Big ML Models Instead of Commercial and Transactional Searches

Previously we explored the notion of big ML models like GPT-3 standing in place of search engines for some types of search queries. We looked at navigational and informational queries first, and are now going to look at commercial or transactional queries.

If you’re new to the series stop by at the Table of Contents.

Commercial queries

Commercial queries are those where we’re looking at buying a type of product and are looking to narrow down which one to buy. Right out of the gate, ML models will be at a disadvantage due to their age. They are often trained on data that could be one or two years old, so if you are looking at getting recent information it’s not a good choice.

But if you are looking for products that don’t change often that could still work. Let’s use cars as an example. Models don’t change that much from year to year and you’re likely to get useful information from GPT-3.

In the next prompt we set it up so that the output is in a spreadsheet-like format. Note that this idea is taken from the GPT-3 examples page. Let’s suppose we’re looking for a small SUV, and we want to know the gas mileage and turning radius.

A four-column spreadsheet of vehicle models in the “small SUV” category, showing the make, the model, the gas mileage, and turning radius

make | model | gas mileage | turning radius

 Which generates the following output:

Honda CR-V | 28 mpg | 36.1 ft

Mazda CX-5 | 26 mpg | 35.4 ft

Ford Escape | 23 mpg | 39.2 ft

Nissan Rogue | 26 mpg | 37.4 ft

 Now I don’t know how accurate that might be, but it does provide a starting point for doing research.

Here’s another one, that supposes we’re looking to source some plastic drum barrels and want to know what the standard sizes and dimensions are.

A threecolumn spreadsheet of common sizes of plastic drum barrels, listing the capacity in gallons, the diameter, and the height of the barrel


(gallons) | diameter | height (inches)

And the response is:

55 | 22 | 33

30 | 18 | 30

15 | 15 | 23


Transactional queries

Last up are transactional queries, which the language models are poorly suited to handle. The information they have will be stale, and at best will output lists of stores and prices from when the training data was scraped. It could still work given the right subject, but this is a case where search engines have it beat.

Let’s first set it up with a very generous prompt that it should be able to make a good completion for:

A threecolumn spreadsheet of stores and their websites, which have a location in Michigan and sell car batteries. Store | Sells car batteries | website

Which gets us:

AutoZone | Yes |

Advance Auto Parts | Yes |

O’Reilly Auto Parts | Yes |

Pep Boys | Yes |

The results are technically correct, but it’s just a list of the major auto parts suppliers in the area. It’s missing a few I can think of just off the top of my head.

Here is another one where the prompt is not so generous:

Buy children‘s swimming pool online. Online stores:

And the output:

– Walmart

– Amazon

– Target

– Toys “R” Us

Which is a little bit funny, since one of those stores doesn’t exist anymore. Definitely not great, this one is easily bested by search.

Replace, no. How about augment?

The premise of GPT-3 replacing Google is certainly exciting for some of us, but it doesn’t seem like it’s quite there yet for most of what someone would search for. It will probably get better in the future, but even now there is still hope. The models can be used as tools to help augment your search queries, or think of alternate keywords to use when looking for something.

For example, here is a prompt where we get some ideas for search queries to use when wanting to learn more about plastic drum barrels:

I‘m looking for detailed information on plastic drum barrels. Here are some Google searches with related keywords to try using for research:

With the output:

-Plastic drum barrels

-Barrels for storage

-Water storage barrels

-Drum storage

-Barrel storage

-Food grade barrels


And here is another, supposing that we’re looking to buy barrels for storing liquids but aren’t sure what the important considerations are. Prompt:

Here is a list of the most important things to consider when choosing barrels for storing liquids:

And output:

1. The type of liquid you are storing:certain liquids require specific types of barrels. For example, wine needs to be stored in an airtight barrel, while water can be stored in any type of barrel.

2. The amount of liquid you are storing:the size of the barrel will determine how much liquid you can store.

3. The length of time you need to store the liquid:some barrels are not meant for long-term storage.

This is good to know, and I can imagine this being helpful for people who have to make decisions about things they don’t have much knowledge of.

Stick with search engines, for now

It looks like big ML models won’t be replacing search engines any time soon, but they can be used to augment searches by helping with related keywords or terms to search for. We’ve been seeing some amazing improvements in the models over the past few years so even though it may not be right around the corner, it will get better and better and should be something to re-evaluate from time to time. Who knows, maybe one day it will be part of your search flow.

The series continues next week with more interesting prompts to help create ideas for new articles, products, and businesses.

< Table of Contents


  • CRMs/ERPs
  • Payment Gateways
  • Accounting Systems
  • Lead Managements Tools
  • Web Apps
  • Legacy Systems
  • E-Commerce Systems
  • Freight/Shipping Systems
  • Social Media
  • Email Services

Get In Touch