All issues
Discovery Digest · June 15, 2026

Issue 03. The week the agent layer went to work

This week, AI surfaces stopped waiting for queries. Google's Information Agents began monitoring the web and notifying users proactively. Meta's WhatsApp agent started handling product discovery for three billion users. OpenAI opened catalogue-scale product advertising inside ChatGPT. Anthropic brought its most capable public model to the same audience as ChatGPT and Gemini. And Google made Preferred Sources the first user-controlled visibility lever inside AI answers. The consistent signal: AI is now acting on behalf of users, not just responding to them, and brands that have not built for this shift are starting to feel it.

Issue 03. The week the agent layer went to work
01 · OpenAI

OpenAI opens ChatGPT to product feed advertising with two-million-item catalogues

What
OpenAI launched product feed campaigns in its ChatGPT Ads Manager Beta on 12 June 2026, allowing retail advertisers to upload catalogues of up to two million product items and generate ads automatically from feed data. Feed-based ads have been the strongest-performing format in the beta since the platform opened to all US businesses in May 2026. The launch follows the pilot's expansion to the UK on 6 June and an announced rollout to Japan, South Korea, Brazil, and Mexico.
When
Announced in the OpenAI Ads Manager Beta, Thu 12 June 2026, reported by Search Engine Land.
How it shifts discovery
ChatGPT is no longer only an answer surface: it is now a commerce surface at catalogue scale. For brands that compete on product discovery, this changes the calculus on AI visibility spend. A consumer who asks ChatGPT which product to buy can now receive a listing from your catalogue rather than a citation from a review site. The question is no longer whether to advertise in AI assistants, but how to structure feeds, bids, and creative for a conversational purchase journey where the ad appears inside an answer, not alongside one.
Questions to ask
  • Are our product feeds structured to maximise relevance inside a conversational context, or are they built only for traditional comparison-shopping engines?
  • How do we measure attribution when a ChatGPT product ad influences a purchase that completes on our own site, outside the assistant?
  • Do we have a creative and pricing strategy for AI purchase-intent moments that is distinct from our Google Shopping and Meta catalogue campaigns?
Sources
02 · Google Search

Google adds Preferred Sources to AI Overviews and AI Mode, doubling click-through for selected publishers

What
Google extended its Preferred Sources feature into AI Overviews and AI Mode, with a rollout beginning in June 2026. Users who have selected favourite sites in their Google account settings now see content from those sites labelled and highlighted inside AI answers, and in new topic carousels for developing and breaking stories. Google reports that users are twice as likely to click through to a Preferred Source result compared with unlabelled AI citations.
When
Announced by Google and reported by 9to5Google, Wed 27 May 2026, rolling out through June 2026.
How it shifts discovery
Preferred Sources introduces a user-controlled citation layer inside AI answers: the first mechanism in Google's AI surfaces that rewards existing user loyalty rather than solely model judgement or content structure. For publishers and brands, being selected as a preferred source by enough users becomes a compounding visibility multiplier that operates independently of ranking or model citation logic. Brands with strong direct audiences are better placed to accumulate preference signals than brands that rely on discovery as the primary acquisition route.
Questions to ask
  • How many of our existing users have us set as a Preferred Source in Google, and are we communicating clearly why they should?
  • Is Preferred Source status tracked as a distinct visibility channel in our reporting, or conflated with organic ranking?
  • Which content formats are most likely to earn Preferred Source selection from a returning user, and are those formats receiving sufficient investment?
Sources
03 · Google Search

Google Search Information Agents reach all AI Mode languages for Ultra subscribers

What
Google confirmed on 12 June 2026 that Information Agents in Search are now available across all AI Mode languages and markets for Google AI Ultra subscribers. Agents let users define monitoring criteria, then run autonomously in the background, watching the web for relevant changes and delivering digest notifications. The feature was announced at I/O 2026 in May and reached full AI Mode language coverage in June, with a broader rollout to all users planned for later in the year.
When
Confirmed by Google VP of Search Robby Stein and reported by ppc.land, Thu 12 June 2026.
How it shifts discovery
Proactive monitoring agents change the competitive context for brands: visibility is no longer only about being found when a user searches, it is about being surfaced when an agent runs a pre-set watch on a topic. A user who has set an agent to track their category, a competitor, or a keyword set will receive brand mentions and product developments as notifications rather than search results. Brands that do not appear in agent-retrieved results are absent from a growing share of the decision-making context that consumers bring to purchase, without any query ever having been typed.
Questions to ask
  • Have we tested what our brand returns inside a Google Information Agent monitor for our primary product categories, and is the output accurate and complete?
  • How does our content perform for agent-driven monitoring tasks versus a direct user query on the same topic?
  • Are we treating content optimised for proactive agent retrieval as a distinct workstream, or assuming standard search optimisation is sufficient?
Sources
04 · Anthropic

Anthropic releases Claude Fable 5 publicly, adding a third major AI answer surface for brand visibility

What
Anthropic released Claude Fable 5 to the public on 9 June 2026, bringing the model that had been available only to trusted-access users under the Mythos programme to all users on claude.ai, the Anthropic API, and Claude Code. Fable 5 delivers Anthropic's strongest publicly available performance across reasoning, long-context tasks, and agentic workloads, with enhanced safeguards developed during the restricted preview period.
When
Released publicly on claude.ai and the Anthropic API, Mon 9 June 2026.
How it shifts discovery
With ChatGPT and Gemini already established as citation surfaces that search teams track and optimise for, Fable 5 raises the minimum viable AI search audit to three model families. Citation patterns, brand entity recognition, and how factual claims about a product are represented behave differently across Claude, GPT, and Gemini because each model trains on different data and applies different retrieval logic. Teams running AI visibility measurement on one or two models are working with an incomplete picture, and Fable 5 is materially more capable than previous Claude releases, which means its outputs will carry more weight with the users who rely on Claude for research and decision-making.
Questions to ask
  • Do we have a visibility baseline for how our brand is represented in Claude Fable 5 responses, and does it differ meaningfully from ChatGPT or Gemini?
  • Is our AI citation monitoring covering three major model families, or are we still tracking only one or two surfaces?
  • Which product categories or factual claims about our brand are most likely to be handled differently by Claude versus OpenAI, and do we understand why?
Sources
05 · Meta

Meta's AI agent for WhatsApp Business goes global, making three billion users a brand discovery surface

What
Meta made its AI agent for WhatsApp Business available globally on 3 June 2026, following regional pilots. The agent handles customer enquiries, product discovery, appointment scheduling, and commerce tasks for businesses through WhatsApp, backed by Meta's Llama models and the WhatsApp Business Platform. WhatsApp serves more than three billion monthly active users, making this the largest single rollout of a conversational AI agent to a business discovery surface to date.
When
Announced by Meta and reported by TechCrunch, Wed 3 June 2026.
How it shifts discovery
WhatsApp has functioned as a dark-funnel channel for brand-consumer interaction in markets where messaging is the primary digital interface: South Asia, Latin America, the Middle East, and Africa. The global availability of Meta's AI agent turns that channel into a structured, AI-mediated discovery and transaction layer. For brands operating in those markets, the first point of AI contact with a consumer may now be a WhatsApp conversation rather than a search query, and the infrastructure requirements for that layer are different from web search or app-based discovery.
Questions to ask
  • In which markets does WhatsApp represent a primary customer communication channel for us, and have we deployed a WhatsApp Business AI agent in those markets?
  • How does AI-mediated product discovery inside WhatsApp fit into our attribution model, given the private and conversational nature of the interaction?
  • Is our brand information, product catalogue, and FAQ content structured for a WhatsApp AI agent, or only for web and app surfaces?
Sources

Key takeaways

What to walk away with this week

  1. ChatGPT product feed ads turn AI conversation into a commerce surface at catalogue scale. Treat the ChatGPT Ads Manager as a third commerce channel alongside Google Shopping and Meta this quarter, not a future experiment.

  2. Google Preferred Sources rewards existing user loyalty inside AI answers, independent of model citation logic. Make accumulating Preferred Source status from your current audience a measurable objective before the feature reaches the full Search user base.

  3. Google's Information Agents and Meta's WhatsApp agent share one implication: AI is now proactively surfacing brands to users without a query. Content that is not structured for agent retrieval will be absent from this layer entirely.

  4. Claude Fable 5 means AI citation monitoring on one or two models is no longer sufficient. Add Fable to your brand visibility audit and treat ChatGPT, Gemini, and Claude as three independent citation surfaces with different logic.

  5. The discovery journey in WhatsApp-first markets is about to be intermediated by Meta AI. If you have customers in South Asia, Latin America, the Middle East, or Africa and no WhatsApp Business Platform infrastructure, close that gap before competitors do.

Previous issue

Issue 04. The week measurement finally caught up with AI traffic

Read issue

The Discovery Digest · Every Friday

Stay ahead of AI Search

Five updates a week across ChatGPT, Claude, Gemini, Perplexity, Copilot, Grok and Google AI Overviews, with the questions worth asking.

Free5 updates weeklyUnsubscribe anytime