Information Agents are background AI agents Google runs continuously on behalf of users (launching summer 2026 to Google AI Pro and Ultra subscribers first) that monitor blogs, news sites, social posts, and real-time feeds for user-defined criteria. If your content updates frequently, an Information Agent might revisit you every hour, every day, for months. You did not ask for this traffic. You will get it anyway.
Per Elizabeth Reid's I/O 2026 post on Search, Information Agents will "intelligently look across everything on the web, like blogs, news sites and social posts" along with real-time finance, shopping, and sports data. The trigger is a user setting up a persistent monitoring task. The mechanism is a continuous, low-volume but never-stopping background crawl pointed at the sources Google deems relevant. The implication, if your site is one of those sources, is 5 to 10x your normal crawl load from a class of bot that didn't exist eighteen months ago.
> Information Agents monitor for user-defined criteria across blogs, news, social posts, finance, shopping, and sports. Pro/Ultra-tier first in summer 2026. The crawl is continuous; the 2 MB byte limit per fetch still applies; freshness signals decide whether you're in the monitored set.
What an Information Agent actually does
The user sets up a persistent task in AI Mode. Examples Google has used in announcements:
- "Notify me when the iPhone 17 Pro drops to $899 or below"
- "Alert me when a three-bedroom listing appears in the Glebe neighborhood under $900K"
- "Tell me when there's a Toronto FC ticket below face value"
- "Watch for new posts from these five local-news blogs that mention the city council"
Each one of those tasks compiles into a background-agent loop. The loop fans out across the data sources Google ranks as relevant for that query. It re-checks them on whatever schedule the model decides is reasonable. It reads the new data, compares it to the user's criteria, and either sits quiet or pings the user with an update.
Multiply that by every Pro and Ultra subscriber running multiple persistent tasks at once, with Google's own data showing AI Mode at 1 billion monthly active users one year after launch and planning-focused queries growing 80% faster than the overall AI Mode growth rate per Shivani Mohan's data post, and you get a continuous background crawl pattern that looks nothing like Googlebot's traditional rhythm.
The crawl-pressure implication
Googlebot of 2018 visited your site according to a schedule shaped by content-change signals, sitemap priority, and backlink velocity. It was bursty and predictable, something you could budget for.
Information Agents are different. They visit when the user's task says they should, not when your site's update cadence suggests they should. For a local-news blog tracked by a hundred different users with overlapping criteria, that can mean an agent checking the homepage every fifteen minutes during news cycles. For a real-estate listing site monitored by buyers across dozens of neighborhoods, every new listing gets indexed within minutes.
We expect, conservatively, 5 to 10x the bot traffic for sites that publish high-frequency content in monitored categories: finance, real estate, local news, sports, shopping, and shopping-adjacent commerce. For sites publishing stable, evergreen content (the cabinet maker, the family law firm, the dentist with two articles a year), Information Agents are barely a blip.
This deserves its own post on what's coming for crawl budget. The short version: the crawl-budget rules Google has historically published assumed a Googlebot economy. The Information Agent economy is layered on top, and the 2 MB byte limit per fetch still applies on every visit. If your pages exceed 2 MB and you're a high-frequency publisher, you're feeding agents truncated content many thousands of times a day.
How to be discovered by them
The discovery model for Information Agents is the same as for AI Mode synthesis at large, but stricter, because the agent re-validates source quality on every monitoring loop. Four things matter:
Clean structured data. Schema.org markup that accurately reflects current content. For news, NewsArticle with proper datePublished and dateModified. For real estate, Offer and Product types with current price and availability. For local business, LocalBusiness with current hours and contact. The structured data tells the agent what to extract; the freshness of that data tells the agent whether to keep monitoring you.
Fresh content signals. Last-Modified headers that match reality. Sitemap entries that update when content does. Stable URLs that don't break. The agent reads update signals before deciding to re-fetch the body, so a stale Last-Modified header gets you skipped on the next pass.
RSS or JSON feed availability. This is the single most useful thing a publisher can do. An RSS or JSON feed gives the agent a structured changelog without re-crawling the homepage every fifteen minutes. We see this dropping crawl pressure by 60 to 80% on sites that ship it well, while keeping the agent in the loop.
Semantic clarity. The agent needs to extract a specific fact (price, address, score, availability) and verify it against the user's criteria. Pages that bury the fact in marketing copy or load it via JavaScript that the agent isn't waiting to render lose. Pages that present the fact in a single, parseable element win.
Cover those four and you're discoverable. Skip any one of them and you're either invisible or accidentally heavy.
The 2 MB byte limit still applies
This is where the crawl-budget and Information Agent stories meet. Per Google's March 2026 crawler blog post, every Googlebot fetch, and every fetch by the bots running on the Information Agent stack, uses the first 2 MB of the page for indexing. Past 2 MB, content is silently truncated.
If you publish 4 MB pages because you're shipping a 1.5 MB hero image, a 600 KB analytics suite, and 800 KB of unused JavaScript, half your page never reaches the agent. Multiply that across thousands of fetches from monitored sources and you have a content-visibility problem you can't fix by writing better copy.
This is why the 2 MB rule became a foundation-level concern at the same time Information Agents started shipping. The two are not coincidental: the byte limit is the cap, the agents are the demand.
What this means for high-intent local
For contractors, home builders, real estate teams, and anyone with high-intent local search exposure (see our contractor playbook), Information Agents are the channel to actively design for. The user query "find a contractor in my area who specializes in basement waterproofing and has availability in the next month" used to be a one-time search. It is now a persistent task. The agent will check your site once a day, look at your availability calendar, look at your service area, and either ping the user or sit quiet.
Three things matter for local. First, your service area and specialties have to be in structured data, not just prose on the homepage. Second, your availability, if you publish it, has to be machine-readable. Third, your contact information has to be consistent across Business Profile, schema, and the actual page content. Inconsistent data gets you dropped from the monitored set.
The cost of blocking them
You can block Information Agents via robots.txt the same way you can block any other Google crawler. We don't recommend it.
The traffic from Information Agents is not the goal. The exposure inside the monitored alerts is. When a user's Information Agent pings them with "three new listings in the Glebe under $900K," your listing is either one of those three or it isn't. The bot traffic that gets you into the alert is the cost of admission. Block the bot and you're not in the alert. You aren't surfaced in your customer's monitored set. You are, in a real sense, invisible to a category of high-intent users who never search the way they used to.
For Pro and Ultra subscribers in summer 2026, monitored alerts will become the dominant way these users hear about anything that updates frequently. That is the user base most likely to be researching a major purchase, a service contract, a property, or a local provider. Blocking the agent is blocking those users.
What we tell clients
Three things, repeatedly.
First, get your structured data accurate and fresh, today. Not as a nice-to-have. As the precondition for Information Agent monitoring.
Second, ship an RSS or JSON feed for anything that updates more than weekly. This is the cheapest, most useful move available. It cuts crawl pressure and improves your odds of staying in the monitored set.
Third, audit your page weight. If you're over 2 MB on any page you care about being monitored, get under 2 MB. The math is unforgiving and the cost of fixing it is much smaller than the cost of being silently truncated thousands of times a day.
The bot traffic is coming. Whether it counts as exposure or just overhead depends on what you ship now.
Sources
- A new era for AI Search, Elizabeth Reid, Google
- How AI Mode is changing the way people search in the U.S., Shivani Mohan, Google
- AI in Search: Going beyond information to intelligence, Elizabeth Reid, Google
- I/O 2026: Welcome to the agentic Gemini era, Sundar Pichai, Google
- Personal Intelligence expansion, Google