That is the gap for many growing firms. Prices live in PDFs. Buttons are styled divs. Cookie walls cover the form. Catalogue facts sit in images. Humans muddle through; agents stall, skip you, or recommend a competitor whose site exposes the facts in ordinary HTML.
What "agent-ready" means (service firms vs retailers)
Google's web.dev guidance frames how agents see a page: screenshots, raw HTML and the accessibility tree — the structured map of roles, names and states that assistive technology already uses. Clean signals across those views beat prettier-but-opaque interfaces.
Kristian's September 2026 audit at Dog on the Table puts the commercial split neatly:
- Lead-generation sites (agencies, consultants, most B2B): the agent should understand the offer, find useful price guidance and reach a labelled enquiry form. Bot protection may still stop the final submit — the person can finish that step.
- Transactional sites (retail, ticketing, bookable appointments): the agent needs accurate product data, live prices/availability and controls it can operate along the buying path.
You do not need Universal Commerce Protocol experiments on day one. You need the shortest path from homepage to the action that makes money — with every fact and control on that path machine-legible.
A practical checklist for UK SMEs
Work through these blocks in order. Most of them also improve accessibility and conversion for humans.
1. Readable HTML facts and prices
Put service descriptions, deliverables, "from" prices or ranges, shipping/returns headlines and product attributes in page text and HTML, not only in hero images, carousels or downloadable specs.
Dog on the Table's scan of 50 UK agency homepages (15 September 2026) found 68% published no price for their own services. Vague "contact us for a quote" pages force agents into market guesses — or omission. A defined package range (for example a brochure site rebuild with named deliverables) is enough to qualify interest without publishing every custom day rate.
2. Semantic links, buttons and labelled forms
Prefer real <a> and <button> elements over clickable div/spans. Link every form field to a persistent <label>. Keep layouts stable so "Add to basket" or "Send enquiry" sits in a predictable place. Remove ghost overlays that cover controls — including consent layers that arrive on top of the money button.
Google's agent-friendly checklist and Lighthouse's Agentic browsing checks all point the same way: accessible names, operable controls, visible state changes after actions. The Dog on the Table sample's most common critical failure was clickable elements without a proper role or keyboard access — classic "looks like a button, isn't one."
3. Product / catalogue data, schema and feeds (retailers)
For ecommerce, agents and AI shopping surfaces lean on structured catalogue facts: title, variants, price, availability, identifiers (GTIN/SKU/brand), shipping and returns as data where possible.
- Implement and validate Product structured data (Google Search Central's product markup docs remain the baseline).
- Keep Merchant Center / shopping feeds accurate — Varn's ecommerce GEO guidance treats feed quality and crawlable product pages as first-class GEO work, not a side project.
- Move purchase-critical attributes out of prose and into fields metafields/feeds can expose.
Shopify merchants: treat this as catalogue hygiene, not another storefront chatbot. Agent-facing protocols and apps change; messy variants, missing GTINs and stale stock still fail the audit. Keep Shopify's agentic-checkout story as a later option once the data layer is trustworthy — it is not the whole readiness programme for most UK SMEs.
4. robots.txt and crawl access
Confirm legitimate search and AI crawlers can reach the pages and feeds that matter. Accidental blanket bot blocks, broken sitemaps, JavaScript-only product shells and aggressive CDN defaults can hide a perfectly good catalogue. Log analysis (which bots hit which paths, which status codes) is more useful than guessing.
5. Checkout and login traps (ecommerce)
Test the path an agent or agent-assisted shopper must take: login walls, CAPTCHAs mid-funnel, promo logic that assumes human browsing, inventory lag, and two-factor prompts that need a phone in hand. Gladly's ecommerce readiness checklist is blunt — find whether you would lose a sale today before you obsess over emerging checkout protocols.
For lead-gen sites, the analogue is simpler: one clear form, labelled fields, minimal overlay theatre. Let the human press send if bot protection demands it.
6. Measuring AI referrals
Separate ChatGPT, Perplexity and similar referrers in analytics. Baseline now. Pair that with Search Console structured-data health and (for retailers) Merchant Center diagnostics. You cannot improve what you lump into "direct."
What not to over-invest in
- Schema spam — baseline Product / Organisation markup is cheap hygiene; dumping every Schema.org type on every page is not a substitute for readable facts and a clean catalogue.
- Flashy site chat widgets before catalogue and form quality — a bot that hallucinates your returns policy is worse than a quiet, accurate page.
llms.txtas vanity — useful as a pointer file for some teams; Dog on the Table's agency sample saw adoption rise, but it rarely fixes unnamed buttons or missing prices.- Experimental WebMCP / agent APIs before the accessibility tree works — watch the W3C community work; repair landmarks, labels and HTML first.
A 30-day starter plan for a lean UK team
Week 1 — Map the money path. Homepage → main conversion action. List every fact, overlay, form field and handover. Decide lead-gen vs transactional.
Week 2 — Fix legibility. Publish (or clarify) prices/ranges and key offer facts in HTML. Replace placeholder-only forms with labelled fields. Kill or defer overlays that block the primary control.
Week 3 — Data and crawl. Retailers: Product schema + feed audit on top sellers; confirm robots/sitemaps. Service firms: one service page per core offer with clear deliverables and guidance pricing.
Week 4 — Measure and retest. Tag AI referrers; spot-check the path with accessibility tree tools / Lighthouse Agentic browsing where available; ask ChatGPT or Perplexity a real customer question in your category and note whether your facts appear correctly.
Repeat quarterly. Agent behaviour moves fast; unnamed buttons and stale stock do not age well.
Why this matters for Clarity Growth
Clarity Growth builds and improves websites that have to work for humans and the machines now researching on their behalf — structured content, performance, and clean connections between catalogue, CRM and ops. Agent readiness is mostly disciplined fundamentals: readable offers, operable forms, trustworthy product data. When those foundations are solid, AI search and shopping surfaces have something worth recommending — and your team spends less time chasing vanity AI features that paper over a muddy site.
Sources
- Is your website ready for AI agents? — Dog on the Table (15 September 2026)
- Build agent-friendly websites — web.dev (Google)
- Ecommerce GEO: optimising for AI search — Varn
- AI agent readiness checklist for ecommerce — Gladly
- Is your Shopify store AI-ready? A 2026 audit checklist — what.digital
- Introduction to Product structured data — Google Search Central