Who owns the website? Why AI agents stall between marketing and product

At our booth this October, one of the most common doubts about a website AI agent wasn't price or fear of AI. It was a simpler sentence: "I don't own the website." One visitor told us product owned the site, and wasn't sure which chat tool was running on it. Another described so many overlapping tools that nobody was clear who used what. A third, wondering aloud whom we should talk to next, named marketing leadership, revenue operations and product in the same breath. I think that sentence explains more stalled AI agent projects than any feature gap, and it has a structure worth naming.

I call it the ownership gap: a website AI agent draws on content from marketing, product truth from docs and support, routing rules from RevOps and the site itself from web ops, so it stalls whenever nobody owns the whole.

What is the ownership gap?

The ownership gap is the distance between the one team that buys a website AI agent and the four teams whose work the agent depends on. A chat widget could live inside marketing because it only asked for a name and an email. An agent answers questions, and the answers live all over the company.

Here are the four lanes an agent touches, and who usually owns each:

• Content: what the website says and how it says it. Usually marketing and product marketing.
• Product truth: what the product actually does, how setup works, what the integration syncs, where data is stored. Usually docs, support, solutions engineering and security.
• Routing: which visitor goes to which rep, which leads count, what happens to pricing questions. Usually RevOps.
• The site: where the agent runs, how it loads, consent banners, page speed. Usually web ops or whoever holds the CMS keys.

Each lane has an owner. The agent has none, unless someone decides it does.

Where does a website AI agent get its answers?

Not from the pages marketing controls. The Storylane Buyability Index scanned 529 B2B websites with 10,580 buyer questions between Sep 8 and Oct 20, 2026, and recorded which page supplied each full answer. A caveat first: the questions were generated for each company (what a buyer would plausibly ask), not logged from real buyers, and the sites were chosen by the people who ran scans, so this is not a random sample of B2B.

Of the 6,204 questions answered in full, the homepage supplied 113 (1.8%) and blog, help centre and docs together supplied 38%. I wrote about that pattern as answer distance. The ownership question sits one level down: how many different kinds of page, and so how many teams, does one website need to answer its own buyers?

The security row is the same pattern I described as the assurance gap: the questions a buyer most needs to be sure about are the ones marketing pages answer least.

"Marketing pages" here means the homepage, product, pricing, comparison and case study pages. A quarter of full answers came from pages the scan couldn't classify, so read the shares as a shape, not exact counts.

The shape is clear enough. On a typical site, a complete set of answers is spread across five kinds of page. An agent that answers buyers well is reading the work of several teams at once, and it is only as current as the least maintained of them.

Why do AI agent projects stall between marketing and product?

The obvious explanation is budget, or skepticism about AI. Both came up at the booth. But they don't fit the ownership doubt. The visitor who told us product owned the site was interested enough to agree to a follow-up. The blocker wasn't wanting the agent. It was that they couldn't say yes to something that would speak with product's voice, write into RevOps' CRM and run on web ops' site.

I have a theory for why this is new. A website has always been a federation of pages, each written by the team that had to answer that question most often. Support wrote the help centre for customers. Security wrote the trust page for procurement. Marketing wrote the homepage for a visitor who hadn't asked anything yet. The seams between those pages were invisible, because the buyer did the stitching. If the answer was in a help article, the buyer either found it or left, and no team was blamed for the trip.

An agent does the stitching itself, in one reply. When a buyer asks whether the product works with their CRM and stores data in the EU, the agent pulls from an integrations page, a docs article and a trust page, and answers in your brand's voice. That one reply now has three owners. If the docs are a release behind, the agent repeats them with full confidence, and nobody in marketing knew the docs had changed.

That is why a visitor evaluating agents asked us how good the answers can be if the website is out of date. It is the right question, and it isn't a question about the agent. It is a question about who keeps the sources true.

So the project stalls in a predictable place. Marketing can sign the contract. It can't promise that product will keep docs current, that RevOps will define which conversations become leads, or that web ops will approve a new script. Each of those teams has a reasonable veto and no reason to use its time on someone else's project.

Shouldn't IT or a committee own it?

This is the strongest objection, and it is half right. A cross-team agent does need sign-off from every lane, and security and legal should see it before it goes live.

But a committee approves; it doesn't maintain. The ownership gap shows up three months after launch, when pricing changes, a new integration ships and a docs page moves. Someone has to notice that the agent's answer is now wrong, and a committee that meets once a quarter will notice late.

Routing makes the same point. In the Buyability scan, 102 questions ended with the site sending the buyer to sales instead of answering, and 90 of those were pricing questions. That is a decision somebody made, usually in RevOps, about which questions are worth a sales conversation. An agent has to make that decision hundreds of times a week. If no one owns the rule, the agent either answers what sales wanted to hold back or holds back what the buyer needed to keep going. There is more on that choice in our piece on putting pricing on your website.

Who should own the website AI agent?

I think one person should own the agent end to end: whoever already answers for the website's conversion number. In most B2B teams that is the head of demand gen or the web lead inside marketing. Not because marketing knows the product best, but because that person feels it first when the agent gets an answer wrong and the meeting doesn't book.

Everyone else becomes a source owner, with a short written agreement about what they give the agent and how often they check it.

The owner's job is to read conversations every week and send each wrong or thin answer back to the lane it came from. That loop is the whole program. A website AI agent is the first tool that shows you, question by question, which team's page let the buyer down.

What can you do about it this week?

Before you shortlist a single vendor, fill in that table with real names. If you can't name the source owner for product truth or routing, you have found the reason your last chatbot project drifted, and the next one will drift the same way.

Then ask each source owner one question: "If the agent said something wrong about your area tomorrow, how would you find out?" The answers tell you which lane needs a review loop first. If you want the build steps after that, our step-by-step guide to implementing an AI agent covers them, and the AI to human handoff piece covers the routing lane in detail.

How RepX is built for a single owner

I work at Storylane, which sells a website AI agent, so weigh this part with that in mind. We built RepX for the team that owns the website's conversion number and has to draw on everyone else's knowledge.

• Content and product truth: RepX is trained on your website, docs, decks, call scripts and interactive demos, so docs and support material feed the agent directly instead of being rewritten by marketing.
• Guardrails: you set what RepX will and won't say on pricing, competitors and compliance, once, and it applies everywhere.
• Routing: Playbooks give each page its own prompts, CTAs and calendar (your pricing page can book an AE while your docs page books support), and qualified leads and conversation summaries flow to HubSpot, Salesforce and Slack.
• Review: you run evaluations before anything goes live, and conversation analytics show the topics buyers ask about most, which is the weekly loop the owner needs.

It is best for B2B teams where one person in marketing owns website pipeline and wants the agent to speak with the product team's accuracy. (Product facts from storylane.io/repx, checked Oct 30, 2026.)

The agent is the easy part. The hard part is deciding who owns the whole website, and an agent forces that decision whether you are ready or not. Make it on purpose: one owner, four source owners, one weekly loop.

If you want to see what that looks like on your own site, book a RepX demo and bring the person who owns your docs.