A normal AI answer cites a handful of sources. A deep research report can cite dozens. When someone asks ChatGPT, Gemini or Perplexity to "research the UK market for heat pumps" or "compare payroll software for a 50-person company", the tool plans the job, runs many searches, reads a large number of pages and writes a structured report, with citations, that the user may then forward to a boss or a client.
These reports are a different kind of visibility. They are long, they are often shared, and they reward exactly the kind of content that rarely wins a featured snippet: original data, detailed documentation, methodology and reference material. For B2B brands, research publishers and anyone selling a considered purchase, being a cited source in a deep research report can matter more than being mentioned in a two-line answer.
This guide explains how the deep research modes in ChatGPT, Gemini and Perplexity gather sources, what that implies about which content gets included, and a practical process for becoming a source these reports keep citing.
Key Takeaways
- Deep research tools turn a question into a research plan, run many searches and read many pages, then write a long report with citations.
- Gemini says Deep Research can "automatically browse up to hundreds of websites"; Perplexity's launch post said its version "performs dozens of searches, reads hundreds of sources".
- ChatGPT deep research can be limited to, or prioritise, specific sites chosen by the user, and can also read uploaded files and connected apps.
- No vendor publishes how sources are selected. Based on how the tools describe their output, pages with specific facts, data and clear structure are the most usable raw material.
- Crawler access matters: blocking the relevant search or grounding crawlers can remove you from the pool these tools draw on.
How the three main deep research tools work
| Tool | How it gathers sources | Output | Notable controls |
|---|---|---|---|
| ChatGPT deep research | Proposes an editable research plan; reads the public web, uploaded files and, where enabled, connected apps such as Google Drive or SharePoint | Report with table of contents, sources section and activity history; download as Markdown, Word or PDF | Users can restrict research to chosen sites or prioritise them while still searching the web |
| Gemini Deep Research | Turns the prompt into a multi-point plan; browses up to hundreds of websites; can use Gmail, Drive and Chat with permission | Multi-page report in minutes; can become Audio Overviews or interactive content in Canvas | User reviews and adjusts the plan |
| Perplexity Deep Research | Performs dozens of searches and reads hundreds of sources (per its February 2025 launch post) | Full report, most runs under three minutes at launch | Selected from the mode menu; free with daily limits, more for Pro (at launch) |
Two features stand out for brands. First, OpenAI lets users point deep research at specific domains. If buyers in your industry routinely tell ChatGPT to prioritise certain trade publications or standards bodies, those sites (and the brands they mention) gain weight. Second, every ChatGPT deep research output includes citations or source links, which is where your visibility shows up.
“Deep research takes more time to read and analyze many sources”
— OpenAI Help Center, Deep research in ChatGPT
Why deep research cites different content from quick answers
A quick answer needs one good passage. A report needs to support many claims: market size, definitions, comparisons, methods, examples, risks. That pushes these tools toward sources that can carry a specific claim on their own. We have not seen a robust public study that ranks content types by deep-research inclusion, so treat the table below as informed practice rather than measured fact.
| Content type | Why it is useful to a research agent | Example |
|---|---|---|
| Original data and surveys | Gives a unique, citable number with a date and method | "2026 UK SME payroll survey: methodology and results" |
| Statistics pages | Many facts in one place, each with a source | A maintained page of industry statistics with dated sources |
| Product and technical documentation | Precise specs, limits and compatibility for comparisons | Public docs listing integrations, limits and pricing rules |
| Comparison and methodology pages | Shows criteria the report can reuse | "How we evaluate X: our 12 criteria" |
| Case studies with numbers | Real outcomes with context | Before/after metrics, timeframe and constraints |
| Glossaries and definitions | Clean definitions for the report's background section | An industry glossary with one-paragraph entries |
Our guides on statistics pages that earn AI citations, case studies for AI search and help centre docs for AI answers cover each format in depth. For original data, see how to create original research content.
A step-by-step process to get cited in deep research reports
- Run the reports yourself. Ask each tool the five research questions your buyers would ask. Save the reports and list every cited domain.
- Classify the sources. Mark each cited source as your site, a competitor, a publisher, a review site, a government or standards body, or a forum. This shows where the report's "evidence" comes from.
- Find the claim gaps. Note claims in the report that cite weak sources, or that have no source at all. Those are openings for content you can own.
- Publish the evidence. Create pages that answer those claims with dated data, clear methodology and plain-language summaries.
- Structure for extraction. Use descriptive headings, short paragraphs, HTML tables and a "last updated" date. See AI-friendly tables and lists.
- Get cited by the cited. Pitch your data to the publishers and associations already appearing in reports. Our guide on why AI cites other sites instead of yours explains this pattern.
- Check access. Make sure key pages are crawlable, not hidden behind scripts or logins, and that you have not blocked crawlers you need.
- Re-run quarterly. Repeat the same research prompts and track how often your domain is cited.
Crawler and access considerations
These tools only cite what they can reach. OpenAI says that for inclusion in ChatGPT search results, sites must allow OAI-SearchBot; OpenAI does not document a separate crawler for deep research, so treat search access as the baseline. On the Google side, Google's crawler documentation says the Google-Extended token controls whether crawled content may be used for training Gemini models and for grounding in Gemini Apps, while not affecting inclusion or ranking in Google Search. Because Deep Research runs in the Gemini app, blocking Google-Extended may limit how your content is used there. Our explainer on Google-Extended covers the trade-offs, and our guide to AI crawlers covers the wider list.
Paywalled and gated content is a special case. A research agent generally cannot read what sits behind a login, so if your best research is gated, publish a substantial public summary with the key numbers.
Common mistakes
- Gating all your research. If the agent cannot read it, it cannot cite it. Keep a public summary with the headline findings.
- Undated statistics. Reports need to say when a figure applies. Always show the date, sample and method.
- Burying numbers in PDFs or images. Put key data in HTML text and tables as well.
- Writing only top-of-funnel posts. Reports draw on specs, comparisons and methodology. Thin "what is" posts rarely carry a claim.
- Measuring only clicks. Readers of a report may never click your citation but still remember your brand. Track citations as well as traffic; see AI search KPIs.
Related Guides
- How Brands Show Up in the Gemini App
- Search Live: Voice and Camera Conversations in Google Search
- Google Web Guide: The AI-Organised Results Experiment Explained
Frequently Asked Questions
What is a deep research mode?
It is an agent-style feature in tools such as ChatGPT, Gemini and Perplexity that plans a research task, runs many searches, reads many sources and writes a long, cited report, rather than a quick answer.
How many sources does a deep research report use?
It varies. Google says Gemini Deep Research can browse up to hundreds of websites, and Perplexity's launch post said its tool reads hundreds of sources. OpenAI does not give a number.
Can users tell ChatGPT deep research which sites to use?
Yes. OpenAI's help centre says users can limit research to specific domains or prioritise them while still searching the wider web.
Does blocking Google-Extended affect Gemini Deep Research?
Google says Google-Extended controls use of content for Gemini training and for grounding in Gemini Apps, but not inclusion in Google Search. Since Deep Research runs in the Gemini app, blocking it may limit how your content is used there.
What content is most likely to be cited?
No vendor publishes selection criteria. In practice, pages that state specific, dated facts, such as original data, documentation, methodology and statistics, give a report the evidence it needs.
Conclusion
Deep research reports reward brands that publish evidence: data, documentation and clear methodology that an agent can lift into a long report. Audit which sources the reports cite today, fill the claim gaps, and keep your best material public and crawlable. Our content writing team can help you build research-grade pages that earn citations. Get a free quote to start.
