Chapter 12

Gift lists differed for husband and wife

A gift stall. A robot stands behind a table of parcels. Two children hold up handmade cards. The robot has two speech bubbles, each with three toy cars, and only the green car is in both.

Imagine you ask a robot at a gift stall what to buy for a husband. Then you ask again and change one word: wife. The robot names things both times. Are they the same things?

Your customers can say who a thing is for, and one word like that may change the names they get back.

So I compared gift answers from three robots. The questions for husband and wife differed by one word, but the answers came from separate runs.

The short version. The husband and wife answers named many different brands. Robots also differ a lot in how many brands they name, and some answers name almost none. So ask your question with each word your customers use, on more than one robot.

What I did

I counted the saved Christmas and Valentine’s answers again. A brand counted when two reading robots both found it.

Few brands came up for both words

Three piles of parcels. The left pile has 200 white parcels, the middle pile has 63 green parcels, and the right pile has 112 white parcels.
63 of 375 brands came up for both husband and wife

Picture it. Each parcel is one brand named at Christmas. The 200 on the left came up for husband alone, and the 112 on the right for wife alone. The green pile came up for both, and those brands took about half of all the times a brand was named. The left pile is bigger partly because I asked with husband almost twice as often.

At Marta’s: like a shop with 375 products, where 63 sit on both the gifts-for-him and the gifts-for-her page.

The two sides had different numbers of answers, and names that appear once can land on one side by chance. Separate groups of husband answers shared far more names with each other than with wife answers. The answers came from separate runs on two dates, so this does not show how much of the difference the word itself caused.

One robot named far fewer brands than the other two

At Christmas, GPT named about 1 brand per answer on average, and most of its answers named none. Gemini and Grok averaged 13 to 18.

Picture it. Each parcel is one brand in an average answer, rounded. Top row husband, bottom row wife. Gemini averaged more for wife, yet half its answers named 17 or more on both sides. Grok named more for husband. GPT did too, by under a parcel, so its rows match. The side with more names differed by robot.

GeminiTwo rows of parcels: 14 on top, 18 below.14, 18
GrokTwo rows of parcels: 15 on top, 13 below.15, 13
GPTTwo rows, each with 1 parcel.1, 1

At Samira’s: like three staff choosing where a client’s ads go. One names about 1 TV channel or website per plan on average, and most of their plans name none. The other two average 13 to 18.

Some answers name almost no brands

Three rows of 50 tags. The top row is all green. The other two rows are all grey.
At Valentine’s, 100 of 150 answers had no agreed brand

Picture it. Each tag is one answer, in rows for Gemini, Grok and GPT, from the top. Green tags have a brand that both reading robots found.

At Giulia’s: like 150 clients asking three advisers which invoice software to buy. One adviser names a product in all 50 answers. The other two give long advice in all 100 answers, and two checkers find no product name.

Grok named many brands at Christmas and had no agreed brands here. The question and the way the answers were collected also changed, so the occasion alone cannot explain this.

How to check this for your business

Portrait of Marta, who runs an online shop.Marta KowalczykOnline shop, PolandMade-up example

Marta sells gifts online. They ask a robot about a gift for a husband, and the shop comes up. They change that word to wife, and the shop is gone.

Is that chance? One try cannot say. So they ask many times with each word, then compare the two sides.

Other owners can test their own pair of words. Daniel can try tenant and landlord. Kadri can try first home and family home.

Try this on Monday: ask with both words

Take a question from your sheet and add a word for a kind of customer, such as family or student. Ask it ten times on each of two robots, in a new chat each time. Then swap the word and do the same. For each robot, count how many answers name each company on each side. A name seen once may be chance.

This can show names that may depend on the word you use. Answers to the same word differ too, so ask again before you trust a gap.

References

Chapter 12. Gift lists differed for husband and wife

Data for the reanalysis. Dmitrij Żatuchin (2026). Supplementary Materials: Volume and Distribution as Separable Dimensions of Gender Bias in Large Language Model Brand Recommendation. Zenodo, version 1.0. 10.5281/zenodo.21904788.

What does AI say about your business?

Put the book into practice. Tell us your company and website to request an AI visibility audit. We’ll reply by email to agree what to check.

What the audit shows

  • Where your company appears in the answers, and next to whom
  • The pages AI cites in your market
  • What to publish first on your own site

Fields marked * are required.

Who is it for?

Your choice about cookies

We use cookies this site needs to work, and others only if you say yes. Choose what you allow, and change your answer whenever you like. Read our privacy policy