I tested Claude, ChatGPT, and Gemini on my research notes. Here’s who won.

Not all AI tools perform the same. I ran a head-to-head test on analyzing my research notes and one failed completely.

For years, I kept research logs because we were supposed to. I wasn’t sure I’d ever look back at them, but I kept writing things down—notes on records, search attempts, names that could matter but I hadn’t placed yet.

Recently I asked myself:

What if AI could extract real value from all of this?

Could it sift through all this information and show me what I’ve actually done? Could it help me see what’s still missing? Could it make connections and find evidence?

So I tested it.

The setup: one spreadsheet, three tools, one prompt

I took one of my actual research logs—a big spreadsheet with real notes I’ve compiled over time. (Yes, if you click that link you can see it.)

Then I gave it to three of the top AI tools available right now:

• Claude Sonnet (Anthropic)

• ChatGPT 5 Thinking mode (OpenAI)

• Gemini Flash 2.5 Pro (Google)

All three got the same prompt, asking for a summary of what the research log showed, what questions it raised, and what it suggested I do next.

Here’s the prompt so you can try it yourself:

Analyze my genealogy research log and provide a comprehensive summary organized as follows:

## Record Types Searched
List all record types that have been searched (census, vital records, deeds, probate, military, immigration, etc.) and note the frequency of each type.

## Individual Summaries
For each person researched, provide:
- Name and key dates (birth, death, marriage)
- Summary of findings from all record types
- Confirmed facts vs. unverified information
- Conflicting information that needs resolution

## Individual Timelines
Create a chronological timeline for each person showing:
- Life events with dates and sources
- Geographic movements
- Family relationships established

## Research Analysis
Provide:
- **Research Gaps**: What records haven't been searched yet for each person
- **Next Steps**: Prioritized list of records to search next
- **Source Quality**: Assessment of source reliability and completeness
- **Breakthrough Opportunities**: Relationships or time periods that need more focus
- **Geographic Focus Areas**: Locations that require deeper research
- **Brick Walls**: Persistent research challenges and potential strategies to overcome them

## Research Statistics
- Total number of individuals researched
- Date range of research activities
- Most productive research periods
- Record types yielding the most information

Format the analysis clearly with headers and bullet points for easy reference.

Yes, I am asking AI to do a lot here with this one prompt. And from my experience with these models, I thought they could handle it.

The results: faster isn’t always better, but sometimes it is

Here’s what happened when I gave each AI model that research log with the prompt:

ChatGPT 5 Thinking

Tried twice. Each attempt ran for 11-13 minutes.

Both times, it stalled out completely. I was so shocked and disappointed! I felt like my dependable research assistant had gotten a lobotomy. The previous reasoning model o3 didn’t struggle with requests like this.

Claude Sonnet

Took 7 minutes.

Here’s part of the analysis Claude provided:

Delivered a solid summary with insight into which research paths I had explored, what kinds of sources I used, and where the gaps were. Nothing particularly insightful or helpful to move forward, but I was genuinely impressed by the summary. Sonnet is the lower tier model for Claude, and I suspect Opus 4.1 would do as good a job or better.

Gemini Flash 2.5 Pro

Finished in under 2 minutes.

Here’s part of the analysis Gemini provided:

It exceeded Claude’s output in terms of quality, and it did it faster. I was particularly impressed by the mention of primary vs. secondary sources which I had not asked for explicitly. It also added the term “brick walls” which shows it has knowledge of current genealogy culture. The noting of conflicts, which are inevitable in research, was particularly helpful. I hadn’t used Gemini in about 2 months and am pleasantly surprise by these results!

What this means for genealogists using AI

Not all tools are created equal. And what worked six months ago might not be your best option today.

The speed and usefulness of Gemini in this task really surprised me. Claude continues to be one of the most reliable options for historical and research-based work, but its restrictions on memory across projects is frustrating for me. ChatGPT 5 Thinking mode is clearly less powerful than the previous reasoning model and I hope future updates bring the old function back.

The tool you choose really does matter. If you try something with one AI model and it doesn’t work, use another. It’s also helpful to be a part of a community using AI for genealogy so you can keep up with what’s going on in AI.

What else can AI do with a research log?

After this first test, I ran a second prompt on my research log. This prompt connected the entries in the log back to my original research question.

AI went through my research log line-by-line evaluation the information as evidence to answer to question. Fascinating to watch. I then got to compare its answers with what I thought my evidence was. Improved my proof 10x.

Prompts and steps like these are what I’ll be sharing with members of the Research Lab. We’ll compare results and ensure we are getting most out of AI together.

Want to keep up with what’s working now with AI?

Research Lab opens in September.

It’s where we test things like this together, so you can stop feeling behind and start using AI the way it actually works today, not last year, or even last month. But most importantly, you get to work on your genealogy research problems, and finally move forward in solving them.

Get on the waitlist to save 20% off the list price when it opens (plus get my full AI prompt library).

👉 Join the waitlist here

Happy researching!

—Denyse

P.S. Research Lab opens in September and will remain open for enrollment year-round, so you can start whenever it works best for you.

But (and there’s always a but)… waitlist members get the lowest price, early access, and the bonus AI prompt library. So join the waitlist to claim it!