ChatGPT Deep Research Prompts: Briefs That Produce Useful Evidence

ChatGPT's deep research mode runs a more thorough investigation than a standard chat reply: it follows multiple sources, checks claims across them and produces a longer report with citations. A vague question returns a general survey. A well-structured brief narrows the sources, names the specific question and tells the AI how to flag uncertain claims.

The templates below cover the most common research brief types. Each one is written to be filled in and reused. fenbs keeps tested briefs and the sources they found in a shared place, so repeat investigations start from a known baseline rather than a blank prompt.

What a deep research brief should include

A research brief has five parts. Leave any of them out and the report becomes either too broad or too shallow to be useful.

• The question: one specific question you want answered. Not a topic, a question.
• Required sources: the kinds of sources that count as good evidence for this question (peer-reviewed research, official agency data, vendor documentation, court records, etc.).
• Exclusions: what to ignore. Roundup articles, opinion pieces, sources older than a certain date, or specific sites you do not trust for this topic.
• Evidence check: how the AI should handle conflict between sources and what to do when a claim appears in only one place.
• Output format: how long, how structured and what to include for each claim (source name, publication date, the specific claim it supports).

Template: competitive or market research

Use this template when you want a factual picture of a market, a competitor's public position or an industry's current state. The evidence check is especially important here because market data becomes outdated quickly.

Question: [Specific question about the market, competitor or trend]

Required sources: Official company announcements, industry analyst reports, SEC filings or equivalent regulatory filings, and news coverage from outlets that cite primary documents.

Exclude: Opinion columns, roundup articles without primary citations, and any data older than [timeframe].

Evidence check: If two sources give different numbers for the same metric, report both with their dates and sources. Do not average them. Flag any claim that appears in only one source.

Output: A report of [length]. For each key finding, list the source, its publication date and a direct quote or link where available. End with a section listing the claims you could not verify from primary sources.

Template: legal or regulatory research

Use this template when you need to understand a regulatory requirement, a court decision's implications, or the current state of a law. This template does not replace legal counsel, but it produces a clearer starting point for a conversation with one.

Question: [Specific question about a regulation, statute or court decision]

Required sources: Primary legal sources (statute text, regulatory text from official bodies such as the FTC, SEC, FINRA or relevant agency, and published court opinions). Secondary sources: law review articles and ABA publications for interpretation.

Exclude: Legal news summaries without citations to the underlying text, blog posts and any source that paraphrases a rule without linking to it.

Evidence check: If you cannot find the primary text, say so clearly rather than summarizing a secondary source as if it were primary. Note the jurisdiction for every rule cited.

Output: A structured summary identifying the primary rule, its current status, the key obligations it creates and the open questions it leaves. Note any recent enforcement actions or rulemakings that may change the picture.

Template: technology or product research

Use this template when you want to understand a technology's current capabilities, a product's documented features or how different tools compare on a specific dimension. Vendor documentation is a primary source here; review articles are secondary and should be treated as opinions.

Question: [Specific question about a technology or product]

Required sources: Official vendor documentation, technical standards documents (IETF, W3C, NIST where relevant), and peer-reviewed technical papers. For capabilities: the vendor's own feature pages and release notes.

Exclude: Review articles, affiliate content, and any claim about a feature that is not supported by a link to the vendor's documentation. Do not infer a feature exists from a competitor's description of it.

Evidence check: For any claimed capability, note whether it is documented by the vendor or inferred from a third-party description. Flag the difference clearly.

Output: A factual summary covering [the specific dimensions you care about]. For each claim, note the source. End with the limitations: what you could not confirm from primary documentation.

For context on how ChatGPT compares to other research tools on longer tasks, the claude vs chatgpt guide covers the differences that matter for research work. For saving and reusing the prompts that worked, chatgpt prompts for work covers a broader prompt library approach.

How to evaluate what deep research returns

A deep research output is a useful starting point, not a finished product. Check each key claim against the cited source. Verify that the date on the source is correct. Look up any source you have not seen before; some AI systems invent plausible-sounding citations. The investment in checking is proportional to the stakes of the decision the research informs.

Pay attention to the "could not verify" section if you added that section to your prompt. Unverified claims are often the ones that most need checking: they may be novel, recently updated, outdated or simply wrong. A report that is honest about its gaps is more useful than one that papers over them with confident-sounding hedges. When a key claim cannot be verified, treat it as a lead, not a finding: it needs more investigation before you rely on it.

Cross-check the most important claims against a different source than the one cited. Two independent sources agreeing is more reassuring than one confident source, particularly for statistics or figures that will be cited in a decision. If you are using the research to support a decision, note in your records which claims you verified independently and which you accepted from the AI's citation.

For alternatives to ChatGPT for deep research tasks, the chatgpt alternatives guide covers the main options and their different strengths.

How to do this with fenbs

fenbs keeps your tested research briefs in a shared Skills library alongside the links and sources from previous investigations. Any connected AI app can load a brief by name before starting research.

1. Create a free account. Set up a project for the research topic or work area these briefs belong to.
2. Add each tested brief as a skill. Include the template text, a note on what kind of question it suits, and the evidence standards it enforces.
3. When a research run produces useful sources, add them to the project's Links and listings. Note the claim they support and the date you accessed them.
4. Connect your AI app over MCP at Integrations. The app can load any brief by name and read the previous sources before starting a new investigation.
5. After each run, record what the prompt missed or what you would add for next time as a lesson in the project. The prompt improves with each use.