What desk research is
Consultants and analysts often begin an engagement with desk research: reading about an industry, a competitor, a regulation or a process before talking to anyone. The aim is to get oriented quickly, find what is already known, and form questions worth asking. It is a good match for AI assistants because it involves lots of reading and writing, but it is also an area where subtle errors can reach a client report.
Useful jobs for an assistant
Getting oriented. Asking an assistant to explain unfamiliar terms or outline how an industry works can build a rough map in minutes. Treat this as a starting vocabulary, not as authoritative content.
Structuring the question. Assistants can help break a broad question into sub-questions, list possible hypotheses and suggest what data would test each one. This overlaps with brainstorming and idea generation, and the value is in generating options you then evaluate.
Summarizing documents you supply. If you have a report, transcript or set of articles, an assistant can condense them and pull out key points. Providing the text yourself is usually safer than asking the tool to recall facts, a topic covered in guidance on document summaries.
Drafting. Outlines, first versions of sections, interview guides and email drafts can be produced fast and then rewritten in your voice and with your judgment.
Organizing notes. Messy notes can be sorted into themes, tables or timelines, and meeting notes can be turned into action lists.
Reviewing your own work. Asking an assistant to find gaps, inconsistencies or unclear logic in a draft can be a useful second pair of eyes, though it will not catch everything.
Where the risks are
Invented details. Language models can produce statements that sound right and are not, including made-up sources, quotes, figures and names. This tendency is explained in the discussion of why tools make things up, and in the context of research it is the main reason for discipline. Never cite something you have not opened and read yourself.
Stale or missing knowledge. An assistant may not know recent events or may reflect outdated information. If the topic is moving fast, check dates on every source.
Shallow synthesis. A neat summary can hide disagreement among sources or flatten nuance. If a claim matters, go to the underlying material.
Confidentiality. Client information often falls under confidentiality terms. Before pasting any of it into a tool, find out where it is sent and stored, and check the engagement agreement and your firm’s policy. Giving a tool the context it needs without oversharing is a skill of its own.
Homogenized thinking. If everyone asks similar questions of similar tools, the answers can become similar too. Original insight usually comes from primary sources, fieldwork and the analyst’s own perspective.
A reliable workflow
- Define the question first. Write down what decision the research supports. A vague question produces vague output.
- Use the assistant to scope, not to conclude. Ask for angles, terms and possible sources to look up.
- Collect sources yourself. Gather reports, filings, statistics from recognized bodies, academic papers and interviews. Prefer primary sources over summaries of summaries.
- Feed material in and ask for extraction. Ask for specific points tied to specific passages so you can verify each one.
- Verify every number and quote. Open the source, find the passage, and confirm the figure, date, units and context. This is the heart of fact checking.
- Keep a source log. Record where each claim came from, with date and link, so a colleague can follow it.
- Write the conclusions yourself. The reasoning, the recommendations and the caveats should be yours, because you are accountable for them.
- Have someone else review. A colleague reading with fresh eyes will often catch what you missed.
Handling numbers
Be especially careful with calculations. Language tools can make arithmetic mistakes or mix up units and time periods. Do any calculation that appears in a deliverable in a spreadsheet or calculator, check it, and keep the working visible. If a market size or growth figure appears without a clear source, treat it as unverified until you find where it came from.
Telling clients
Whether and how to disclose AI use depends on the engagement terms and your firm’s policy. Some clients ask for it explicitly. Being open about the tools used and the checks applied tends to build trust, and the principle is discussed further under AI disclosure.
A balanced view
An assistant can shorten the slow early phase of research and help structure the work. It cannot replace source evaluation, interviews, domain experience or the professional judgment a client is paying for. The time saved depends on the task, and the time spent verifying should not be skipped to capture it.