Google unveiled three significant updates to its Search and AI Mode this week, as detailed by Roger Montti for Search Engine Journal. His article inspired an examination of these advancements, the overarching trends they signal, and their potential impact on search functionality this year.
Upon closer inspection, these updates suggest a shift from what was traditionally a results-page-centric experience towards a more integrated approach to task execution.
Recent Announcements from Google
The tech giant has introduced individual hotel price tracking features in Search, now accessible globally for users signed in and searching in English and Spanish. Users receive email alerts notifying them of price fluctuations during selected timeframes.
In addition, Canvas trip planning transitioned from a Labs preview to general availability in the U.S. this March, allowing users to describe their travel plans and receive personalized itineraries complete with flights, accommodation, and attractions that save automatically.
Furthermore, agent-powered store calling, originally launched in classic Search, will soon be available in AI Mode. This innovation empowers Google’s AI to contact nearby stores to verify inventory by utilizing Gemini models and Duplex technology.
Product leader Rose Yao shared these developments on X. Further information can be found in a dedicated Google blog post.
Identifying the Pattern
These updates illustrate a continuing product direction from Google, evident in its research, patent filings, and executive commentary since the year’s outset.
In January, Google released the SAGE research paper, which explored training agents for reasoning across a four-step process—this laid the groundwork for executing multi-step tasks within Search.
Pichai’s interview in April made this direction public. He commented, “Many queries that were traditionally information-seeking will become agent-driven in Search.” Our analysis noted a shift in his terminology from vague predictions of change to explicit mentions of task execution.
Earlier this month, Montti asserted that task-oriented, agent-driven search was already reshaping SEO, particularly highlighted by Google’s global rollout of agent-based restaurant booking—evidence that the future Pichai described has already manifested in current offerings.
Recently, the U.S. Patent Office published a continuation patent from Google titled “Autonomously providing search results post-facto.” This filing outlines a system that waits for answers when none are immediately available and subsequently provides them through assistant interactions.
These recent updates are consistent with this ongoing trajectory. The Canvas feature’s broader U.S. rollout occurred approximately five months after its November launch.
The introduction of store calling in AI Mode follows its initial release in Search last November. Additionally, the individual property price tracking feature is now available in Search.
Microsoft’s recent advancements align with this evolving trend. Sumit Chauhan, President of Microsoft’s Office Product Group, noted in a company blog that Copilot’s agent-driven capabilities are now fully available across Word, Excel, and PowerPoint:
“Copilot creates the most value when it performs the work—formatting, restructuring, building visuals, and transforming data—rather than simply suggesting steps.”
These features are now standard for Microsoft 365 Copilot and Premium subscribers and are also available to Personal and Family plans. However, it remains uncertain whether businesses will receive analogous reporting for agent-driven features, as this was not addressed in Microsoft’s post.
Unsettled Terminology
Google employs “agentic” in its product announcements, characterizing features such as AI Mode and calling them task-focused.
The partnership with SeatGeek was branded as “Google’s Agentic AI Search Experience.” Other companies are similarly adopting a framework that emphasizes agency in the language.
Pichai envisions a future wherein ‘Agent manager’ serves as Google’s role, positioning Search as ‘an agent manager’ responsible for overseeing various tasks. This concept frames Google as an orchestration layer rather than direct competition.
Montti has referred to this phenomenon as “task-based agentic search,” occasionally abbreviated as TBAS—his shorthand for this evolving discourse, not yet standardized industry language.
The term “agentic” refers to capability, while “Agent manager” indicates a specific functional role that Google claims. “Task-based” emphasizes the user’s objectives. The presence of three distinct labels within a single month indicates that the industry is still grappling with precise nomenclature.
Implications for Search Practitioners
The features rolled out this week redefine visibility across numerous business sectors.
Local retailers are now faced with a novel discovery layer. In AI Mode, Google’s agents will reach out to businesses, confirming stock and other details, rather than relying on users.
Google has yet to reveal which stores will initially be contacted, the criteria for eligibility, or whether specific business attributes influence this selection process.
An analysis of 68 million AI crawler visits across 858,457 Duda-hosted sites has shown that sites connected with Yext, Google Business Profile, and review systems experienced more frequent crawls than those lacking such associations.
These findings merely reflect crawler dynamics and do not provide clarity on agent-driven calls. It remains uncertain whether similar indicators will affect which stores are approached.
Hotels and travel-related enterprises must now contend with property-specific price tracking. Itineraries generated by Canvas rely on its inherent selection criteria.
However, there’s no comprehensive report indicating if a particular hotel appeared in a Canvas-selected plan, triggered an alert, or was recognized in an AI Mode response.
Publishers continue to face intensified scrutiny from AI-generated content summarization. An analysis by Index Exchange of 1,200 publishers showed that 69% encountered year-over-year declines in advertising opportunities, with an average reduction of 14%.
These declines varied by sector, with health and career publishers experiencing drops of 40-50%, while news and politics publishers observed mere 7% decreases.
Vanessa Otero, Founder and CEO of Ad Fontes Media, stated in the same report:
When it’s critical to be accurately and thoroughly informed about significant events, a reputable news source is still far superior to an AI chatbot, which may provide generic or misleading information.
Users recognize this, which explains why most news consumers lean toward their trusted websites.
Historically, news has performed well for advertisers, and if the trend of resilience among news sites persists, this segment of inventory will likely be the most coveted in the forthcoming open web.
Travel publishers are similarly pressured as Canvas generates itineraries without acknowledging sources, leaving publications uninformed on whether their content influenced trip arrangements.
Ecommerce retailers find themselves in the dark regarding which stores are contacted, impeding their ability to ascertain if inventory feeds, listing accuracy, or Google Business Profile indicators are effective.
The multi-platform nature of discovery creates complications in strategy. Google favors structured data and verified profiles for its agents; meanwhile, Perplexity Computer coordinates across 19 models with diverse retrieval preferences.
ChatGPT Atlas extracts content directly from browsers, while OpenAI’s Operator employs GUI vision to interact with rendered pages.
Each business is represented across multiple discovery avenues, each with unique technical requirements. A single-strategy optimization no longer suffices.
Areas of Opacity
Since our initial coverage identified the measurement deficit, this gap has further widened.
Search professionals remain uninformed about whether their establishments were included in Canvas trip plans.
They lack visibility on agent outreach efforts or whether their hotel was highlighted in a price-tracking alert. Furthermore, it’s unclear how often their content was utilized to establish someone else’s itinerary.
Alongside these updates, no new reporting mechanisms have been introduced. Alphabet reported $63.1 billion in Google Search & Other advertising revenue during Q4 2025, reflecting a 17% year-over-year increase, with management attributing success to advances in Search, Cloud services, and AI utilization.
However, no new reporting tools have emerged to assist businesses in tracking their contributions to this AI-enhanced search environment.
This pattern emerges across platforms. ChatGPT referral statistics are strictly what OpenAI chooses to disclose. Visibility regarding citations in Perplexity is limited to its own platform. Google’s agent functionalities do not directly align with the Search Console interface.
Academic research concerning agent training is continuously evolving, as evidenced by two recent papers on arXiv spotlighting the rapid progress.
One, CW-GRPO, authored by Junzhe Wang and colleagues, suggests enhancements to multi-turn search agents through reinforcement learning.
The other, SKILL0, created by Zhengxi Lu and collaborators at Zhejiang University, trains agents to internalize skill packages, enabling them to operate autonomously without instructional overhead during functioning.
The pace of evolution in the training infrastructure has outstripped that of measurement frameworks on which businesses rely.
Closing this disparity is not within the power of search professionals alone; Google, OpenAI, Perplexity, and Anthropic must all provide equivalent reporting for their agent-related functions. However, none have publicly committed to doing so.
Future Projections
Pichai anticipates that 2027 will mark “a significant inflection point for various components.” He referenced non-engineering workflows and certain agent-driven business processes. Our analysis has traced through that timeline.
May heralds the Google I/O and Microsoft Build events, where both companies are expected to unveil expansions to their agent-driven functionalities—this makes tracking and reporting tools an urgent priority.
Without visibility into their role in task-oriented search, businesses cannot effectively optimize or advocate regarding financial obligations.
Two enduring questions linger beneath this discussion. Pay-per-click models were viable when user interactions generated clicks. However, features like store calling, Canvas planning, and price tracking do not yield clicks, and no platform has proposed a substitute for this paradigm.
Schema.org was established for crawling by search engines and is not tailored for agents requiring real-time inventory, booking access, and actionable endpoints. The standards for agent-readable business data have not yet advanced to meet these needs.
Future developments hinge on whether any platform is willing to build out reporting capabilities alongside its functionalities.
Thus far, none have articulated how they would approach this. Until a shift occurs, businesses will continue to navigate optimization for interactions they cannot monitor. Upcoming signals will emerge at I/O and Build in the next three weeks.
Source link: Searchenginejournal.com.
