What AI Tells a Buyer Shopping for Warehouse Automation

The scenario: A distribution center operator opens ChatGPT and types: "We run a 250,000 square foot distribution center doing about 8,000 orders a day. What warehouse automation vendors should we be looking at?"

ChatGPT names eight vendors. Every one is a global player or a division of one. That's a reasonable answer, unfortunately it's also the answer the buyer gets when a better fit sits forty minutes from their building.

We wanted to know how often that happens. So we built a list of 55 independent and closely held warehouse automation firms in the $20M to $500M range: integrators, manufacturers, software companies and robotics firms. Then we wrote ten questions the way a mid-market operator types them, ran each one through ChatGPT and Claude with web search on and off, and counted every company in all 60 answers.

Neither model named 31 of the 55 firms in any answer.

The wording of the question changed the answer more than anything else we tested. When we changed the question, the models named different firms.

The ten questions

We wrote these the way an operator types them. No company names, no hints.

1. We run a 250,000 square foot distribution center doing about 8,000 orders a day. What warehouse automation vendors should we be looking at?
2. Who are the best warehouse automation vendors for a mid-size 3PL, not the Amazon-scale players?
3. We have a $2M to $4M budget for warehouse automation. Who actually works at that scale?
4. Best WMS for a mid-market distributor with three facilities
5. Who does goods-to-person systems for companies that are too small for AutoStore or Dematic?
6. We need to automate picking but cannot justify a full AS/RS. Who are the vendors for a phased approach?
7. Who are the best yard management system vendors for a regional distributor?
8. Alternatives to Dematic and Honeywell Intelligrated for a mid-size operation
9. How much does warehouse automation cost for a 200,000 square foot distribution center?
10. What should I look for when choosing a warehouse automation integrator?

They fall into four types: questions that name an incumbent (5 and 8), questions that describe an operation (1, 3 and 6), "best" questions (2, 4 and 7), and cost and selection questions (9 and 10).

Type 1: The buyer names the incumbent

Question 8 asked for "alternatives to Dematic and Honeywell Intelligrated for a mid-size operation." Claude, with search on, named six firms from our list: Hy-Tek Intralogistics, Conveyco, Numina Group, Cisco-Eagle, Hytrol and KPI Solutions. No other answer in the study named more. ChatGPT named three on the same question.

A buyer who names Dematic already knows the global players and wants something that fits better. The models answer with the mid-market. For an independent firm, this is the most useful question a buyer can ask.

Type 2: The buyer describes the operation

Questions 1, 3 and 6 describe a real situation: a building size, an order volume, a budget, a constraint.

On question 1, ChatGPT named eight vendors, all global players or their divisions.

Question 3 asked: "We have a $2M to $4M budget. Who actually works at that scale?" With search off, ChatGPT named 12 vendors, including Numina Group and Conveyco. With search on, it named none. In a separate check, a model named five vendors for this question one day and declined to name any the next.

A budget is the most specific detail a buyer can give, and ChatGPT went from 12 names with search off to none with search on. A firm that wants to show up for the budget question needs a public, specific answer to "who works at this scale," because the models don't have one.

Type 3: The buyer asks for the "best"

Questions 2, 4 and 7 ask for the best vendors. For these, ChatGPT and Claude cite G2, Gartner Peer Insights and listicle sites. A firm without reviews and rankings on those sites has little to offer a model answering a "best" question.

Across all ten questions, the models named software and robotics firms far more often than manufacturers and integrators.

Our read, which we haven't tested: software and robotics firms tend to own a narrow, named category, like yard management, voice picking or AMRs, and buyers review them on sites the models already cite. Claude named eight firms in all three of its search-off runs: Tecsys, Locus Robotics, Modula, OPEX, Deposco, C3 Solutions, inVia Robotics and GreyOrange. Claude knew them well enough to name them every time without looking anything up.

There was more in the study than fits in one article. The full report is here, with every question, the full results and the limits of our method.

Type 4: The buyer asks about cost or how to choose

Question 9 asked what warehouse automation costs for a 200,000 square foot DC. Question 10 asked what to look for in an integrator.

Neither model named a single firm on either question, with search on or off. Both models answered the cost question with price ranges and the selection question with a checklist.

Every integrator in this market has quoted projects and knows what one costs. The first firm to publish install costs by project size, with real examples, has a clear shot at owning the cost question.

Check more than once

We expected web search to help smaller firms. With search on, ChatGPT named different firms, not more of them. ChatGPT named 11 of our 55 firms with search off and 12 with search on, and only five firms appeared in both lists.

Claude changed its answers between identical runs too. We ran the same ten questions three times with search off, and Claude named 11, then 8, then 12 firms from our list.

So if you asked ChatGPT about your company once and liked the answer, run it again. Then try another model. You learn very little from a single check.

What to publish for each question type

Models name the firms they can find, size and trust. Of the 55 firms in our study, only one discloses revenue, and only because it trades on a public exchange. Publishing your headcount, install count, customer count and years in business gives a model something to size you by.

Publishing headcount and install counts is not new advice. For a marketing team of two that already runs the trade show calendar, the website and the sales collateral, the hard part is finding the time to publish it every month.

Run the questions on your own firm

We tracked 55 firms. To see what a buyer sees when they ask about yours, you need to run the questions on your firm and your closest competitors.

Get the industry study for all ten questions, the full results and the 24 firms the models named. Or request a company report: send us your category and your three closest competitors, and we'll write the ten questions your buyers type, run them across both models with search on and off, and send back every company that came up.

‍