A deep research AI agent verifies its answers by running a loop instead of a single search. It breaks the question into sub-questions, searches at least three independent sources, reads each result in full, pulls out every factual claim with its source URL, and scores that claim on a four-tier confidence scale before the claim is allowed into the answer. When two sources disagree, the agent shows both sides rather than picking one. That loop is the difference between an answer that sounds right and one that has been checked.
OpenAI launched Deep Research mode in February 2025, Google followed with Gemini Deep Research, and Perplexity added Pro Search. Those products run the loop well for general questions, but you cannot control the loop structure, confidence scoring or source selection, so an agent you build yourself is the one that shows the verification steps in full. This matters because single-shot answers are not reliable enough for professional use. A 2025 Stanford University study of AI legal research tools found they hallucinate between 17 and 33 per cent of the time even when purpose-built for one domain.
We run this loop ourselves. Every post on this blog starts from a research pack of 30 to 40 sources gathered by the same search, evaluate and iterate pattern, and nothing gets published without the verification steps below. Here is how the loop works, exactly how to set it up in the tools your team already uses, and what separates an answer you can trust from one that just sounds right.
A deep research AI agent runs on iteration, not a single search
Most people picture deep research as one very thorough search, but it is actually a loop that runs search, read, evaluate, refine, search again. The agent starts with a question, runs an initial search across at least three sources, reads the full content of promising results, decides if the answers are sufficient, and if not, generates a more specific sub-query and repeats. The loop keeps running until the agent has enough corroborated evidence to answer with a confidence rating.
OpenAI's Deep Research mode operates on exactly this principle. When you feed it a complex question, it runs 5 to 15 search iterations, reads 30 to 100 sources and produces a synthesis with inline citations. The loop structure matters more than the quality of any single search result. A mediocre search engine in a tight loop outperforms a world-class search engine used once.
Step 1: Break the question into sub-questions
The first thing a deep research agent does is decompose your question. If you ask 'What is the market size for AI agents in healthcare?', the agent does not search for that exact phrase. It breaks it into sub-questions: total AI market size, healthcare AI adoption rates, percentage of AI spend going to agents, recent funding rounds. Each sub-question gets its own search loop, and the agent synthesises the answers at the end. The more granular your decomposition, the more likely you catch conflicting data.
In practice: In Hermes, sub-question decomposition lives in a skill file. Create ~/.hermes/skills/deep-research.md with a rule that says: "Before searching, split the user's question into at least 3 sub-questions. Search each one independently, then synthesise." The agent loads this skill at session start and applies the decomposition rule to every research request. In Claude Code, the same logic goes into your CLAUDE.md under a research conventions section. Keep the decomposition rule simple, about five lines, so the agent can follow it without overthinking.
Step 2: Search with multiple sources
A single search engine introduces blind spots. The agent should query at least three sources: a web search for broad results, a news search for recent developments and a specialised source like arXiv for academic research or Crunchbase for funding data. If two sources disagree, the agent flags the discrepancy rather than picking a side and smoothing it over.
Perplexity's Pro Search is a good example of multi-source searching done well, because it queries several indices simultaneously and surfaces conflicting information rather than faking consensus. In a custom agent, you replicate this by wiring multiple MCP tool servers, each connected to a different search API. Hermes supports this natively: define separate web-search, news-search and academic-search tools in your MCP config, and the agent calls them in parallel. The 10-layer AI agent stack covers how tools (layer 6) and skills (layer 5) wire together for exactly this kind of multi-source workflow.
Why it matters: Single-source answers are a leading cause of AI hallucination in enterprise settings. The Stanford study is the clearest evidence, and even purpose-built research tools hallucinate between 17 and 33 per cent of the time. When a claim matters, the agent needs independent confirmation from a second and third source before it counts.
Step 3: Read and extract claims
For each result, the agent reads the full content, not just the search snippet. It extracts specific claims, such as numbers, dates, names and statistics, and records the source URL, publication date and a confidence flag for each claim. This is what separates a deep research agent from a regular chat answer. Every claim has a backlink to its origin. You can click through and verify the agent's work.
In practice: When configuring web_extract or a similar content tool in your agent, set the character limit high enough to capture a full article. A limit of 15,000 characters covers most in-depth pieces. Tell the agent to extract claims in a structured format: claim text, source URL, publication date and a confidence flag. We use JSON lines for this, one claim per line, with the source URL as the key:
{"claim": "AI agent market projected to reach $47B by 2030", "source": "https://...", "date": "2026-03-15", "confidence": "single"}
{"claim": "AI agent market projected $42-51B by 2030", "source": "https://...", "date": "2026-04-02", "confidence": "single"}
When the agent finds two claims about the same topic, it compares them. If they agree within a reasonable margin, confidence goes up. If they conflict, the agent notes the disagreement and searches for a third source to break the tie.
Step 4: Evaluate confidence
The agent scores each claim on a four-tier scale. If the answer relies on a single source, it is low confidence. If two independent sources agree, medium. Three or more independent sources with consistent data is high confidence. And if sources disagree, the agent presents both sides with source URLs rather than guessing which one is right.
This is the scale our own pipeline uses before a single source gets cited:
| Confidence | What it takes |
|---|---|
| High | Three or more independent, credible sources agree on the same claim |
| Medium | Two independent sources agree, or one high-quality source with supporting context |
| Low | One source only, or sources that partially agree with caveats |
| Conflicting | Sources disagree, so the agent presents both sides with their source URLs |
This scale lives in the agent's skill file as a scoring rubric. When the agent returns a research result, it includes a confidence badge for each claim. The user can click through to the sources and verify for themselves. Teams trust the output far more when they can see which claims are solid and which are tentative, because the confidence scale makes the agent's uncertainty visible rather than hidden.
Step 5: Iterate until you have enough
The agent does not stop after one pass. If the combined confidence across all sub-questions does not meet the configured threshold, the agent generates more specific search queries and runs another loop. This is the part most people skip when they build their first deep research agent. They set up one search, one read, one answer. The loop is what separates a competent deep research agent from a basic search tool.
Most research questions need 3 to 5 iterations before hitting a high-confidence answer. Simple factual questions resolve in 2 iterations. Complex market-sizing questions can take 6 or 7. Set a hard ceiling of 10 iterations per question so the agent does not loop forever on unanswerable queries. The iteration count and confidence threshold go in the same skill file as the decomposition rules.
Why deep research matters more in 2026 than it did in 2025
There are two reasons. First, the volume of AI-generated content on the web has made single-source answers less reliable. When a search engine returns ten results, some of them may be generated by another AI, which means they repeat the same incorrect claims with different wording. Multiple-source verification catches this. If five sources all say the same thing and they are all derived from the same original flawed source, they only count as one independent claim.
Second, the cost of inference has dropped enough that running a 10-iteration research loop costs roughly the same as a single long chat response. In early 2025, running 15 iterations with 50 source reads would have cost several dollars per query. By mid-2026, the same loop runs for cents. The economics now favour thoroughness over speed.
Setting it up in your tools
Here is exactly how to set up deep research in the three most common agent platforms in mid-2026.
In Hermes
Deep research is configured through the MCP tool server that connects to a search API like SearXNG or Firecrawl. The agent receives a skill file that defines the loop structure, sub-question decomposition rules and confidence scoring criteria. Create ~/.hermes/skills/deep-research.md with the loop instructions, the four-tier confidence scale and sub-question decomposition rules. The agent loads this at session start and applies it to every research request. Configure the search tools in ~/.hermes/config.yaml under the MCP section, pointing to your chosen search API endpoints. For a complete breakdown of how skills and tools work together in an agent stack, see our guide on the 10-layer AI agent stack.
In Claude Code
You can wire the same pattern through a custom tool definition in your CLAUDE.md. Define a deep_research tool that the agent can call, which runs a search, reads the results, evaluates confidence and returns a structured answer. The loop logic lives in a bash script or Python module that Claude Code invokes. The key is defining the iteration count and confidence threshold in the tool description so the agent knows when to stop searching and start answering. We typically set a minimum confidence of "high" for financial or legal research and "medium" for general business intelligence.
Using OpenAI Deep Research
OpenAI's built-in Deep Research mode handles the loop for you. It runs 5 to 15 iterations automatically, searches 30 to 100 sources and produces a cited synthesis. The limitation is you cannot customise the confidence scoring or sub-question decomposition logic. For quick answers this is fine. For research that needs to meet an internal audit standard, the Hermes or Claude Code approach gives you more control over the verification process.
What we have learned from building these agents
Three things consistently surprise teams building their first deep research agent.
One: The loop structure matters more than the search engine quality. Teams often spend weeks tuning their search query syntax when what they actually need is a loop that searches moderately well five times instead of perfectly once. A loop with three mediocre searches and a simple confidence check beats a single perfect search every time.
Two: Source diversity is more important than source quantity. Twenty sources all from the same type of publication (say, tech blogs) give you one perspective. Three sources from different types (academic paper, industry report, government data) give you genuine triangulation. Configure your agent's search tool set with diversity in mind, not just volume.
Three: Teams trust the output more when they can see the uncertainty. A research agent that says "I am medium-confident about this figure, two sources agree but one is a vendor report" gets used more than one that confidently states the wrong number. The confidence scale is the feature that makes the agent useful in a professional setting.
For a practical walkthrough of automating multi-step research workflows with n8n templates, see our guide to n8n ad automation. And if you want to go deeper on how the full agent stack fits together, from identity to memory to safety, read the 10-layer AI agent stack.
Frequently asked questions
How do deep research agents verify their answers?
A deep research agent verifies its answers by corroborating every claim across independent sources before it uses that claim. It reads each source in full rather than the search snippet, records each claim with its source URL and publication date, and scores it on a four-tier confidence scale: high needs three or more independent sources that agree, medium needs two, low means a single source, and conflicting means the sources disagree and the agent presents both sides. It keeps running search-and-read iterations until the combined confidence clears a set threshold, and every claim in the final answer links back to where it came from so a person can check it.
What is a deep research AI agent?
A deep research AI agent is an automated system that searches multiple sources, reads full content, extracts specific claims, evaluates confidence levels across at least three independent sources, and iterates until it finds enough corroborated evidence. Unlike a standard chatbot, it shows its work with inline citations. By mid-2026, deep research mode had become a standard feature in the major AI assistants, from OpenAI and Google to Perplexity, because single-shot answers are not reliable enough for professional use.
How do you build a deep research agent?
Build a research loop that breaks the question into sub-questions, searches at least three independent sources (general web, news and a specialised database), reads the full content of each result, extracts claims with source URLs, scores each claim on a four-tier confidence scale (high, medium, low, conflicting), and iterates until enough corroborated evidence exists. In Hermes, configure this through an MCP tool server connected to a search API and a skill file at ~/.hermes/skills/deep-research.md. In Claude Code, wire it through a custom tool definition in your CLAUDE.md.
What separates deep research from regular AI search?
Regular AI search runs one query and summarises the top result. Deep research runs a multi-iteration loop: search, read, evaluate, refine, search again. The agent continues until it finds enough corroborating evidence, and every claim has a backlink to its source. A 2025 Stanford University study of purpose-built AI legal research tools found they still hallucinate between 17 and 33 per cent of the time, which is why a verification loop against multiple independent sources matters. OpenAI's Deep Research mode runs 5 to 15 iterations and reads 30 to 100 sources per question.
Which tools can you use to build deep research agents in 2026?
Hermes supports deep research via MCP tool servers connected to search APIs like SearXNG or Firecrawl, with loop logic in a skill file at ~/.hermes/skills/deep-research.md. Claude Code supports custom tool definitions for search-and-iterate loops defined in CLAUDE.md. OpenAI's Deep Research mode offers built-in multi-step research with 5 to 15 iterations and 30 to 100 sources. Perplexity's Pro Search queries multiple indices simultaneously. The essential infrastructure is a search API, a content extraction tool and a loop controller that manages iteration logic and confidence scoring.
How much time can a deep research agent save?
A thorough competitive research brief that typically takes a human analyst 3 to 8 hours can be completed by a deep research agent in 15 to 30 minutes. In our own pipeline, the most common finding is about coverage. The agent routinely turns up sources a manual pass would have missed, because it searches more broadly.