Use a research mode (Claude Research, ChatGPT deep research, Gemini Deep Research or Felo's Pro Search), write a short brief that says what to find, how to lay it out and "cite every claim", let it work for a few minutes, and then open the sources behind the numbers you will act on. The free lesson Research: a cited competitor price scan takes seven minutes and ends with you catching an out-of-date price the report missed.
A normal chat answer is fine for one quick fact. A research mode is different: it plans, runs many searches, reads the pages and writes a report with a citation on each claim. That saves the afternoon of reading. It does not replace your check, because a report can quote an old page or a second-hand article as if it were current.
What are the steps?
- Pick the mode. In Claude, click + and choose Research. In ChatGPT and Gemini, choose Deep research from the + menu (in Gemini it may sit under more tools). In Felo, turn on Pro.
- Write a four-part brief. Who you are, what to find, how to lay it out (a table with named columns) and what to compare. Add "Cite every price and claim" and "flag anything old or second-hand".
- Read the plan. ChatGPT and Gemini show a research plan first. Edit it if a step is wrong, then start.
- Let it run. Expect a few minutes. You can leave and come back.
- Open the sources for the numbers that matter. Check the line, the date and who published it.
- Ask a follow-up. Tell it what you found and ask for a corrected table or talking points.
- Export or copy the report with its citations so whoever reads it can check too.
The post on Claude research prompts has brief templates you can copy.
What does this look like in a real lesson?
In Research: a cited competitor price scan, a made-up landscaping company wants to know what nearby lawn care firms charge. You turn on Research and ask for a table of competitors and prices with a source link for every price, and to flag old or second-hand prices. Claude plans, searches, reads pages and writes a report. The report already warns that one price came from a review site. Then you open one competitor's source yourself, and the small print at the bottom shows the prices are from March 2024. The report never said so. You tell Claude, and it marks the old row and writes three talking points for customers.
The ChatGPT deep research lesson does the same for a software choice: a brief comparing three real time trackers against a made-up studio's needs, a five-step plan you approve, a report with a source link on every need, and a spot-check of one price on the vendor's own pricing page before you export the PDF. In the Gemini version you notice that some very exact survey numbers trace back to a vendor blog post that never says how many people were asked.
Which tool should I use?
- Felo is an AI search tool that puts a numbered source on every claim, and its Free plan includes a daily number of Pro searches. In Felo: Price a New Service with a Source for Every Number, a bookkeeping firm prices a fractional CFO package, checks a firm's rate against its page, and spots a price that came from a comparison article instead of the firm. Then it asks Felo where each price came from and exports a PDF.
- Claude Research suits a report you will keep working on in the same chat.
- ChatGPT deep research and Gemini Deep Research write long reports and show their plan first. Gemini can export to Google Docs.
- Perplexity is good for quick cited lookups between bigger research runs.
For side-by-side detail, see the best AI research tools and ChatGPT vs Claude vs Gemini for research reports. If your topic is competitors, which AI tool can research competitors and show its sources narrows it down.
What mistakes should I avoid?
- A vague brief. "Research our market" gets a vague report. Name the columns and the comparison.
- Skipping the plan. It is your one chance to fix the search before it runs.
- Trusting exact-looking numbers. Very precise figures can come from a tiny survey. Find who said it.
- Not checking dates. Prices and rules change. Old pages look current in a summary.
- Forwarding an unchecked report. Spot-check at least the numbers a decision depends on.
Key takeaways
- Use a research mode for anything bigger than one quick fact.
- Write a brief: who, what to find, layout, comparison, cite every claim.
- Read and approve the plan when the tool shows one.
- Open the sources behind the key numbers, and check dates and publishers.
- Keep the citations when you share the report.
Frequently asked questions
- How long does AI research take?
Usually a few minutes, sometimes longer for a big question. The lessons' runs took about five minutes in ChatGPT and Gemini and about two minutes in Felo.
- Can I use AI research for a client report?
Yes, as a first draft. Check the sources behind every number, rewrite the conclusions in your own words, and have a person sign off before it goes to the client.
- How do I get citations from ChatGPT in a normal chat?
Ask it to search the web and cite each fact. The page on how to make ChatGPT show where each fact came from has the steps.