Following the Leaves: What I Learned from Unexpected Ancestry Hints

Have you ever noticed an Ancestry hint for someone who isn’t actually mentioned in the record?

While reviewing hints for my great-great-grandparents, I came across three Civil War–related hints that puzzled me. These records were being suggested for Mary Foster Crawford, the wife of Washington Marion Crawford. At first, I thought they might include a reference to Mary — perhaps as a widow or family contact. However, when I examined the actual record images, her name was nowhere to be found.

Using the “quick compare” feature, I discovered two important things:

• The records were already correctly attached to her husband, Washington Marion Crawford.
• The records named Washington M. Crawford directly, yet the hint was also appearing for Mary Foster Crawford.

Curious about why these hints showed up for Mary, I asked ChatGPT to help me understand.

---

📄 How Ancestry’s hints get shared

When you attach a record to someone in your tree, that action can help generate hints for others who have the same person (or a similar profile) in their own trees. This process is called tree-based hinting and works like this:

1️⃣ Record-based matches
These are generated when facts in your tree (like names, dates, and places) match details in Ancestry’s record collections.

2️⃣ Tree-based matches (Member Trees system)
When a record is attached to an individual, that link is added to Ancestry’s hinting network. If someone else has a similar person in their tree — even if they haven’t added certain records yet — Ancestry might suggest that same record as a hint to them.

---

🤖 AI and “phantom” hints

I also asked ChatGPT whether AI could suggest a record for someone even if that person isn’t named in it. The answer? Yes.

Ancestry’s hinting system uses AI and machine learning algorithms that rely on a mix of factors:

• Names (exact or similar)
• Dates (birth, marriage, death, residence)
• Places
• Family relationships (spouse, parents, children)

Even if someone’s name doesn’t appear directly in a record, the system tries to “triangulate” possible matches. For example:

• A census might list only the head of household by full name, yet AI suggests attaching it to all household members.
• A marriage record may name just the couple, but hints might appear for their parents, inferred from tree relationships.
• A city directory may list only one family member, but hints appear for others presumed to be living at the same address.

---

⚠️ Why this can be risky

AI-based hinting relies on probabilities — it looks for patterns that might match, rather than verifying direct evidence. This can lead to hints that look plausible but are incorrect, especially when:

• Names are common
• There are multiple families in the same area
• Records contain partial or ambiguous information (like “Mrs. John Smith”)
• Indexing errors or variant spellings exist

---

💬 An example

If your tree includes “Mary Jones,” born in 1880 in Missouri, and there’s a record for a “John Jones” household in the 1900 census, AI might suggest that record for Mary. It infers she could be a child in that household — but this isn’t always correct, especially if there are many Jones families.

---

🔎 What you can do

✔️ Always examine the original record images and full transcriptions.
✔️ Check household context and relationships carefully.
✔️ Use “Ignore” or “Not a Match” on incorrect hints to improve your tree and reduce future confusion.
✔️ Add notes or citations to explain why you did (or did not) attach a record.

---

✅ The bottom line

Yes — Ancestry’s AI can suggest hints for a person even when they aren’t explicitly named in the record. These hints are designed to guide your research but should never replace careful evaluation.

In my case, it isn’t entirely clear whether these hints to Mary Foster Crawford came from tree-based hinting, AI inferences, or a combination of both. However, based on Ancestry’s Hint Assist documentation, it seems likely that these suggestions weren’t directly generated by Hint Assist AI, but rather by the broader hinting system relying on tree connections and inferred relationships.