Your sales reps are already asking ChatGPT about your products

Picture of Noam Naveh
Noam Naveh

CEO @Stylib

Sales reps at architectural product brands are already asking ChatGPT about their own products. Nobody approved it and nobody is checking the answers.

From the rep's seat, it is the obvious move

A designer on a senior living project wants three navy upholstery fabrics compared: double rubs, CAL 117 and NFPA 260, bleach cleanability, and which have an EPD that counts towards LEED.

The answers sit across five PDF tech sheets, a SharePoint folder and a product manager who is in meetings until Thursday. ChatGPT replies in ten seconds, with a neat table.

Nobody would call that lazy. The rep has a client waiting and a tool that answers straight away. The internal route takes three days.

ChatGPT has never seen your catalogue

The catch is what’s in that table. ChatGPT has never seen your catalogue. It has seen your website, and your website is only a sample of the catalogue.

It has some colourways, not all of them. It has the specs typed into the product page, not the ones buried in a PDF or behind a trade login. Whatever it can’t find, it fills in, and the table looks just as tidy either way.

Industrial distribution has started measuring this. Reshape Automation, an AI software company working in industrial distribution, put 100 product questions about 14 industrial manufacturers to leading AI models. Even the best configuration, with web search switched on, got 42 of 100 wrong. Across all 500 answers, Reshape estimated that over a third of the wrong ones stated a fact, figure or product code with no hedging at all.

That is the real problem. The table doesn’t show which cells were found and which were filled in.

In most industries it's an awkward email. In ours, it's a specification

Get a lead time wrong and someone sends an apology. Get a flammability rating wrong and it goes into a specification.

From what we see at Stylib, the hardest data to find is usually the data that matters most: test results, certifications, cleaning codes. Those are the fields that live in PDFs rather than on product pages. They are also the fields where a guess turns into a liability once it’s in a spec and the fabric is on a care home chair.

Management looks the other way

Hardly anyone bans it. Nobody approves it either. Management knows the reps use it, or half knows, and says nothing. There’s no policy, no training, and nobody checks what goes out to clients.

It’s easy to see why. Reps already have too many systems to learn, and this one they picked themselves. It fixes something the official tools don’t: it answers in plain language, straight away. Saying nothing keeps the reps happy and costs nothing up front.

But staying quiet doesn’t fix anything. The PDFs are still scattered and the product manager is still in meetings. The guesses keep going out under the company’s name, and nobody can tell which answers were checked and which were filled in.

The useful question is narrower than “AI or no AI”. It’s about the source. A model that answers from your full product data, PDFs and test reports included, and cites the document each figure came from, is a different tool from one that guesses using your homepage.

What you let it read

So the question isn’t whether your reps use AI. It’s what you let it read.

If you want reps, and the designers they work with, to get answers from your full product data rather than your website, see how Stylib’s AI sales assistant for your product catalogue works.

Sources:

Share the post:

Related posts

Article

Your sales reps are already asking ChatGPT about your products

Article

The technical data sheet is never finished

Article

You can date a manufacturer’s website by where its technical data lives

What challenges could Stylib help you solve?

Whether its improving operations, empowering sales consultants or enhancing your site, we’d love to hear about it.

This site is protected by reCAPTCHA and the Google
Privacy Policy and Terms of Service apply.

Thanks for reaching out.

We received your message.
Someone from our team will soon be in touch.