01 / Understand how agents buy
Autonomous AI agents — ChatGPT agents, Perplexity agents, Claude with tool use — research, compare and transact on behalf of humans. They never see a results page. They build a shortlist and act on it.
Do this
- Give an AI agent a real buying task from your category: 'Compare the 3 best vendors for X and recommend one.'
- Watch which sources the agent reads and which vendors it names.
- Note the reasoning — agents explain their choice.
- Repeat the task in ChatGPT, Perplexity and Claude.
Example: A B2B buyer asks ChatGPT to shortlist 'the three best time-tracking tools for trade businesses'. The agent reads review portals and comparison pages — and recommends exactly one vendor.
What you get: You see the new buying process with your own eyes — and where you do (or don't) appear in it.
02 / Run your agent shortlist audit
Before you optimize, you need the baseline: for which tasks does an agent pick you — and where does it pick a competitor, and why?
Do this
- Define 10 realistic agent tasks ('find', 'compare', 'recommend', 'book').
- Run each task in at least 3 agent environments.
- Log: shortlisted? recommended? excluded with what verbatim reasoning?
- Repeat monthly — agents change with every model update.
Example: A software vendor learned from the agent's reasoning: 'no transparent pricing found'. The fix was a public pricing page — not a marketing budget.
What you get: A measurable agent-inclusion rate — the agentic equivalent of rank tracking.
03 / Make your product data machine-consumable
Agents act on structured facts, not prose. Price, availability, specifications and terms must be retrievable without guessing.
Do this
- Implement Product, Offer and Service schema with stable @ids.
- Publish prices, delivery times and terms as clear text and structured data.
- Provide a product feed (JSON or CSV) an agent can fetch.
- Validate everything with a schema validator.
Example: A shop added complete Offer schema with prices and delivery time. Shopping agents started including it in price comparisons — before, it was silently skipped.
What you get: Agents can fetch and trust your facts — instead of guessing or leaving you out.
04 / Build verifiable trust signals
An agent won't stake its recommendation on claims it can't verify. It weights independent reviews, consistent company data and third-party mentions.
Do this
- Build reviews on G2, Trustpilot, ProvenExpert or Google — real ones, continuously.
- Keep name, address and company data exactly identical everywhere.
- Get mentioned in industry directories and trade media.
- Link all profiles via sameAs in your Organization schema.
Example: Two vendors, same offering: the one with 40 verifiable Google reviews gets recommended by the agent. The one without doesn't even appear in the reasoning.
What you get: The proof an agent needs before it prefers you over a named rival.
05 / Write comparison content agents quote
Agents compare. If you own the honest comparison page ('Us vs Alternative A vs Alternative B'), the agent quotes your criteria — and your positioning.
Do this
- Build an honest 'you vs alternatives' page with real criteria.
- Say what the competitor is better at, too — that's what makes the page credible.
- Structure the comparison as a table plus short, extractable paragraphs.
- Keep the page current (visible date).
Example: A SaaS vendor published 'X vs Y vs Z — the honest comparison'. Agents adopted the page's criteria as their comparison framework — and the vendor defined the playing field.
What you get: You supply the comparison framework instead of drowning in someone else's.
06 / Make every task resolvable in one retrieval
An agent that needs three pages to find price, audience and contact gives up and takes the next vendor. Key pages must answer the whole task in one pass.
Do this
- Answer on every key page: What? For whom? What does it cost? How to start?
- Put a 40–80 word TL;DR block at the top.
- Make the next step (demo, contact, purchase) machine-readable as a clear link.
- Remove pop-ups and barriers that stop an agent.
Example: An agency put services, price range and contact path on one page. Agent answers started reproducing all three correctly — including the link.
What you get: Decision-ready content an agent can use in a single retrieval.
07 / Open crawler access for agents
Agentic fetchers (ChatGPT-User, Perplexity-User, Claude-Web) are the agents' eyes. Block them in robots.txt and you simply don't exist to agents.
Do this
- Check your robots.txt for blocks on GPTBot, ChatGPT-User, PerplexityBot, ClaudeBot & co.
- Explicitly allow the agentic user agents.
- Publish an llms.txt listing your key pages.
- Test it: ask an agent to read a specific page of your site.
Example: A retailer wondered why ChatGPT never knew its prices — robots.txt blocked all AI crawlers. One line changed; two weeks later prices were quoted correctly.
What you get: Agents can see you at all — the precondition for everything else.
08 / Prepare APIs and MCP readiness
The next level: the buyer's agent talks directly to your systems — checking availability, verifying price, booking. The Model Context Protocol (MCP) is becoming the standard for this.
Do this
- Document existing APIs publicly and clearly.
- Assess which use cases suit an MCP integration (inventory, pricing, booking).
- Start with a read-only endpoint — transactions come later.
- Watch which agent platforms integrate your category first.
Example: A booking provider exposed a simple availability endpoint. Agents could check live and link directly — the competitor without an endpoint was merely 'mentioned'.
What you get: Your systems are ready when buyers' agents start transacting directly.