Website chat should work out who is asking before it answers

Picture a paying customer whose upgrade didn't go through. They can't use the features they paid for. They open the chat box on the vendor's website, because it's the only thing on the page that looks like it might reach a person. The chat is a sales agent. It sees words like "upgrade" and "plan", answers as if it's talking to a buyer, offers a demo and logs a new lead.

Every website with a sales agent sees some version of this. The agent understood every word. It just never asked the question that mattered first. Who is this? My argument in this piece is simple. Website chat should work out who is asking before it decides what they asked. Partners, MSSPs, existing customers and new buyers can type the same words and need completely different answers.

How often do non-buyers show up in website chat?

Less often than you'd fear, which is why the problem lasts. We looked at internal Storylane data covering every conversation on the RepX agent on storylane.io between July 28 and October 23, 2026. There were 1,396 conversations in which the visitor typed at least one message. We searched the transcripts and summaries, then read each match by hand.

That is 45 conversations, about 3% of the total. These are lower bounds from keyword searches on one company's website, so treat the exact counts as a rough guide. Even if the true share is double, the main point holds. Almost everyone who talks to a website agent is a buyer.

So why care about the other 3%? Because they include your most frustrated customers and your future partners. If a sales agent treats them as leads, two things go wrong. The customer gets a sales pitch when they wanted help. And your lead data fills up with people who will never buy.

Why does a sales agent treat a customer like a lead?

The obvious answer is missing training. Teach the agent your support content and it will answer support questions. I don't think that's the whole story. Most sales agents already know where the support page is.

I have a theory for why it still goes wrong. A sales agent sorts questions by topic, and topics don't tell you who is asking. "How does seat pricing work?" is a pricing question from a buyer. From a customer who wants to remove a seat, it is a billing question. On storylane.io, 18 visitors asked exactly that in the period above (it is one of the suggested starter questions). The words alone can't tell you which of them were customers.

The Storylane Buyability Index shows the same overlap from the other side. Across 10,580 buyer questions generated for 529 B2B websites, 1,127 (10.7%) were about support. Examples include what happens after go-live, who to call at 2am, and whether you get a customer success manager. Those are buyers asking support questions before they buy. Websites answered them fully 44% of the time, against 59% for all questions. Two caveats apply. The questions are generated per company to reflect what a buyer would plausibly ask, not logged from real buyers. And the sites were chosen by the people who ran scans between September 8 and October 20, 2026. The full dataset is in the Storylane Buyability Index.

So "support" is not only a customer topic, and "pricing" is not only a buyer topic. Any routing rule built on topic will send some customers to sales and some buyers to the help desk. Identity has to come first, because it changes what the same sentence means.

What does each kind of visitor need?

The useful split isn't buyers versus everyone else. There are four kinds of visitors with four different jobs. Each one has a failure that looks fine in the chat log.

Most teams don't plan for the last row. A partner can be worth more than any single buyer that week, yet they look like a bad lead by every end-buyer rule. For example, a ten-person MSP fails a "200+ employees" rule even if it manages security for forty clients.

Why do partner-led and MSSP vendors need this most?

Because their website visitors are a different mix. When partners do most of the selling, the vendor's website becomes the place where customers, partners and partners' customers go to check something.

In that mix, the usual sales flow goes wrong in three ways.

• The MSSP gets treated as an end customer. They ask about multi-tenant management or reseller pricing and get a pitch sized for a single company. In the Storylane Buyability Index, partner-type questions (MSP programmes, reseller pricing, white-labelling) came up on 28 of 529 sites, 32 questions in all. They are rare, but not strange.
• A partner's customer gets sent to your direct sales team. This is where routing turns into channel conflict. If a visitor says their MSSP manages the product for them, booking them with a direct rep can undercut the partner who brought them in. Send them to the partner manager, who decides.
• The partner's own customers arrive needing support. The vendor may not even own that support relationship. "Contact your provider" is a fair answer, and an agent should be allowed to give it.

If you sell through MSSPs, write the partner route before you tune a single qualification question. We covered the demo side in how security vendors help MSSPs and resellers demo. Routing is the website half of the same problem.

Isn't a "Customers, log in here" link enough?

This is the strongest objection, and it's mostly right for most visitors. Customers who want to log in will find the login button. Keep it easy to see.

But the customers in our table didn't miss the login button. They had a problem the login page couldn't solve, like a charge they didn't recognise, an upgrade that didn't take, or an account they wanted deleted. The chat box was the only thing on the page that looked like a person. Visitors type into chat whatever they came with. Internal Storylane data from 1,332 RepX conversations across five B2B software companies (June 1 to September 9, 2026) found that 3.1% of 2,225 buyer questions were simply asking to talk to a human. A link in the menu doesn't catch that. The agent has to.

How do you route by who is asking?

Five changes, in the order I'd make them.

1. Decide who the person is first. Before the agent decides what the question is about, it should decide if the person is a buyer, a customer, a partner or none of these. Phrases like "my account", "our clients" and "we already pay" are stronger signals than any topic.
2. Give every non-buyer a clear exit. Support gets the support address or a ticket. Finance gets finance. Partners get the partner manager. Job seekers get the careers page. Write these as rules the agent can't override, the same way you'd write rules about pricing or competitors.
3. Never send a meeting link to someone with a problem. This is the cheapest fix on the list. A demo calendar offered to a customer with a strange charge reads as "we only answer people who might pay us more".
4. Keep customers and partners out of lead counts. If your qualified-lead number includes support cases, the number is wrong in the direction everyone likes, so nobody questions it. Score qualification only after you know who the person is.
5. Read the conversations that don't fit. Once a week, pull the chats your agent marked as off-topic, asked for a human, or mentioned an account. That is where routing gaps show up first.

For where a lead goes after it qualifies (round robin, territory, account ownership), our lead routing guide covers the rules. The point here is narrower. Those rules only work on people who are actually leads.

Where does a website agent like RepX fit?

I run marketing at a company that sells a website agent, so weigh this with that in mind. RepX is built for websites where most chat comes from buyers. It answers product questions, shows interactive demos during the conversation, qualifies against criteria you write in plain English, and sends qualified leads and conversation summaries to HubSpot, Salesforce and Slack.

Its Playbooks let each page have its own prompts, calls to action and calendar. So a pricing page can book a sales rep while a docs page books support. Guardrails let you set what it will and won't say. With any agent, including RepX, write the customer and partner exits before you write the qualification criteria.

If most of your chat is support, start with a bot built to resolve, not one built to sell. Website bots do one of three jobs: routing, resolving or selling. If you are weighing scheduling-led routing against conversational agents, Chili Piper Chat vs Qualified compares the two approaches.

The one thing to change on Monday

Open your chat transcripts from the last month. Search for "my account", "invoice", "partner" and "our clients". Count how many of those conversations ended with a meeting link or a lead record. If the answer is more than zero, your agent is answering what was asked without knowing who asked it. Work out who first. The what gets easier after that.

If you want to see how RepX playbooks, guardrails and CRM handoffs could map to the buyers, customers and partners on your own site, book a walkthrough of RepX.