---
title: AWS Just Gave Agents Domain and Date Filters for Web Search. Most Marketing Agents Are Still Citing Whatever Ranks.
description: AWS expanded Web Search on Bedrock AgentCore with domain allow-lists, published-date filters, and new EU and Asia Pacific regions. Here is why marketing agents grounded in unfiltered web search are a brand safety problem, and how to configure the controls that ship with the tool.
author: LETSGROW Dev Team
date: 2026-08-29
category: AI Tools
tags: ["AI Tools", "AWS Bedrock", "AI Agents", "Brand Safety", "Data Residency"]
url: "https://letsgrow.dev/blog/bedrock-agentcore-web-search-domain-date-filters"
---
# AWS Just Gave Agents Domain and Date Filters for Web Search. Most Marketing Agents Are Still Citing Whatever Ranks.

Grounding an AI agent in live web search sounds like a solved problem. It is not. Most marketing teams that wired a browsing or search tool into an agent gave it one instruction: go find an answer. Nobody told it where to look, how recent the answer needs to be, or which domains count as authoritative. That gap has been sitting in production for a year, quietly deciding which sources your agent treats as fact in front of a prospect, a support ticket, or a competitive research brief.

AWS just closed part of that gap. In August 2026, Amazon expanded Web Search on Bedrock AgentCore, the managed grounding tool it launched in June, with per-request domain filtering and published-date filtering, and pushed the service into Europe and Asia Pacific. If your team runs agents on Bedrock, you now have controls you did not have two months ago. If you do not run on Bedrock, the update is still worth your attention, because it previews the control layer every agent platform is about to ship, and most marketing orgs have no plan for actually using it.

## What Actually Shipped

Web Search on Bedrock AgentCore is a fully managed tool that lets an agent query the live web and get back cited snippets, source URLs, titles, and publish dates, all without customer data leaving the AWS account. It runs as a built-in connector target on the AgentCore Gateway over MCP, which means any agent already speaking Model Context Protocol can pick it up without a custom scraping layer. It launched in June priced at seven dollars per thousand queries, available only in US East.

The August update adds two things marketing teams should actually care about. Domain filtering lets you scope a query to an allow list or block list, so an agent researching a competitor's pricing can be told to only trust the competitor's own site, or explicitly excluded from citing a specific low-trust domain. Published-date filtering lets you set a recency window, so a query about current pricing or a compliance deadline cannot pull an answer from a three-year-old cached page. AWS also expanded the service into Europe (Ireland) and Asia Pacific (Tokyo), which matters for any team that has to keep customer-facing agent traffic inside a specific region for regulatory reasons.

## The Accountability Gap Nobody Configured

Here is the part most teams skipped when they first bolted a search tool onto an agent: an unfiltered web search connector has no concept of trust. It ranks by relevance, not by whether the source is your competitor's marketing copy, an outdated support article, or a content farm that scraped your own site and got the numbers wrong. An agent answering a prospect's question about your pricing, your integrations, or your security posture will happily cite whatever ranked highest, and until now, Bedrock-based teams had no lever to stop it.

This is not a hypothetical risk. It is the same failure mode marketing teams have already lived through with early RAG systems that pulled from stale internal docs, except now the source pool is the entire open web instead of a controlled knowledge base. The blast radius is bigger, because the agent is not misquoting your own outdated slide deck, it is misquoting a stranger's blog post and presenting it as researched fact.

::checklist{title="Signs your agent's web search has no accountability layer"}
- Your agent can cite any domain on the open web, including competitor sites and forums, with no allow list or block list
- Nobody has set a recency window, so a query about current pricing or policy can surface a page from years ago
- There is no logging that captures which source URL an agent actually cited for a given customer-facing answer
- Your legal or brand team has never reviewed which domains an agent is permitted to treat as authoritative
- You are running customer-facing agents in the EU or APAC on infrastructure that processes the search query in a different region
::

If two or more of those are true, your agent's grounding is a liability wearing a helpfulness costume.

## Why This Is a Marketing Problem, Not Just an Infrastructure One

Engineering teams will read this as an AgentCore Gateway configuration update. Marketing teams should read it as a brand safety and compliance control that finally exists. Every agent your team ships that touches a prospect, a customer, or a piece of public content is making implicit editorial decisions about which sources it trusts. Domain filtering turns that implicit decision into an explicit, auditable setting. Date filtering stops your agent from confidently repeating information that was true a year ago and is not true now. Regional availability means a team bound by GDPR or similar data residency rules can finally use a managed grounding tool instead of building one in-house just to keep query data in-region.

::compare-table{title="Grounding options for a customer-facing marketing agent"}
| Approach | Source control | Maintenance burden | Best for |
| --- | --- | --- | --- |
| Unfiltered web search tool | None, ranks by relevance only | Low to set up, high risk later | Prototypes and internal research only |
| Managed search with domain and date filters | High, explicit allow lists and recency windows | Low, configuration not code | Customer-facing agents that need current, trustworthy sources |
| Roll-your-own scraper and filter logic | High, but you own every edge case | High, ongoing maintenance and legal exposure | Teams with a specific need the managed tool does not cover |
| RAG over owned content only | Total, but limited to what you have indexed | Medium, needs the re-embedding discipline covered elsewhere | Answers about your own product, docs, and policies |
::

Most marketing agents need a mix of the last two rows: RAG over owned content for anything about your own product, and filtered web search for anything that requires current, external information. The mistake is deploying either one unfiltered and assuming relevance equals trustworthiness.

## What to Configure Before Your Next Agent Goes Live

Do not wait for a bad citation to force this conversation. Build the allow list and block list for any agent with web search access before launch, not after a prospect gets fed a competitor's numbers as fact. Set a default recency window scoped to what the query type actually needs, tight for pricing and compliance questions, looser for general research. Log the cited source URL on every answer your agent gives, so when something goes wrong you can trace it to a specific page instead of shrugging at a black box. And if you operate in the EU or APAC, move your agent traffic to the region that matches your customers before a compliance review makes you do it under deadline pressure.

The tool finally exists. The teams that configure it before something breaks are the ones who get to call this infrastructure. Everyone else is going to call it an incident.