How AI assistants pick vendors — and what that means for your company
Your buyers have changed how they build a shortlist. Before anyone fills in a contact form, a Head of Ops has often already asked an assistant “who does meeting-to-CRM automation for B2B teams in Southeast Asia?” and read the answer.
That has produced a wave of “AI SEO” packages. Most of them are hype. Here is the honest version for a B2B company that wants to be found and understood.
What Google actually says
Google is still where most B2B research starts. Google’s own guidance for generative AI features in Search is clear on a few points:
- The same fundamentals that help you rank in normal search apply to AI Overviews and AI Mode. There is no separate AI ranking system to game.
- There are no special technical requirements or magic files for appearing in Google’s AI features.
- Google Search does not use
llms.txtor similar files to rank or surface your pages.
So if a vendor tells you an llms.txt file will get you into Google’s AI answers, that is not how it works. What gets you into AI answers is the same thing that gets you into normal results: content that can be crawled, read, and trusted.
What genuinely helps
None of this is glamorous, and all of it works:
- Crawlable, fast pages. If a crawler can’t load your page quickly, nothing else matters.
- Specific, checkable content. Real answers to what your buyers ask, in text, with numbers you can stand behind.
- Clear structure. One topic per page, sensible headings, descriptive titles, internal links between related pages.
- Structured data that matches the page. Schema helps machines classify you — but only if it reflects what a human actually sees.
- Named, consistent facts. What you sell, who it’s for, what it costs, how to start. Assistants reconcile your site against other sources; contradictions get you dropped from the shortlist.
That last point does most of the work in B2B. An assistant recommending a vendor is making a small reputational bet. Ambiguity loses.
Where machine-readability actually fits
Not everything is Google. Assistants — ChatGPT, Claude, Perplexity and the rest — read the open web directly, and increasingly connect to structured endpoints rather than scraping pages.
This is the part worth taking seriously, and it is not an SEO tactic. In December 2025 the Model Context Protocol was donated to the Linux Foundation’s Agentic AI Foundation, backed by AWS, Google, Microsoft, OpenAI and Cloudflare. By mid-2026, a large majority of enterprise AI teams were running MCP-backed agents in production. The direction is settled: systems are becoming things agents talk to, not just pages humans read.
For your website that means a modest, cheap hedge — clean HTML, accurate schema, and optionally a machine-readable summary of your business facts — so that when an assistant does read you, it gets you right.
For your internal systems, it means something bigger, and it is the reason we care about this at all. An automation only a human can trigger is a dead end. One that exposes a clean API, a CLI, or an MCP endpoint is something every agent you adopt later can reach. Build the plumbing that way and you keep your options open.
That is the honest framing: machine-readability is a sensible layer on solid fundamentals, not a shortcut around them — and it matters far more inside your operations than on your marketing site.
The bottom line
- A clear, fast site with a real page for each thing you do.
- Content that answers what buyers actually ask, with facts that don’t contradict each other.
- Structured data that matches the page.
- Then — as a hedge — machine-readable content for the assistants.
- And in your operations, build automation that agents can reach.
Do those in order and you have a genuine chance of being found and correctly described. What you will not get from anyone honest is a guarantee of rankings or AI mentions — those depend on your content, your competition, and how the engines work. Be wary of anyone who promises otherwise.
If you want a read on how your systems look to an agent, send your site through the Agent Readiness Audit.
Official sources worth reading
Dealing with this in production? The AI Systems Diagnostic maps where AI pays in your operations — and where it doesn’t.
More guides
- Stop running your business on WhatsApp + ExcelYou run your business on WhatsApp and Excel, and it works because you hold it together. Here's what to fix yourself first — and when a system pays off.
- Enquiry ops: getting leads from first message into the CRM without losing themMost B2B teams lose leads between the first message and the CRM. Here's how to design the enquiry path, what to track before automating, and when a scoped build is actually worth it.
- Where AI actually pays in a Malaysian SMEWhere AI saves hours or wins customers in a 20–200-staff Malaysian company — and the popular ideas that quietly waste money. Honest, no hype.