Share the checklist:
0/X
Articles on the topic of AI agents and AI search: AI Agents and UX How AI Search Changes SEO Best AI SEO Tools Checkout UX How to Reduce Abandoned Carts

🧭 Strategy, Ownership & Agent Audit

  • Find Out How AI Agents See Your Business Today
    Before changing anything, ask ChatGPT, Gemini, Perplexity, Copilot and Claude the questions your buyers ask: “best [product] under $200”, “compare [your brand] vs [competitor]”, “book a [service] near me”. Record whether you appear, which facts are wrong, and which competitor gets recommended instead. This 30-minute test is your baseline and usually reveals the most expensive gaps immediately.
  • Map Which Buying Journeys an Agent Could Complete
    List your top revenue journeys (reorder a consumable, buy a best-seller, request a quote, book a demo or appointment) and mark for each one: can an agent research it, compare it, add to cart, pay and track the order without a human? Start with simple, repeatable, low-risk purchases — they are what shoppers delegate to agents first. Conversion Funnel
  • Assign One Owner for Agent Readiness
    Agentic commerce touches SEO, ecommerce, IT, payments, legal and customer support — so by default nobody owns it. Name one person responsible for the roadmap, the quarterly agent test and decisions on which agents to allow or block. Without an owner, bot settings get changed by the security team while marketing wonders why AI referrals disappeared.
  • Decide Your Stance: Open, Selective or Closed to Agents
    Make a written business decision per agent type: training crawlers, AI search crawlers, user-triggered browsing agents and checkout agents. Blocking everything protects content but removes you from AI recommendations; allowing everything can expose pricing and inventory to scrapers. Most stores choose “visible for discovery, verified for transactions”. Revisit the decision every quarter as platforms change.
  • Prioritise the Products and Services Agents Should Sell First
    Pick 10–50 items with clear specs, stable pricing, good margin, strong reviews and low return rates. Agents favour offers they can describe confidently, so these become your pilot catalogue for feeds, schema and protocol integrations. Launching with your whole catalogue at once multiplies data errors and makes it hard to see what works.
  • Set KPIs Before You Invest
    Record today’s AI referral sessions, AI-sourced orders and leads, conversion rate of AI traffic, share of voice in AI answers for 20 key prompts, and agent-related errors in server logs. Agentic sales often happen off-site through APIs, so define up front how each order will be attributed. Conversion Rate Calculator · ROI Calculator

🤖 Agent Access, Crawlers & Bot Management

  • Check robots.txt Rules for Each AI User-Agent Separately
    AI companies use different bots for different jobs. OpenAI: GPTBot (training), OAI-SearchBot (search), ChatGPT-User (actions for a user). Anthropic: ClaudeBot, Claude-SearchBot, Claude-User. Perplexity: PerplexityBot, Perplexity-User. Google-Extended and Applebot-Extended control AI training use only. A single “Disallow: /” for one bot name can silently remove you from AI search results. Robots.txt Generator
  • Make Sure Your CDN or WAF Isn’t Blocking Good Agents
    Many sites allow AI bots in robots.txt but block them at Cloudflare, Akamai, Imperva or the hosting firewall — often after a default “block AI bots” toggle was switched on. Review your bot-management rules and firewall logs for 403 and 429 responses to AI user-agents. What robots.txt allows, the firewall may still deny, and agents simply move on to a competitor.
  • Support Verified Agent Identity Instead of Guessing by User-Agent
    User-agent strings are easy to fake. Major agents increasingly sign their requests cryptographically (Web Bot Auth, based on HTTP Message Signatures) and card networks have launched trust frameworks such as Visa Trusted Agent Protocol and Mastercard Agent Pay. Ask your CDN and payment provider whether they can verify these signals, so you allow real shopping agents and block impersonators. Cloudflare Signed Agents
  • Don’t Put CAPTCHAs on Pages Agents Need to Read
    A CAPTCHA on product, pricing, stock or store-locator pages stops legitimate assistants just as effectively as scrapers. Keep challenges for high-risk actions (login, account creation, gift-card redemption, repeated failed payments) and use rate limits or verified-bot allowlists elsewhere. Every unnecessary challenge is a sale the agent can’t complete.
  • Render Key Content in HTML, Not Only via JavaScript
    Many AI crawlers fetch raw HTML and don’t execute JavaScript. If price, availability, variants, shipping costs or reviews load only client-side, the agent sees an empty shell. Check with “View source” or curl: product name, price, stock status and main description should be present in the initial HTML response. HTTP Header Checker
  • Keep XML Sitemaps Complete and lastmod Accurate
    AI search systems rely on the same discovery signals as search engines. Include every indexable product, category, policy and FAQ page, remove 404 and redirected URLs, and update lastmod only when content really changes. Submit sitemaps in Google Search Console and Bing Webmaster Tools — Bing’s index also feeds several AI assistants, and IndexNow gets changes to it faster. IndexNow
  • Consider an llms.txt File — but Don’t Rely on It
    llms.txt is a proposed Markdown file that points AI tools to your most important pages and documentation. It is cheap to add and useful for developer-facing and documentation-heavy sites, but major search engines don’t use it as a ranking or retrieval signal. Treat it as a nice-to-have after robots.txt, sitemaps, schema and feeds are solid. llms.txt Proposal
  • Monitor Agent Traffic in Server Logs Every Month
    Analytics tools that rely on JavaScript miss most bot and agent visits. Pull server or CDN logs and track requests, status codes and top URLs per AI user-agent. A sudden drop in OAI-SearchBot or Claude-SearchBot hits, or a spike in errors, is an early warning that you are disappearing from AI answers — weeks before revenue shows it. Error Tracking

🏷 Product Data, Feeds & Structured Data

  • Add Complete Product and Offer Schema to Every Product Page
    Agents read structured data to extract facts without guessing. Mark up name, brand, GTIN/MPN, SKU, image, description, price, priceCurrency, availability and itemCondition, plus product variants (ProductGroup) where relevant. Validate with Google’s Rich Results Test and make sure the markup matches the visible page — contradictions make agents distrust the whole page. Schema Markup Generator
  • Publish Shipping and Return Policies in Structured Data
    “Can it arrive by Friday?” and “Can I return it for free?” are questions agents answer before recommending a store. Add OfferShippingDetails (costs, handling and transit times by country) and MerchantReturnPolicy (return window, fees, method) at product or Organization level, and configure the same in Google Merchant Center. Missing policies make you look riskier than competitors who publish them.
  • Use Global Identifiers (GTIN, MPN, Brand) Wherever They Exist
    Agents compare the same product across many stores. Without a GTIN or MPN they can’t be sure your item is identical to the one they found elsewhere, so they may skip your offer. Fill identifiers for every branded product in your catalogue, feed and schema, and keep them consistent across marketplaces.
  • Write Attribute-Rich Descriptions, Not Marketing Fluff
    Agents shortlist by constraints: dimensions, materials, compatibility, capacity, ingredients, certifications, warranty, use case, who it’s not for. Put these as clear attributes and short factual sentences (“Fits 13–15-inch laptops; weight 850 g; water-resistant, not waterproof”). “Premium quality for modern life” gives an agent nothing to match against a user’s request.
  • Keep Price and Stock Identical Across Site, Feed and Schema
    When the feed says $49 in stock and the page says $54 out of stock, merchant programs disapprove products and agents lose trust. Sync price, sale price, availability and delivery estimates from one source of truth, ideally in near real time. Audit a random sample of 20 products weekly across page, schema, Merchant Center and any AI shopping feed.
  • Keep Your Google Merchant Center Feed Healthy and Rich
    Google’s AI Mode and Gemini shopping experiences draw on the Shopping Graph, fed largely by Merchant Center. Fix all disapprovals and warnings, fill optional attributes (product highlights, product detail, size, color, material, age group) and use supplemental feeds for data your platform can’t export. Rich feeds win more of the long, specific queries typical of AI shopping. Merchant Center Product Data Spec
  • Prepare a Product Feed for ChatGPT and Other AI Shopping Surfaces
    OpenAI accepts merchant product feeds for shopping results and Instant Checkout in ChatGPT, and similar programs exist or are rolling out at Perplexity and Microsoft Copilot. Check eligibility on your region and platform, apply, and export a clean feed with identifiers, availability, shipping and return info. Shopify, Etsy and other platforms may connect this for you — verify it is actually switched on. ChatGPT Merchants
  • Add Organization, Brand and Contact Schema
    Agents check who is behind a store before sending a user’s money there. Add Organization markup with legal name, logo, address, contact points, sameAs links to official profiles, and return policy. Keep the same name, address and phone across your site, Google Business Profile, marketplaces and directories so an agent can confirm you are one real, established business.
  • Mark Up FAQs, Size Guides and Compatibility Info as Plain Text
    Agents answer the follow-up questions shoppers ask before buying: “Is it true to size?”, “Does it work with iPhone 15?”, “Is it dishwasher-safe?”. Put these answers on the product page in visible HTML text (not only in images, PDFs or collapsed scripts), with clear question-style headings. Every question answered on your page is one the agent doesn’t need a competitor to answer.

📜 Policies, Pricing & Trust Signals Agents Verify

  • Show the Full Price Early — Including Shipping and Fees
    Agents compare total landed cost, not headline price. If shipping, taxes, handling or service fees appear only at the last checkout step, the agent may calculate you as more expensive — or mislead the user and trigger a return. Publish shipping rates and thresholds on a crawlable page and in schema, and avoid surprise fees in general. Checkout UX Guide
  • Write Return, Refund and Warranty Policies in Plain, Specific Language
    “Returns accepted under certain conditions” can’t be summarised by an agent. State the return window in days, who pays return shipping, refund method and timing, exclusions, and warranty length. Put the policy on a stable URL, link it from every product page and footer, and date it — agents and users both need to know it is current.
  • State Delivery Times as Concrete Ranges by Region
    Agents are often asked to buy something “that arrives before the weekend”. Give order cut-off times and delivery ranges per country or region (“UK: 1–2 business days; EU: 3–5”) on product pages, in schema and in feeds. Vague “fast shipping” loses to any store that states a date.
  • Keep Reviews Genuine, Recent and Accessible
    Agents weigh review volume, recency and rating, and increasingly cross-check third-party platforms (Google, Trustpilot, G2, Amazon, Reddit). Collect reviews continuously, mark up AggregateRating only for reviews you actually host, and reply to negative ones. Fake or incentivised reviews breach FTC and EU rules and can get you excluded from AI shopping programs. What Is Social Proof
  • Make Contact and Support Details Easy to Find and Parse
    Before recommending a merchant, agents look for signs of a real business: a physical address, phone, support email, business hours and company registration details. Put them on a crawlable contact page and in the footer as text, not an image. Stores hidden behind a contact form only look like a risk to both agents and customers.
  • Keep Your Brand Facts Consistent Everywhere
    AI models build their picture of you from your site, Wikipedia/Wikidata, LinkedIn, marketplaces, press and review sites. Mismatched founding dates, prices, locations or product names produce confused or wrong answers. Audit your top 15 external profiles every six months and correct outdated information at the source. How AI Search Changes SEO
  • Publish Honest Comparison and “Who It’s For” Content
    A large share of agent tasks are comparisons. Create factual pages comparing your product with alternatives and explaining which customer each option suits best — including when yours is not the right choice. Agents quote balanced, specific comparisons far more readily than one-sided sales pages. LLM Seeding

🖱 Agent-Friendly Site UX & Checkout Flows

  • Use Semantic HTML and Real Buttons, Links and Form Fields
    Browser agents such as ChatGPT agent mode, ChatGPT Atlas, Perplexity Comet and Gemini in Chrome navigate using the page’s accessibility tree and screenshots. Clickable <div>s without roles, icon-only buttons and custom dropdowns confuse them. Use <button>, <a>, <select> and <label> elements with clear accessible names like “Add to cart – Blue, size M”. It also improves accessibility for people.
  • Make Variant Selection Explicit and Unambiguous
    Size, color and configuration pickers are where agents most often fail: swatches without labels, sold-out variants that still look selectable, prices that only update after a click. Label each option in text, disable unavailable combinations clearly, show variant-specific price and stock, and give each variant its own URL or parameter. Session Recordings
  • Allow Guest Checkout Without Forced Account Creation
    An agent acting for a user can’t (and shouldn’t) create accounts or invent passwords. Mandatory registration stops agent purchases outright — and it already costs you human shoppers too. Offer guest checkout, create the account optionally after purchase, and support wallet payments that carry the address. How to Reduce Abandoned Carts
  • Remove Pop-Ups and Overlays That Block the Purchase Path
    Newsletter modals, spin-to-win wheels, chat bubbles over the “Add to cart” button and aggressive exit popups interrupt agents the same way they interrupt humans — except the agent may give up. Delay promotional popups, keep them off cart and checkout, and make every overlay closable with a clearly labelled button. Heatmaps
  • Make Checkout Error Messages Specific and Text-Based
    “Something went wrong” leaves an agent looping. Show field-level messages in text next to the field (“ZIP code doesn’t match the selected state”), keep entered data after an error and never rely on color alone. Track which form errors occur most — they are the same ones blocking humans. Smart Forms · Error Tracking
  • Support Wallet and Stored Payment Methods
    Apple Pay, Google Pay, PayPal, Shop Pay and Link let a checkout finish without typing card numbers — the path agent platforms prefer, because the user confirms in a trusted wallet. Make these options visible on product, cart and checkout pages and test they work in the browsers agents use (Chromium-based).
  • Keep Checkout Short, Stable and Predictable
    Every extra step, redirect or layout shift is a chance for an agent to lose its place. Aim for a single-page or two-step checkout, avoid A/B tests that radically change the checkout structure mid-session, and keep promo code and upsell steps optional. Measure where both human and agent sessions drop off. Checkout UX With Heatmaps
  • Expose Booking, Quote and Availability Details for Services
    For services, the “purchase” is a booking or a quote request. Publish price ranges or starting prices, service areas, available time slots and what’s included. Use a booking system with a clear public page (or one connected to Google Reserve or similar integrations) and a short quote form with labelled fields. “Call us for pricing” means the agent calls a competitor instead. Local SEO That AI Can’t Ignore

🔌 Protocols, APIs & Agentic Checkout Integrations

  • Understand the Main Agentic Commerce Protocols
    Know the landscape before a vendor sells you a “solution”: Agentic Commerce Protocol (ACP, OpenAI and Stripe) powers checkout inside ChatGPT; Universal Commerce Protocol (UCP, Google with major retailers and Shopify) covers checkout in Google’s AI surfaces; Agent Payments Protocol (AP2, Google) standardises how agents prove a user authorised a payment; Model Context Protocol (MCP) connects AI assistants to your tools and data. Agentic Commerce Protocol
  • Check What Your Ecommerce Platform Already Supports
    Shopify, BigCommerce, WooCommerce, Salesforce Commerce Cloud, Adobe Commerce, Wix and others are shipping agentic channels, AI catalog syndication and protocol support at different speeds. Ask your platform and payment provider (Stripe, PayPal, Adyen, Checkout.com) what is available in your region, what needs to be enabled and what it costs. Often the fastest path is a setting, not a development project.
  • Offer a Real-Time Product, Price and Inventory API
    Agents need fresh facts at the moment of purchase, not yesterday’s feed. Provide (or enable through your platform) an endpoint returning current price, stock, variants and delivery estimate per SKU. Document it, secure it with keys or verified-agent access, and set rate limits so it can’t be abused by scrapers.
  • Support Cart and Checkout via API, Not Only via the Browser
    Protocol-based checkout (ACP, UCP) lets an agent create a checkout, apply shipping, calculate tax and confirm the order through your backend while the user pays inside the assistant. Your order system must accept these orders, return clear totals and errors, and treat them like any other order in fulfilment, emails and accounting.
  • Consider an MCP Server for Your Store or Service
    An MCP server lets AI assistants search your catalogue, check order status, book appointments or pull account data through defined, permissioned tools instead of scraping pages. Several ecommerce platforms and payment providers already offer ready-made MCP servers. Start read-only (catalog search, order status) and add actions like cart creation only with strict authentication. Model Context Protocol
  • Build Idempotency and Clear Order Webhooks Into Integrations
    Agents retry requests when a connection drops. Without idempotency keys, one retry can create two orders and two charges. Require idempotency keys on order and payment creation, send webhooks for status changes (confirmed, shipped, delivered, cancelled, refunded) and make sure the agent platform receives them so it can update the user.
  • Make Post-Purchase Actions Agent-Accessible
    Agents won’t stop at checkout: users will ask them “Where is my order?”, “Change the delivery address” or “Return this”. Provide order-tracking pages that work with order number plus email (no login required), tracking links in confirmation emails, and a self-service return flow. Each of these that an agent can’t handle becomes a support ticket for your team.
  • Keep a Human-Readable Changelog of Your Agent Integrations
    Document which agents, feeds, protocols and APIs are live, which products are included, who approved them and which credentials they use. Protocols and platform programs are changing fast; when something breaks at 2 a.m., a one-page record saves hours and prevents accidentally switching off a revenue channel.

💳 Payments, Fraud & Legal Readiness

  • Tell Your Fraud Tool That Agent Orders Are Coming
    Agent orders look unusual to legacy fraud rules: new device fingerprints, data-center IPs, instant form filling, no mouse movement. Without tuning, you will decline good orders. Ask your payment and fraud provider how they identify agent-initiated transactions (network tokens, agent signals, protocol metadata) and review declines of orders marked as agent traffic every week.
  • Use Tokenised Payments and Never Store Raw Card Data
    Agentic payment flows are built on network tokens and delegated, scoped payment credentials (for example, Stripe Shared Payment Tokens, Visa Intelligent Commerce, Mastercard Agent Pay). Make sure your PSP supports them and that no agent flow ever passes full card numbers through your systems. It reduces PCI scope and fraud risk at the same time.
  • Keep Strong Customer Authentication Where It Is Required
    In the EU, UK and other regulated markets, SCA/3-D Secure still applies when an agent pays. Check with your PSP how agent-initiated payments handle authentication and exemptions, and make sure the challenge reaches the user (in their wallet or bank app) instead of silently failing the order.
  • Define Your Policy for Disputed Agent Purchases
    “My AI bought the wrong size” will become a common refund reason. Decide in advance how you handle agent errors, duplicate orders and purchases the user claims not to have authorised, and document proof you keep (order confirmation, payment mandate, agent identity). Consistent handling protects your chargeback ratio and customer relationships.
  • Update Terms of Service for Automated and Agent Access
    Your terms should state which automated access is allowed, that orders placed by a user’s agent are binding on the user, how pricing errors are handled, and what scraping or reselling is prohibited. Have a lawyer review it for your markets — consumer protection rules still apply to purchases initiated by agents.
  • Check Privacy Notices and Data Sharing With AI Platforms
    Agentic checkout means sharing customer names, addresses and order data with AI platforms and payment partners. Update your privacy notice, data processing agreements and records of processing under GDPR and similar laws, and share only the data needed to fulfil the order. Don’t let a quick integration become a compliance incident.
  • Protect Promotions and Limited Stock From Automated Abuse
    The same automation that helps shoppers can drain limited drops and stack coupon codes. Use per-customer limits, single-use codes, verified-agent rules and velocity checks on high-demand items — without blocking verified agents on regular products. Review coupon usage patterns after every major campaign. Ecommerce Discount Strategy

📊 Measurement, Testing & Ongoing Monitoring

  • Track AI Referral Traffic as a Separate Channel
    Create a channel group in GA4 (or your analytics) for referrers such as chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com and claude.ai, and capture the utm_source=chatgpt.com parameter ChatGPT adds to many links. Compare conversion rate and order value with other channels — AI referrals often convert above average, which justifies further investment. UTM Builder
  • Attribute Orders Placed Inside AI Assistants
    Orders completed via ChatGPT Instant Checkout, Google AI surfaces or other protocols may never create a session on your website. Tag them with a distinct sales channel in your ecommerce platform, CRM and reporting, and include them in revenue dashboards. Otherwise your analytics will show falling traffic while the real sales channel grows unseen. Ecommerce Sales Performance
  • Run a Monthly “Agent Mystery Shop”
    Every month, give two or three agents the same task: “Buy [product] in size M, delivered by Friday, under $80”. Record where they get stuck — variant pickers, popups, CAPTCHAs, payment, wrong facts — and fix the top issue. Repeat after major site releases, theme updates and new bot-protection rules.
  • Watch Session Recordings of Automated Visits
    Recordings of browser-agent sessions show exactly where an agent hesitated, clicked the wrong element or abandoned checkout, just like a human visitor. Filter for sessions with unusual behaviour (very fast form fills, no scroll, repeated clicks on one element) and compare them with heatmaps of the same page. Session Recordings · Heatmaps
  • Track Your Share of Voice in AI Answers
    Pick 20–50 buying prompts for your category and check monthly whether AI assistants mention, recommend or link to you, and what they say about price, quality and policies. Log competitors that appear instead. Improvement in these prompts is the leading indicator for agentic sales. Best AI SEO Tools
  • Set Alerts for Changes That Silently Break Agent Access
    A robots.txt edit, a new noindex tag, a firewall rule or a broken schema field can cut you off from AI agents overnight. Set automated alerts for changes to robots.txt, meta robots, canonical tags, structured data errors and spikes in 4xx/5xx responses to AI user-agents. SEO Alerts
  • Review the Checklist Every Quarter
    Agentic commerce is changing month by month: new protocols, new merchant programs, new browser agents and new rules. Set a quarterly review with your owner, IT and payments teams, re-run the agent test from the first section and update this checklist with what changed. Businesses that iterate steadily build an advantage competitors can’t copy overnight.