-one customer service and support leaders found ninety-one percent faced executive pressure to implement AI. But job architecture and hiring sit with HR, while platforms and budgets sit with IT or a central AI function. The mandate exists; the authority doesn't.

Sam: Accountability without control. And how ready are organisations to govern what they're deploying?

Alex: A separate Grant Thornton survey of nine hundred and fifty senior leaders found seventy-eight percent lacked confidence they could pass an independent AI governance audit within ninety days. Forty-eight percent of boards had never set governance expectations for AI at all. And nearly three in four organisations have given agentic AI access to their data and processes, while only one in five has a tested incident response plan.

Sam: That last number is alarming. Three in four have agentic AI in their systems; one in five has a plan if something goes wrong. There was also a very personal story this week that brought AI identity risks into sharp focus.

Alex: Yes — Adweek reported on September twenty-ninth that Rodney Rambo, who had served as global CMO at Converse, departed in January twenty twenty-six — months before a disputed advertising campaign ran. But AI answer engines continued to associate his name with the brand, and he told Adweek the erroneous association exposed him to online abuse and threats.

Sam: So the AI wasn't making a sophisticated interpretive error — it was just working from stale data?

Alex: Exactly. The State of AI Marketing's analysis noted the wrong association came from an out-of-date corporate relationship, not a difficult interpretation of the campaign itself. An AI answer engine delivers a conclusion — a name paired with a brand — without surfacing the stale directory listing that supplied the connection. The recommended fix is in the public record: keep current leadership pages with accurate tenure dates and publish transition notes when senior leaders depart.

Sam: That's actionable. And rounding out the week, there was a bigger argument about where brand governance needs to go — upstream into the systems themselves?

Alex: CMSWire argued on September twenty-ninth that the pool of systems now capable of producing customer-facing content has grown to effectively unlimited scale, making post-production review a bottleneck rather than a safeguard. The call is to designate explicit sources of truth with named owners, replace screenshots and slide decks with design tokens and structured content models, and tier human oversight by risk level.

Sam: And Ibexa's CEO made a similar point earlier in the year, didn't he?

Alex: He did. At Ibexa's Orchestration Day in June twenty twenty-six, CEO Bertrand Maugain warned that AI compounds martech fragmentation rather than resolving it, because a lack of context across fragmented SaaS stacks means AI can surface messaging an organisation has long since abandoned. Access to all those sources, he argued, does not equal authority over them.

Sam: The common thread across nearly every story this week seems to be the same: the capability is moving faster than the governance.

Alex: That's exactly right, and I think that's the story of this week. Agents are live, AI search is rewriting visibility economics, creative production has scaled into the mid-market, and regulation is starting to arrive — but the governance infrastructure to manage all of it is still catching up. Thanks so much for listening to The Authority Report. We'll be back next week with more.