The Agentic Web, How AI Agents Will Shop, Research, and B...
Active Document
The agentic web is here — AI agents browsing, researching, booking, and purchasing on behalf of users. What this means for Edmonton businesses and how to prepare.
Loading document...
Something specific is happening in 2026 that most Edmonton business owners haven't absorbed yet. The AI interfaces their customers use — ChatGPT, Claude, Gemini, Perplexity — are evolving from answer engines into action engines. Instead of telling the user "here's a list of Edmonton plumbers," they're starting to call the plumber, book the appointment, and update the calendar on the user's behalf.
This is the agentic web. And it changes what "being online" means for an Edmonton business.
Traditional web: Customer searches → sees results → clicks a link → visits your site → converts.
Agentic web: Customer asks AI → AI agent browses, compares, and acts on their behalf → your business either participates or doesn't.
The agent never visits your marketing site the way a human does. It reads structured data. It calls your booking API. It requests a price list. It makes a booking. If your business can be understood and interacted with by an agent, you get the conversion. If not, you don't.
Browser-based agent that can shop, fill forms, and complete multi-step tasks on behalf of a user. Released in limited beta in 2025, broader availability through 2026.
Real use: user asks "find me a dentist in Edmonton that takes Sun Life and has Saturday availability, and book me in." Operator opens browsers, searches, reads clinic sites, checks insurance lists, looks at availability, books.
Similar agent — Claude can operate a computer, browse, fill forms, act. Stronger at structured reasoning tasks, weaker at some visual navigation.
Integrated with Google's ecosystem — Calendar, Maps, Gmail. Google has structural advantages here because it already knows where you're going (Maps), when you're free (Calendar), and what you do (Gmail). Gemini's agent surface is built on top of that.
Perplexity added transactional features in 2025 — find products, compare, initiate purchase. More commerce-oriented than general-task than the others.
Open protocol from Anthropic that lets AI agents connect directly to business systems (CRMs, databases, APIs). Growing ecosystem in 2026 — businesses that expose MCP servers become discoverable and actionable by any MCP-compatible agent.
2023: Customer googles "Edmonton dentist Saturday availability," clicks three sites, eventually calls one, leaves voicemail.
2026 agentic: Customer asks ChatGPT, "Find me a dentist in Edmonton who takes my insurance and has Saturday availability in the next two weeks, book me in." Agent queries AI-discoverable clinics, checks schemas, calls or API-books.
Clinics that show up in this flow need:
Clinics that don't? Completely invisible to agent flows.
2023: Customer googles, calls a couple, gets availability, picks one.
2026 agentic: Customer asks an agent to find "a plumber in Edmonton who does emergency work and has reviews over 4.5 stars, get me a quote for a leaking hot water tank." Agent pulls from Google Business Profile data, reviews, individual plumber sites. Some plumbers have agent-accessible quote forms; they get contacted first.
2023: Customer visits 5 agency sites, reads pricing pages, fills two contact forms.
2026 agentic: Customer asks, "Compare the three biggest Edmonton AI agencies on pricing for voice agent deployment." Agent pulls pricing data from public sources, compares, summarizes. Agencies with clear public pricing tables and service schema are quoted; agencies whose pricing requires a discovery call are skipped entirely in the initial round.
llms.txt or MCP endpoint for direct interactionA stunning Edmonton web design with zero structured data is visible to humans and invisible to agents. A plain but well-structured site gets agent traffic.
llms.txtAgents can't use this site at all. These businesses miss 100% of agentic queries.
Agents can discover the business and pass information back to the user. But they can't transact directly. Good enough for informational queries, weak for action queries.
dateModified signalsllms.txt with curated content indexAgents can get real information back — "yes this clinic accepts Sun Life and has Saturday slots." Book via a link the agent hands to the user, not directly through the agent.
Agents can complete the transaction directly. The user's agent books the appointment; the business's system accepts and confirms.
Based on our observation of Edmonton business sites in early 2026:
llms.txt, some published content but no structured commerce.llms.txt, clear pricing. Visible and queryable.The shift from Level 1 to Level 2 is the highest-ROI step for most Edmonton businesses in 2026.
llms.txt at your domain root with key pages indexed. See llms.txt complete guide.dateModified on key pages via automation (file mtime, CMS hooks) so freshness signals stay current./.well-known/mcp.json or similar endpoint advertising MCP capabilities if relevant to your business.GEO (Generative Engine Optimization) is about getting cited by AI answer engines. Agentic web optimization is about being used by AI agents taking action.
The skills overlap:
llms.txt helps bothWhere they diverge:
Most Edmonton businesses should focus GEO first (cheaper, faster payback) and add agent-specific capabilities later.
Some Edmonton businesses panicked in 2024-2025 and blocked GPTBot, ClaudeBot, PerplexityBot, etc. in robots.txt. In 2026 this removes them from both GEO and agentic contexts entirely. Rarely the right move.
The exception: businesses whose IP is the product (legal research databases, paywalled publishing, proprietary data tools). For those, selective blocking makes sense. For a dentist, plumber, or agency — blocking is usually self-harm.
Agents don't care about your brand voice in the first conversation. They want facts: hours, services, pricing, availability, insurance accepted. Keep the data clean; save the brand voice for the human-facing touchpoints.
Hiding different content for bots vs. humans. Cloaking. Keyword-stuffing for AI. These all get filtered, and modern AI engines are adversarial about detecting manipulation.
In 2026, many agentic flows will skip "call the business" entirely if a calendar link or booking form exists elsewhere. If your only conversion path is a phone call, you lose agent-mediated traffic. Add at least one self-serve or low-friction path.
Agentic web is more favorable to small businesses than traditional SEO was. Reasons:
It's operational but early. Real usage volumes are small fractions of traditional search in early 2026 but growing fast. By late 2026 / early 2027, agent-mediated traffic is projected to be a meaningful share of transactional queries for many categories.
Only if your current site is Level 0 (invisible). If you're at Level 1, incremental upgrades to schema + llms.txt + published pricing get you to Level 2 without a rebuild. Rebuilds are justified for sites already being considered for rebuild for other reasons (performance, modernization).
MCP (Model Context Protocol) is an open protocol for AI agents to connect to business systems. For most Edmonton SMBs, no — MCP implementation is overkill. For businesses with complex offerings (agencies, B2B SaaS, scheduling-heavy operations), yes — MCP servers become a real advantage in 2026-2027.
No, agents act on behalf of human customers. The human still makes the buying decision; the agent handles the legwork. Marketing to humans stays relevant — you still need to win the "which of these should I consider" phase. But now you also need to be findable and actionable by the agent doing the research.
Log-file analysis. Check for user agents like GPTBot, ClaudeBot, PerplexityBot, Google-Extended, OAI-SearchBot, and various "Operator" variants. Frequency trends reveal whether your site is being crawled for agent purposes.
Agents typically identify themselves and respect robots.txt. If an agent misbehaves (scraping aggressively, ignoring robots rules), you can block it. Most production agents play fair because their providers have reputational risk.
A small cohort of tech-forward clinics, a few law firms with modern websites, and some mid-size SaaS companies. Most traditional-industry businesses (plumbing, construction, retail) are 12-18 months behind. Not a criticism — a specific opportunity for those who move now.
Structural. The 2024-2025 generation of AI answer engines was surprising; the 2026 generation of agent engines is an order of magnitude more so. The change in how customers find and transact with businesses is real, even if the headline numbers still favor traditional search in 2026.
Want an agent-readiness audit for your Edmonton business? We'll score your site from Level 0 to Level 3 and give you a specific upgrade path. Book a free audit. See our AI SEO Edmonton service or the Edmonton AI agencies directory.