Why do agents waste tokens on web search today?

Traditional search APIs return only URLs and thirty-word snippets, forcing agents to fetch raw HTML and strip markup before any reasoning can begin. On multi-step searches, agents re-bill the entire growing context window on every hop, paying a "retrieval tax" that compounds across query iterations.

How much more expensive is snippet-based search than indexed documents?

A single web search loop answering "What was Y2K?" cost roughly 3,750 tokens on the first hop, but three hops climbed to roughly 28,700 tokens (more than 4× the cost of an owned index, which took roughly 6,900 tokens total).

What questions can only be answered with full documents?

Questions that require joining data across multiple records—for example, finding every target account that hired someone into a data or AI leadership role in the last quarter requires reading full role histories plus recent news together, work that neither a search snippet nor a neural search tool can do on the first pass.

Large Language ModelsAI Stocks & MarketsDaily Dose of Data ScienceJul 16, 2026Jul 17, 2026, 06:00 JST

Agents Need Better Web Search—Full Documents, Not Snippets

Send on LINE

3 Key Points

What happened
A analysis of how AI agents search the web shows that traditional search APIs returning only snippets and links force agents to fetch and clean raw HTML on every query. Running the same question through three retrieval methods—plain search results, neural search, and a pre-indexed web service—revealed that looping through snippets costs roughly 4× more tokens than accessing full pre-processed documents. For a research task requiring multiple searches, agents repeatedly repay a "retrieval tax" on content they've already seen.
Why it matters
As coding agents and LLM-powered systems move to production, validation shifts from offline tests to live telemetry—meaning agents must handle real-world queries they were never trained on. Web search is how agents stay current with markets, news, and organizational changes, but today's search APIs waste tokens on overhead instead of reasoning. For teams building agent loops (research pipelines, GTM briefings, market monitoring), the choice of search tool directly controls cost and speed—not the model or prompt.
What to watch
Seltz, a purpose-built web index for agents, returns complete structured documents (people profiles, full article text, Wikipedia entries) in a single call rather than forcing agents to reconstruct content from snippets. It works best for deep-dive lookups after discovery is done; chaining discovery queries (open web search) with deep-dive queries (indexed full documents) lets each loop iteration use the right retrieval method for what it's actually trying to do.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Get news like this every morning, free

Context & Analysis

The article frames a structural mismatch in how AI agents interact with the web. Coding agents have collapsed the cost of software generation, but validation capacity—the ability to catch failures before production—has not kept pace. AI-generated code fails on unknown-unknowns because no human formed a mental model while writing it, forcing validation to happen live on real traffic. The same challenge applies to prompt changes and retrieval-augmented generation (RAG): offline tests cannot predict production breakdowns. This shift to production-first validation demands raw, high-cardinality telemetry that can be sliced after the fact by any attribute the operator needs.

Within that frame, web search emerges as a critical bottleneck for agents dealing with present-day data—market changes, new hires, published news, price updates. The article quantifies a hidden cost: every search call that returns snippets instead of full documents forces the agent to fetch, parse, and clean HTML, repaying this cost on every loop iteration. The token audit (28,700 tokens for three hops through snippets vs. 6,900 for a single query of an indexed document) makes visible the compounding tax on context-window size. The article argues the fix is not a smarter model or prompt, but a change to the search interface itself—returning finished, structured documents so agents reason rather than preprocess.

FAQ

Why do agents waste tokens on web search today?: Traditional search APIs return only URLs and thirty-word snippets, forcing agents to fetch raw HTML and strip markup before any reasoning can begin. On multi-step searches, agents re-bill the entire growing context window on every hop, paying a "retrieval tax" that compounds across query iterations.
How much more expensive is snippet-based search than indexed documents?: A single web search loop answering "What was Y2K?" cost roughly 3,750 tokens on the first hop, but three hops climbed to roughly 28,700 tokens (more than 4× the cost of an owned index, which took roughly 6,900 tokens total).
What questions can only be answered with full documents?: Questions that require joining data across multiple records—for example, finding every target account that hired someone into a data or AI leadership role in the last quarter requires reading full role histories plus recent news together, work that neither a search snippet nor a neural search tool can do on the first pass.

Daily Dose of Data ScienceRead Original Article

Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Get it free

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.