AI Chatbot for Manufacturers and Distributors: Specs, Quotes, Dealers

At eleven at night, a maintenance engineer wants to know whether your stainless ball valve holds 63 bar at 180 °C and how long 200 of them take. The answer is on page 112 of a 180-page PDF catalogue. A purchasing manager in Monterrey wants the nearest distributor, and a contractor wants a price on 2,000 fittings by Friday. Nobody in sales is at work.

This guide is for manufacturers, industrial distributors and wholesalers who want a website chatbot that answers from their own documents and passes real buying intent to sales. It covers buyer questions, catalogues and product data (with a splitting test), quote intake, what never to promise, live stock lookups, dealer routing, other languages and setup.

In a Gartner survey of 646 B2B buyers, run in August and September 2025 and published in March 2026, 67% said they prefer a rep-free experience and 45% had used AI during a recent purchase. Your website now carries more of the sale. PepoChat is our product; its behaviour below was checked in its code and documentation on 22 October 2026.

What does an AI chatbot do for a manufacturer or distributor?

An AI chatbot for a manufacturer or distributor does four jobs on your website: it answers technical and commercial questions from your catalogues and spec sheets, takes quote requests, points buyers to the right distributor, and hands everything else to a person.

An AI support agent is a chatbot that searches your own content before every reply and can take actions, such as filing a ticket, within limits you set. In this sector it works like a well-read inside sales assistant: it knows the catalogue, the terms of sale and the dealer map, and it knows that prices, stock and delivery promises belong to someone with system access. It is not a configurator, a quoting engine or an ERP; it sits in front of those and links to them.

What do B2B buyers ask a manufacturer's chatbot?

B2B buyers ask a manufacturer's chatbot about specifications, compatibility, certifications, documents, lead times, minimum order quantities, price breaks, where to buy and how to get a quote. Documents answer the first group; the rest need a live system or a person.

In the table, answer means reply from your content, act means collect details or call a system, and hand off means pass the conversation to a person.

Put simply, documents answer "what is it", live systems answer "what can I have and when", and people answer "what will you do for me".

Equivalents deserve their own rule: parts with matching dimensions can differ in material, pressure rating or tolerance, so the chatbot should only name equivalents from a cross-reference list you publish.

How do PDF catalogues, spec sheets and CSV files become chatbot answers?

PDF catalogues, spec sheets and CSV files become chatbot answers in three steps: the chatbot extracts their text, splits it into short passages, and searches those passages for every question. Together these files form the knowledge base, the private collection of text the chatbot searches before it replies. In PepoChat a passage is about 1,000 characters, and each search returns up to eight of them.

That method is retrieval-augmented generation (RAG), a language model paired with a search over your own text. PepoChat's search compares meaning rather than exact characters, so a bare part number such as BV-1013 gives it little to match; a passage that also says "two-piece stainless steel ball valve" gives it much more.

From PepoChat's code and documentation, checked 22 October 2026.

Plan limits. A 300-page catalogue at about 3,000 characters a page is roughly 0.9 MB of text, so the Free plan's 20 MB of knowledge-base text holds about twenty; Starter allows 200 MB and Growth 1 GB. You meet two other limits first: 25 MB per file, because product photos inflate PDFs, and the Free plan's five knowledge sources, since each uploaded file counts as one. Export text-only PDFs, and put a datasheet library into one sitemap import or one Google Drive connection, each a single source.

Scanned catalogues. A scanned file goes to a vision model as a whole, so after uploading a long scan, ask about something on its last pages; if the answer is missing, split the file and upload the parts.

Freshness. Website pages are re-fetched weekly and Sync website pages refreshes them on demand, Google Drive files sync daily, and an uploaded file is replaced by uploading one with the same name. That suits specifications and terms; it is far too slow for stock.

How should you format product data for a chatbot?

Format product data for a chatbot as one self-describing line per product: the part number, the product name and every attribute with its label and unit. A CSV header row reaches only the first passage, so most rows lose their column names.

We tested this on 22 October 2026 by running PepoChat's text splitter on the same 60 ball valves, nine attributes each, in four formats.

Tested with PepoChat's text splitter on 22 October 2026.

A row such as BV-1013,Ball valve 2-piece,316 stainless,1",63,180,CE; WRAS,25,4 means little without its header: 63 what, 180 what, 25 what? The same product as a labelled line answers on its own:

BV-1013 two-piece ball valve, 316 stainless steel, 1 in (DN25) port: maximum pressure 63 bar (914 psi), maximum temperature 180 °C (356 °F), certifications CE and WRAS, minimum order 25 units, standard lead time 4 weeks.

A spreadsheet can build these lines: add a formula column that joins each label to its value, and export only that column. In Google Drive, put that column on the first tab, because only the first sheet syncs.

Three more rules help. Put the product family and the words buyers use next to every part number. State both unit systems in the data, so the chatbot never converts. Keep one file per product line and replace it by uploading a file with the same name, so two versions never disagree.

Know the limit, too: retrieval is not a database query. Asked "which of your valves are rated above 100 bar?", the chatbot gets at most eight passages per search, about 48 products in this format, not your whole range. Send parametric questions to your site's product filter, or build a search action like the stock lookup below.

How should a chatbot take an RFQ?

A chatbot should take an RFQ by collecting the buyer's company, name, email, phone, part numbers, quantities, required date and delivery location in the conversation, reading the request back, and then alerting sales and filing it where your team works.

An RFQ (request for quotation) is a buyer's request for a price and delivery terms on specific parts and quantities. The chatbot does the intake, not the pricing, which saves sales the first email back asking what the buyer needs.

In PepoChat the flow uses settings that exist on every plan:

1. Ask for details at the right moment. Under Lead Capture, choose When the agent needs a follow-up: the agent answers spec questions straight away and asks once, politely, when a request such as a quote needs someone to get back to the buyer. Email is always collected; tick Name, Phone and Company. The lead generation guide compares the three timings.
2. Write the RFQ checklist into the instructions. In How to sound and behave, add: "For a quote request, collect part numbers or descriptions, quantity per part, required delivery date, delivery country and postcode, and any certificates needed with the shipment. Read the request back before you hand over."
3. Alert sales. Under When a new lead arrives, switch on a Slack card or a signed webhook, both available on every plan, or team emails on Starter and Growth. Each fires once per contact and links to the conversation.
4. Hand over the conversation. When the buyer asks for a person, the agent escalates and every workspace member gets an email; in the team inbox, Assigned to passes the RFQ to the right rep.

Two details trip teams up. Buyers can attach up to three images per message (PNG, JPG, WebP or GIF, 5 MB each), such as a photo of a nameplate, but not PDF drawings or CAD files, so ask for drawings by email; chat images are deleted a week after upload. Replies typed in the team inbox are not emailed to the buyer, so answer RFQs by email or phone.

HubSpot and Salesforce. PepoChat's HubSpot actions find a contact by email and create a ticket; its Salesforce actions find a contact, create a case and check a case's status. Neither creates contacts or leads. Use the lookup to spot existing customers, whose order problems belong in a ticket or case, and send new buyers to the CRM through the new-lead webhook and Zapier, Make or n8n, as the guide to sending chatbot leads to HubSpot or Salesforce shows. For a summary in Slack, change the Slack action's when to use text to "Use when the visitor has completed a quote request".

What should a manufacturer's chatbot never promise?

A manufacturer's chatbot should never promise a price, stock level, delivery date, compatibility or certification that it did not read from a current document or a live system. Catalogue text can be weeks old, and a chat reply reads like a commitment.

Lines like these go in the Never do this field of the agent instructions, up to 2,000 characters. When a buyer asks for one of them, the agent declines briefly and offers what it can do instead, such as taking an RFQ.

Certifications carry legal weight. The European Commission's CE marking page explains that the manufacturer carries out the conformity assessment, sets up the technical file and issues the EU declaration of conformity, and that CE marking does not mean the EU approved the product as safe. Keep current declarations and certificates in the knowledge base, link them in answers, and delete expired ones the day they lapse. This is general information, not legal advice.

Delivery terms follow the same logic. The ICC's Incoterms 2020 rules are eleven three-letter trade terms setting out the tasks, costs and risks of delivering goods from seller to buyer. The chatbot can name the Incoterms on your terms page; freight and duties for a shipment come from sales.

Can a chatbot check live stock and lead times in your ERP?

Yes, a chatbot can check live stock and lead times when you expose a read-only HTTPS endpoint from your ERP or inventory system. In PepoChat that is a custom action: an API call with up to 10 inputs and an allow-list of the response fields the agent may read.

Name:           Check stock and lead time
When to use:    Use when a buyer asks whether a part is in stock or how soon it ships.
Request:        GET https://erp-api.example.com/v1/stock?sku={{part_number}}
Header:         Authorization: Bearer <read-only token>
Input:          part_number, the part number exactly as the buyer typed it
Allowed fields: item.sku, item.stock_band, item.lead_time_weeks

Put a thin endpoint in front of the ERP rather than exposing the ERP itself. PepoChat's calls are HTTPS only, time out after 10 seconds, reject responses over 2 MB and pass at most 4,000 characters to the agent, so a small, fast response works best. The allow-list keeps cost prices and margins out of the agent's view even if the endpoint returns them.

Then decide what a stranger may learn. Integration lookups are not gated by email verification, so anyone on your website can ask about any part. Stock bands such as "over 500" and standard lead times are usually safe; customer-specific prices and order history keyed on a guessable customer number are not. Action calls run at least once, and a rare retry can repeat a write, so keep orders and payments out of the chatbot. The guide to agentic AI in customer service covers these guardrails, and the custom actions docs list every field.

A workspace holds up to 20 actions; a new custom action starts switched off, so test it before buyers see it.

How does a chatbot route buyers to the right dealer or distributor?

A chatbot routes buyers to the right dealer or distributor by matching their location against your dealer list or a locator API, then giving that partner's contact details, while your own team keeps house accounts and export enquiries.

For a few dozen partners, a dealer page is enough if each dealer is one line naming its territory in words, such as "covers Ohio, Indiana and Kentucky", with its phone and email. For hundreds, build a custom action that takes a country and postcode and returns the nearest three. Then add a channel rule to the instructions: "If a buyer is in a distributor's territory, give the distributor's details and do not quote prices." That keeps the chatbot from undercutting the partners who stock your products.

PepoChat's hosted chat page is a standalone link to the same agent that you can print as a QR code on catalogues, trade-show stands or equipment nameplates. If you use an allowed-domains list, switch on Allow the hosted chat page first.

How does a chatbot answer buyers in other languages?

A chatbot answers buyers in other languages by writing each reply in the buyer's language from your existing documents, so a Spanish question about an English datasheet gets a Spanish answer. PepoChat's widget interface also follows the browser in seven languages.

Engineering data has its own traps. Much of Europe uses the comma as a decimal marker, so "1,500 psi" can read as one and a half to a German engineer. NIST's guide to writing SI units uses a dot for the decimal point, spaces rather than commas to group digits, and a space between a number and its unit, as in "37 °C". Write "1500 psi" or "1 500 psi" in data the chatbot will quote, keep part numbers and standard names exactly as published, and add "Never convert units or currencies" to your rules.

Decide before launch who answers an RFQ written in Portuguese or Polish. The multilingual support guide covers handoffs when nobody on the team speaks the buyer's language.

How do you set up a chatbot for a manufacturing or distribution website?

Setting up a chatbot for a manufacturing or distribution website takes about a day in PepoChat, in eight steps from questions to testing.

1. List the questions. Ask sales and customer service for the 30 questions they answer every week, and mark each answer, act or hand off.
2. Import your website. Use a sitemap import with include filters for your product, support and datasheet paths, and list the datasheet PDFs in the sitemap so they arrive as part of the same source, within its 200 addresses.
3. Add catalogues and product data. Upload text PDFs under 25 MB each, and product data as one labelled line per product.
4. Write the instructions. Add the RFQ checklist to How to sound and behave and the six lines from the table above to Never do this.
5. Switch on lead capture and alerts. Choose When the agent needs a follow-up, tick name, phone and company, and turn on a Slack card or webhook.
6. Connect systems. Add the HubSpot or Salesforce actions, and custom actions for stock or a dealer locator once your endpoint exists.
7. Set team availability. Choose My team answers during business hours with your time zone and an away message such as "Our sales team replies to quote requests within one business day."
8. Install and test. Paste the script tag into your site template, then ask every question from your list in the Test the agent drawer. Test chats are free and stay out of Analytics and Contacts, but integration actions run for real.

After launch, count RFQs captured, how many carry a phone number and company, how fast sales replies, and which topics the agent could not answer. PepoChat's Analytics shows only visitor votes and topics, so count RFQs from the Contacts and conversation exports.

What to do next

• Mark the question table against your own catalogue, then convert one product line into labelled lines and test twenty real questions on it.
• Follow how to train an AI chatbot on your own website and PDFs for each import step, from sitemap filters to replacing a document.
• Decide where RFQs land, CRM lead, ticket or Slack card, before you switch on lead capture.
• Create a free PepoChat workspace: every feature apart from new-lead emails to the team is on the free plan, with 200 AI replies a month and five knowledge sources, and the pricing page lists each plan's limits.