AI Pedigree Chart Failures

Have you seen the chalkboard pedigree charts people have been sharing in the Genealogy and Artificial Intelligence Facebook group?

I’ve been trying to create one of my own and, so far, it has been a major failure.

Some of my early attempts used the suggested “Steve Little” style prompt with my ahnentafel embedded. Here’s the version I started with:

Role

You are an expert Genealogical Visualization Designer.

Input Data (Ground Truth)

Use the following Ahnentafel list as the absolute source of truth.

CRITICAL: Do not hallucinate names. Do not omit dates.

Generation 1

1. Marcia Crawford

Generation 2

2. Eugene Crawford born 8 Dec 1927; married 9 June 1951; died 14 Sept 2006.

Generation 3

3. Leon Crawford: born 6 Feb 1894; married 24 Dec 1919 ; died 3 Oct 1976.
4. Winnie Currey: born 30 June 1903; died 11 Feb 1992.

Generation 4

5. Judson Crawford: born 15 April 1866; married 24 Dec 1890; died 19 Feb 1949.
6. Josie Hammond: born 9 Feb 1874; died 27 Sept 1954.
7. Hiram Currey: born 23 Oct 1866; married 13 May 1891; died 15 Sept 1943.
8. Winifred Hutchinson: born 6 May 1871; died 23 Sep 1913.

Task

Generate a Chalkboard Pedigree Chart visualizing the provided data.

Visual Constraints (The “Nano Banana” Protocol)

1. Legibility First: Every name and date from the Input Data must be strictly legible.
2. Layout: Use a standard, horizontal pedigree structure.
3. Medium: Realistic chalk on a slate-grey blackboard.

Color Methodology (The “Lawrence-Little” Scheme)

Apply this specific color coding to the text/borders of each node:

1. Luminance (Generation Depth) changes from fully saturated for oldest generation fading into soft pastels
2. Hue Mixing (Inheritance): The hue of a child node MUST be the visual equal mix* of their parents’ hues.

Output Goal

A diagram where the text is readable enough to be verified against the source list, using color to show genetic flow from vivid ancestors to pastel descendants.

Below is the first attempt by ChatGPT to create the chart.

According to Steve Little, one is supposed to chat and iterate. So I did. I pointed out the errors and eventually got ChatGPT to interpret the ahnentafel correctly—it could tell who were parents and who married whom. Unfortunately, the image output remained strange. Names were misspelled, dates were altered, and sometimes two men were connected as a couple.

Since I knew that Steve Little had created a Gemini “Gem” for this purpose, I tried that next.

Again, I fed it the ahnentafel, provided the pedigree chart, and carefully described the errors. For my final Gemini attempt, I used a very direct prompt:

Create a pedigree chart for Eugene Crawford show his ancestors in a horizontal chart starting with him on the left and branching to the right to show his parents, grandparents and great grandparents.

The result was closer, but still not right. The chart was vertical instead of horizontal. The first three generations were somewhat recognizable, but many names were misspelled and most of the dates were incorrect. The last two generations were completely wrong.

So I made one more try—this time with Google Notebook LM. I uploaded a correct pedigree chart and asked it to create an infographic. This attempt finally produced an accurate depiction of the ancestral lines, but the generation containing Eugene’s grandparents was missing the names entirely. Beautiful lines, but incomplete genealogy.

At this point, I have several attractive but flawed chalkboard charts, all of which you’ll see embedded in this post. They are great examples of why genealogists still have to double-check everything AI produces, especially when our family data is involved.

Hopefully, after attending the upcoming “Navigating the AI Frontier: A Free Symposium …”, I’ll understand more about how these tools work—and maybe I’ll finally coax one of them into drawing an accurate chalkboard pedigree chart for my Crawford family.