When A Search Query Becomes A Standing Instruction

Google's AI Mode monitoring and OpenAI's dots keep working after a request. Tasks, sources, and controls differ, as do the questions about site visibility.

SEJ STAFF Matt G. Southern

53 seconds ago
⋅
12 min read

SEJ STAFF Matt G. Southern Senior News Writer at Search Engine Journal

Bio

When A Search Query Becomes A Standing Instruction

Google started making AI Mode’s information monitoring available to all users on Sept. 28, after initially offering it to Ultra and Pro subscribers. Then, on Sept. 29, OpenAI launched its new always-on ChatGPT agents called dots. These tools serve different roles. One keeps track of changing information within Search and sends an update, while the other manages assigned tasks across connected apps and a cloud computer.

What they share is that the request stays active even after the user leaves. For local businesses or retailers, this means that later updates could include information that appeared after the initial search. I like to think of a request that stays active as a “standing instruction.” That’s my own term, not an official one from Google or OpenAI.

I also covered the Sept. 28 rollout and the help pages for dots on Sept. 30. Google’s Search Console documentation includes AI Mode in its reporting, but the pages reviewed for this article do not explain how monitoring updates are counted or identified separately.

Standing Requests Predate Both Products

Google Alerts helps you stay informed about topics you care about by converting them into regular searches. According to its help page, you enter a topic, select your preferred notification frequency, choose the site types, set a language and region, and specify the number of results. Google will then email you “whenever we find matching search results.”

OpenAI’s side has a precedent too. Its release notes date scheduled tasks in ChatGPT to January 2025, with support for future and recurring tasks. The July 9, 2026 entry introducing ChatGPT Work says Work can use scheduled tasks “that run once, repeat on a schedule or trigger, or monitor for changes.” The current Scheduled tasks help page defines monitoring tasks as checks that “send a notification when a relevant update occurs” and can “use information from previous runs.”

Recurring checks were available on both sides before this week. Google’s monitoring and OpenAI’s dots handle ongoing requests differently, including which sources they can reach and what they return.

Google’s account of the gap in the older tools came from Liz Reid, VP of Search, at I/O in May. In an interview Roger Montti covered in July, she described people who “have to keep checking and checking and checking,” and a stock-tracking request with a specific set of parameters where, in her words, “Nothing’s going to do that.”

How Google Turns A Search Into A Monitoring Task

Google’s I/O post in May introduced information agents that “look across everything on the web, like blogs, news sites and social posts,” plus Google’s real-time finance, shopping, and sports data. Reid wrote that an apartment hunter can “brain dump all of the exact requirements” and the agent “will continuously scan for you.” The company’s I/O roundup said you create one by adding “keep me updated” to a search, and that updates arrive “along with helpful links to explore further on the web.”

The feature reached Ultra subscribers on June 12, across all AI Mode languages and markets. Robby Stein, VP of Product for Google Search, said then that agents would send “detailed updates and links to the web.”

His Sept. 28 post refers to “info monitoring capabilities” rather than “information agents” and explains that Search will “continuously check across changing info on the web” and suggest tasks during searches. It doesn’t repeat the links language from May and June, and talks about updates and suggested tasks instead.

Reid’s May post also mentions that the agent sends a synthesized update “with the ability to take action.” It doesn’t specify what action is taken, who is responsible for taking it, or if approval is needed. Additionally, the September post doesn’t mention any actions. None of the May announcements and June and September posts reviewed here explains how Search decides which sources to include in an update.

How OpenAI Turns A Prompt Into Ongoing Work

OpenAI’s launch post describes dots as “always-on agents” with their own cloud computer, their own browser, and access to a person’s connected apps. The getting started page says a dot “can take on ongoing responsibility and keep making progress between conversations.”

Assigned work is what most resembles Google’s monitoring. OpenAI’s Tasks and memory page describes a dot that “keeps working between conversations and follows through as things change.” It can “decide when to pause and wake up to continue,” and a user “can also specify a time or a supported event that should trigger work.” For recurring tasks, the page suggests that users specify what to check, the timing, the changes that should trigger notifications, and where to deliver results.

Event monitoring relies on the connected service being able to support it. As the same page mentions, “Connecting Slack or another source alone doesn’t create a monitoring task.”

Proactive research runs alongside assigned work and does not require a new request. The privacy and safety page explains that a dot can gather helpful updates by reading permitted connected sources and saving private notes, even before anyone asks. Its research tools can’t directly message others, change content through plugins, or control browsers or computers. Additionally, the Tasks page highlights that this research can connect new information to previous work, like noticing that a release decision conflicts with a draft shared last week.

That limit puts a browser-based website watch outside of proactive research. The computers and apps page describes a cloud computer and browser that the dot uses for its work, and mentions that “Some websites block cloud browsers or require further verification.”

Dots are rolling out to Pro users outside the European Economic Area, Switzerland, and the UK, as well as to Business Premium users across all supported regions. The Enterprise beta is off by default.

The Shared Layer Is Persistent Intent

When you look at them together, Google’s monitoring and OpenAI’s dots have something in common. They both aim for a goal that the user sets, then check the information later to see if it meets that goal. This is my own analysis, not an official statement from either company.

Google’s May post highlights that you’ll receive a notification “when listings meet your needs.” OpenAI’s Tasks page suggests that dot users tell the system which changes deserve a notification. In both cases, once you set the criteria, the system applies it to information that arrives later.

Google chooses the source categories that Search checks. For a dot, access depends on connected accounts and granted permissions. Task instructions can specify which sources to use. However, none of the pages reviewed here explains how a specific source is selected for a given update.

The May post talks about a synthesized update, while OpenAI’s launch post focuses on results retrieved for review, like drafts, analyses, and suggested actions. Google’s June post states that updates happen “the moment new info is available,” whereas a dot operates on a schedule, responds to a supported event, or acts on its own timing between conversations.

The documentation clearly shows that only one side describes whether an action follows and what approvals are in place. OpenAI’s safety page explains features like Custom Rules and Auto-review, including user-defined limits, automatic checks, and steps like password changes that are returned to the individual. Meanwhile, Google’s side only mentions the “ability to take action” from the May post.

Google’s leadership has described this direction before. In an interview Roger Montti covered in April, Sundar Pichai pointed out that “a lot of what are just information seeking queries will be agentic search,” and that users “will do long-running tasks, can be asynchronous.” These are Google’s projections, not confirmed behaviors.

Where The Comparison Ends

Google’s feature is embedded in Search, providing updates derived from web content and real-time data. As described in September, what it returns is an update.

Dots are general-purpose agents capable of using linked apps and a browser, managing their notes alongside ChatGPT’s memory, and executing or preparing actions within OpenAI’s guidelines. The launch announcement and help pages refer to them as agents rather than search tools. They’re separate tools, not versions of each other.

Why This Matters For Website Visibility

For both products, an update prepared later can include information that did not exist when the request was made. This is more of an analysis than a confirmed behavior. The pages I looked at don’t detail the step-by-step process behind an update. So, nothing here indicates exactly when a page is fetched, evaluated, or selected.

For a local business, the September post examples covered nearby new restaurants, pop-ups, and local holiday activities for kids. To illustrate rather than report, a September task might include an update about a business opening in October. The post cites sites, forums, and social media as sources but doesn’t clarify how Search chooses among these for local topics.

For retailers, examples included back-in-stock alerts and price-drop notices, and the post named Google’s Shopping Graph, a product data set, as a source. It doesn’t specify if price updates come via merchant feeds, page crawls, or both. Google’s AI features page advises site owners to keep their Merchant Center and Business Profile data up to date.

For a site that publishes articles, Google’s May post included news sites in the list of sources scanned by agents. An article published after someone set a task could be part of a subsequent update. The example does not establish whether a particular article would be selected, linked, or clicked.

Measurement is broken down into three distinct events, and they shouldn’t be combined. As noted on Google’s AI features page, sites featured in AI Overviews and AI Mode are also included in the Performance report under the Web search type. The Generative AI performance report, which Google states had already reached websites worldwide by Aug. 31, counts impressions, that is, how many times links to a site are shown to users in a generative AI feature. A retrieval alone is not a displayed-link impression. A click-through is another event. Google counts external-link clicks in AI Mode in Search Console, while Analytics measures subsequent activity on the site. As of Sept. 30, neither the AI features page nor the report page mentions monitoring updates.

An agency could hand a dot a recurring check across its connected platforms. That’s internal tooling. It doesn’t reveal how customers find a site or whether any visit follows. Watching a public competitor page through a browser would fall outside proactive research, and as of Sept. 30, OpenAI’s crawler page shows four user agents, but none are for dots.

What The Platforms Have Not Explained

The May announcements and June and September posts reviewed here do not clarify how Search selects sources for monitoring updates, whether every update includes links, or if some tasks involve additional fetching and under which user agents. The September post mentions “globally” but lacks details on the involved languages or when Pro access was added.

OpenAI’s launch post, along with three help pages, two Learn pages, and the crawler page reviewed here, does not specify which browser or fetch identity a dot uses on the public web. They also do not clarify whether updates based on public-web content include citations or how a site owner can distinguish between a dot’s visit and other ChatGPT activities. However, these pages note that proactive research cannot directly control the browser. OpenAI says authorized background tasks follow its usual action rules and safety checks. The reviewed pages do not identify the public-web requests those tasks generate.

Looking Ahead

Google announced the wider rollout on Sept. 28, ending the September post with “Lots more to come!” A Google help page for monitoring or an update in Search Console could clarify how monitoring updates are counted. In its Sept. 29 release notes, OpenAI said it would share dots usage terms for each plan after the first month. A dot-specific public-web identifier could help site owners recognize its requests. Citation guidance could clarify how source links appear in updates.

More resources:

Featured Image: Google and OpenAI. Source: Google.

Category SEO AI Search