{"id":1693,"date":"2026-07-27T08:09:23","date_gmt":"2026-07-27T08:09:23","guid":{"rendered":"https:\/\/maxaeo.ai\/blog\/deep-research-ai-citations\/"},"modified":"2026-07-27T08:09:23","modified_gmt":"2026-07-27T08:09:23","slug":"deep-research-ai-citations","status":"publish","type":"post","link":"https:\/\/maxaeo.ai\/blog\/deep-research-ai-citations\/","title":{"rendered":"Deep Research AI Citations: How Agents Build a 40-Source Report"},"content":{"rendered":"

Deep research AI citations follow different rules than the three or four links a chatbot drops into a quick answer. A multi-step agent breaks one question into a dozen or more sub-questions, runs each as its own search, and fills a report section by section. We tore down 60 of those reports and tagged every source. The pattern was consistent: the pages that win long reports are demonstrably not the pages that win single-turn chat.<\/strong><\/p>\n

This is the mechanical version of that story \u2014 what the agent does between "go" and the finished report, which slots exist inside the document, and which page on your site is eligible for each one.<\/p>\n

\"Diagram<\/figure>\n

What are deep research AI citations?<\/h2>\n

Deep research AI citations are the sources a multi-step research agent reads, keeps and links inside a long-form generated report.<\/strong> Unlike a chat answer that cites 3\u20138 pages, a deep research run decomposes one prompt into many sub-queries and cites a separate set of sources for each one, so a single report can carry 40 or more references.<\/p>\n

The distinction matters because eligibility changes. In a one-shot answer, you compete against every other page on the topic for a handful of slots. In a report, you compete against the handful of pages that answer one narrow sub-question \u2014 and there are twenty such contests inside every run.<\/p>\n

Which products count as "deep research"<\/h3>\n

Not every long answer is a deep research run. Four surfaces produce the report format described here, and they behave differently from their own chat modes:<\/p>\n\n\n\n\n\n\n\n\n
Surface<\/th>\nWhat triggers it<\/th>\nTypical output<\/th>\n<\/tr>\n<\/thead>\n
ChatGPT deep research<\/td>\nExplicit "Deep research" tool selection<\/td>\n1,500\u20134,000 word report, inline numbered citations<\/td>\n<\/tr>\n
Gemini Deep Research<\/td>\nDeep Research mode in the model picker<\/td>\nReport plus an editable research plan shown before execution<\/td>\n<\/tr>\n
Perplexity research mode<\/td>\n"Research" \/ "Labs" toggle<\/td>\nReport with a visible source list and step log<\/td>\n<\/tr>\n
Claude research<\/td>\nResearch toggle in chat<\/td>\nReport with inline links, agentic multi-search<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

Google's AI Mode sits between the two worlds \u2014 it fans out but returns a page-shaped answer, not a report. We cover that split separately in AI Mode vs AI Overviews<\/a>.<\/p>\n

How we tested: 20 prompts, three agents, 2,340 tagged citations<\/h2>\n

The maxaeo research team ran 20 B2B SaaS buying-intent prompts through ChatGPT deep research, Gemini Deep Research and Perplexity's research mode in May\u2013June 2026 \u2014 60 reports total. Prompts covered categories like data observability, HRIS, API gateways and customer data platforms, each phrased the way a real buyer phrases it ("we're a 200-person team, which X should we shortlist and what are the tradeoffs?").<\/p>\n

For every report we saved the visible plan, the full output, and each citation. Two analysts independently tagged all 2,340 surviving citations by report section<\/strong>, page type<\/strong> and the sub-question the citation answered<\/strong>, then reconciled disagreements. We dropped 118 references that resolved to dead links or duplicate URLs.<\/p>\n

Limits worth stating.<\/strong> One vertical (B2B SaaS), one language (English), one two-month window, and deep research output is non-deterministic \u2014 the same prompt re-run a week later returned a 62% overlapping source list on our spot checks. Treat the percentages as the shape of the mechanism, not as constants.<\/p>\n

Headline numbers from the teardown:<\/strong><\/p>\n\n\n\n\n\n\n\n\n\n\n
Metric<\/th>\nMedian<\/th>\nRange<\/th>\n<\/tr>\n<\/thead>\n
Unique URLs cited per report<\/td>\n41<\/td>\n18\u201396<\/td>\n<\/tr>\n
Unique domains per report<\/td>\n26<\/td>\n11\u201354<\/td>\n<\/tr>\n
Visible sub-queries in the plan<\/td>\n14<\/td>\n6\u201331<\/td>\n<\/tr>\n
URLs cited per sub-query<\/td>\n3<\/td>\n1\u20139<\/td>\n<\/tr>\n
Run time to finished report<\/td>\n7m 12s<\/td>\n2m\u201324m<\/td>\n<\/tr>\n
Length of the actual quoted passage<\/td>\n54 words<\/td>\n11\u2013210<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

That last row is the one most teams underestimate. The median cited page ran about 1,900 words, but the agent lifted a 54-word passage from it. You are not getting cited for an article. You are getting cited for a paragraph.<\/strong><\/p>\n

How a multi-step research agent assembles a report<\/h2>\n

The assembly sequence was visible in the plan output of all three agents, and it ran in the same order every time:<\/p>\n

    \n
  1. Decompose.<\/strong> The agent rewrites the prompt as a research brief and splits it into sub-questions \u2014 definition, market landscape, evidence, mechanics, objections, recency. OpenAI describes this planning-then-executing loop in its introduction to deep research<\/a>, and Google documents the same plan-execute-synthesize structure for Gemini Deep Research<\/a>.<\/li>\n
  2. Fan out.<\/strong> Each sub-question becomes one or more live searches. This is the same hidden expansion described in our breakdown of how one prompt turns into dozens of hidden searches<\/a>, except the agent keeps the branches instead of collapsing them.<\/li>\n
  3. Open and skim.<\/strong> The agent fetches candidates and keeps whatever it can parse into a short quotable claim. In our reproduction of the fetched URLs, 12% returned a consent wall, login gate or empty JS shell \u2014 pages that were effectively invisible regardless of how good the content was.<\/li>\n
  4. Fill slots.<\/strong> Each retained passage gets assigned to a section of the draft. A page can fill one slot or several.<\/li>\n
  5. Reconcile and cut.<\/strong> Where sources disagree, the agent either picks the better-evidenced one or reports the conflict. Sources that duplicate an existing claim get dropped without a citation.<\/li>\n<\/ol>\n

    Step 5 is where most brand pages die. They were read, they matched, and they added nothing the agent didn't already have. In our sample, redundancy was the single largest silent loss: of the pages an agent fetched and could parse, roughly two-thirds never appeared in the final citation list, and the overwhelming majority of those were topically correct but duplicative.<\/p>\n

    What the agent can and cannot fetch<\/h3>\n

    Retrieval is the gate before quality ever gets evaluated. Three failure modes accounted for nearly all of the 12% dead-on-arrival rate:<\/p>\n