How We Use AI Deep Research Tools to Decide What to Build Next

Every tool we’ve built started with the same question: is this worth building, and what would make it better than what already exists?

We’re not developers or market analysts, so we lean on AI for that groundwork. All four of the assistants we use (Claude, ChatGPT, Gemini and Perplexity) now have a “deep research” style mode. Here’s how we use them, what each one does, and the one habit that matters most.

What we ask them

We keep it to four questions for any new tool idea:

1. What tools already do this?
2. What features do they offer?
3. What do users complain about?
4. What could we offer that’s simpler or more useful?

Our first tool came from a problem we’d had ourselves. Research like this is how we check whether other people have it too, and whether anyone has already solved it well.

What each tool offers

These descriptions come from each company’s own documentation. Plans, limits and names change often, so check the current details before you rely on any of them.

Claude Research. According to Anthropic’s help page, Claude runs several searches that build on each other, decides what to look into next, and gives citations you can check. It’s available on paid plans, web search must be turned on, and research sessions can use up your usage limits faster than normal chats.

ChatGPT deep research. OpenAI’s help page says ChatGPT proposes a research plan that you can review and adjust before it starts. You can start it from the tools menu or by typing /Deepresearch, and you can point it at specific websites.

Perplexity Deep Research. When Perplexity launched it, the company described a mode that searches repeatedly, reads documents, and refines its plan as it learns. Reports can be exported or shared as a page.

Gemini Deep Research. Engadget reported that Google opened Deep Research to everyone to produce reports on complex topics. Later, The Register reported that it can use your Gmail, Drive and Chat as sources if you allow it, and noted mixed reviews, including concerns about how accurately sources are labeled.

How we actually use them

When we have a niche idea we want to build, we start with Perplexity and ask for a full market analysis. How much demand is there? Which tools already exist? Which features do users value? What frustrates them? We reach for Perplexity first because its deep research mode pulls from many sources and gives us very useful answers to those questions. We read the report and cross-check the sources it used.

Then we take the idea to Claude or ChatGPT and ask it to “validate and test” it. That helps us see the idea from different angles and shows us which direction to move in. Asking an AI to “validate” an idea tends to produce a polite yes. If you also ask it to argue against the idea, you get a more honest picture. Finally, we do our own homework and check that everything we’re relying on traces back to a real, traceable source, not a random open-ended page. That gives us the confidence to move forward.

Our process

1. Ask one narrow question. “What do people dislike about online QR code generators?” gets better answers than “research QR codes.”
2. Review the plan if the tool shows one. It’s a quick way to catch the tool going off in the wrong direction.
3. Test the idea in a second tool. Take the findings to another assistant and ask it to validate them. Where the two agree, you can be more confident. Where they disagree is where to look closer.
4. Open the sources. Don’t trust the summary. Click the links and check whether the page really says what the report claims.
5. Turn complaints into a short feature list. We look for the one problem people mention often that we could fix simply.

The step most people skip

An AI research report is a starting point, not a verdict. Reports can sound confident and still misread a source or cite one loosely. That’s why step 4 matters more than the others. If we can’t find a claim in the original source, it doesn’t go into our plans or into this blog.

What to try this week

Pick one small problem you’ve run into. Ask two of these tools what already exists and what people complain about, and open every source they cite. Note what you find, because that’s often where a good idea starts.

This post was AI-assisted and reviewed by the Veneris Labs team. Images generated with AI.