The AI-Ready Website Playbook: 4 Optimization Layers for Humans, Search Engines, and AI Agents

Your website was probably designed for one audience and judged by one scoreboard: the human visitor and Google’s traditional rankings. That model has broken, and the shift has been quiet enough that you may not have noticed.

Your site now has three audiences (the human visitor, the search engine crawler, and the AI agent acting on the visitor’s behalf), and the gap between them is where most businesses are silently losing pipeline.

When AI Overviews appear for a query, the top-ranking page now sees a 58% lower clickthrough rate (Ahrefs, 300,000 keyword study, December 2025 data).

When a buyer asks ChatGPT “what is the best CRM for a 12-person law firm,” the agent reads structured data, schema, and FAQ blocks, then decides which two or three companies to recommend.

When that same buyer’s autonomous browser, like ChatGPT Atlas or Perplexity Comet, navigates a competitor’s site, it skips anything that requires JavaScript to render the price.

This is the SMB playbook for getting your website AI-ready (without ignoring your other two audiences).

Four working layers, a 90-day retrofit you can run on your existing website, four industry scenarios, and our own receipts to show we’re not just preaching, but also acting on our own advice.

---

Why Your Website Now Serves Three Audiences

The old UX rules still apply. Page speed still matters. Scannable content still matters. Clear calls to action still matter.

None of that changed.

The second audience also has not changed in any fundamental way: search engines have been crawling and ranking your site for two decades, and traditional SEO (Core Web Vitals, schema, internal linking, backlinks, intent-matched content) still drives most of the organic traffic that becomes pipeline.

Googlebot and Bingbot are not going away. The traditional 10 blue links are not going away. The work you have always done for them still matters.

What changed is the third audience: the AI systems that read, summarize, cite, and increasingly act on the content of your website. Google AI Overviews, ChatGPT, Perplexity, Claude, Gemini, Microsoft Copilot, and the new agentic browsers (ChatGPT Atlas, Perplexity Comet, The Browser Company’s Dia) are all part of this third layer.

They are layered on top of search engines, not replacing them, and they have their own rules about what they extract, cite, and act on.

The three signals below are about the third audience specifically, because the third audience is the one most SMB sites are not yet built for.

The First Signal: AI Overviews are Eating Organic Clicks

Google rolled out AI Overviews to most search results during 2024 and 2025. The traffic damage is now measurable, large, and worse than the early studies suggested.

Ahrefs ran the same 300,000-keyword study twice.

In April 2025, AI Overviews correlated with a 34.5% reduction in clicks to the top-ranking page.

In their December 2025 follow-up, that number reached 58%. Half-and-then-some of the clicks the top-ranking page used to get are now answered inside the SERP and never reach the website.

A reduction that large does not mean SEO is dead. It means the organic traffic that does land is more valuable, more buyer-intent, and more concentrated, and the rest is being arbitrated by an AI system that decides whether to surface you in the answer at all.

The Second Signal: AI Assistants are the New Homepage for Research

ChatGPT, Perplexity, Claude, Gemini, and Microsoft Copilot are the front door for an enormous and growing share of product research and vendor comparison.

When a 14-person dental practice asks Perplexity “what is the best practice management software for a small dental office in 2026,” the answer is generated from sources Perplexity selects.

If your site is one of the sources, you are in the consideration set for that practice. If not, you do not exist for that buyer.

This is what the SEO industry started calling AEO (answer engine optimization) and GEO (generative engine optimization).

The labels matter less than the mechanism: AI systems extract passages, not pages. They favor structured content. They cite sources that look authoritative, and they cite them more often when the page makes the citation easy.

The Cornell University GEO research evaluated nine optimization methods across Perplexity, found measurable visibility lifts: +40% from citing sources, +37% from adding statistics, +30% from expert quotes, and +25% from authoritative tone.

The same study measured a negative 10% impact from keyword stuffing. Keyword stuffing was already useless in traditional SEO. In AI search, it actively hurts you.

The Third Signal: Agentic Browsers Have Arrived

Late 2025 and early 2026 saw the launch of an entirely new browser category.

ChatGPT Atlas (OpenAI), Comet (Perplexity), Dia (The Browser Company, the team behind Arc), Opera Neon, and Microsoft Edge with Copilot now ship with assistants that read the page, summarize, compare across tabs, and in Atlas’s and Neon’s case, take actions on the user’s behalf.

These browsers reward the same things:

• clean DOM
• semantic HTML
• exposed pricing
• labeled buttons
• accessible forms.

They punish the same things:

• JavaScript-gated critical content
• login walls
• popups blocking primary content
• infinite scroll without working pagination
• CTAs that look like buttons but are unlabeled div elements

If you have ever watched an agent in Comet’s research mode try to compare two SaaS pricing pages and one of them is a “talk to sales” wall, you already know how this ends.

The wall loses. The transparent competitor gets recommended. The buyer never types your URL.

How This Connects to the Rest of imFORZA’s Playbook

This pillar is the operational layer of two pillars we already published.

The website is where the Brand Brain assets we wrote about in the M10 guide (your ICP file, your voice file, your design system, your competitor file) actually meet the public.

If those assets exist only inside your AI tools, they are invisible to the agent doing the research.

The website is the surface where they become readable.

The website is also the interface of the AI marketing system we described in the ARMS pillar.

Forms feed CRM. CRM feeds nurture. Nurture feeds offers.

None of those loops close if the front door is broken for either audience.

Said differently: this article is about the front door. The other two are about what happens inside the building.

---

The Three-Audience Test

Before the playbook, take the test.

The score you get from the audit is the entry point for the rest of this article.

If your weakest layer is Performance and Accessibility, Layer 1 is your first 30 days.

If your weakest layer is the Machine-Readable infrastructure, Layer 3 is.

The 90-Day Retrofit Sequence below maps directly to which layers your audit flagged.

---

---

The Four Layers of an AI-Ready Website

Every website that performs well for all three audiences has the same four layers in good shape. You do not need a new framework or a new platform. You need each of these four layers working.

Layer 1: Performance and Accessibility Foundation

This is the layer where humans win, search engines win, and AI agents win at the same time. Every fix here serves all three audiences. Core Web Vitals are a Google ranking signal, semantic HTML is what AI agents extract, and accessibility is the experience your human visitors actually feel.

Core Web Vitals 2026 thresholds

Google’s published thresholds for the three Core Web Vitals are unchanged from the 2024 update that replaced FID with INP:

Metric	Good Threshold	What it Measures
Largest Contentful Paint (LCP)	<= 2.5 seconds	When the main content visually loads
Interaction to Next Paint (INP)	<= 200 milliseconds	How responsive the page feels to clicks, taps, and key presses
Cumulative Layout Shift (CLS)	<= 0.1	How much the layout jumps around as the page loads

(Source: Google Search Central, Core Web Vitals reference.)

A site at the 75th percentile across all three metrics on both mobile and desktop is what Google calls “passing.”

Across the few dozen-plus SMB WordPress audits we have run in Q1 2026, the failure pattern is consistent:

• INP fails first (theme JavaScript blocking the main thread)
• CLS fails second (images without dimensions, late-loading ads)
• LCP fails third (oversized hero images, render-blocking fonts)

The fixes are documented and almost always cheaper than rebuilding.

Most sites move into the green inside 30 days with quality managed hosting, a configured caching layer, optimized images, and a theme audit that removes JavaScript bundles the page does not actually use.

WCAG 2.2 AA is the Accessibility Floor

WCAG 2.2 has been the W3C recommendation since October 2023.

It adds nine new success criteria on top of WCAG 2.1, including focus appearance, a 24 by 24 CSS pixel target size minimum, single-pointer alternatives to dragging, accessible authentication that does not require cognitive tests, and consistent help placement.

A screen reader and a Comet agent are reading the same DOM. Labeled inputs, real semantic headings, and descriptive alt attributes serve both at the same time.

How AI Agents Actually Read Your Site

In April 2026, Google’s Chrome team published official guidance on building agent-friendly websites.

Their core observation maps directly onto the three-audience case…

Agents read your site three ways at once and cross-reference all three:

• Screenshots. A vision model captures the rendered page and identifies elements by visual cues (size, color, position, button shape). This is slow and token-expensive, so agents use it as a backup when the DOM is confusing.
• The DOM. The agent reads the raw HTML and infers structure from how elements are nested, what tags are used, and what attributes describe them. A <button> inside a product card belongs to that product. A <div> styled to look like a button does not communicate its function.
• The Accessibility Tree. This is the browser’s built-in semantic summary of the page, originally built for assistive technology like screen readers. For an agent, it is a high-fidelity map of every interactive element with its role, name, and state, with the visual noise of CSS stripped out.

An agent’s success rate goes up when these three views agree.

Build a button as a real <button> with a clear label and a cursor: pointer style, and all three modalities tell the same story. Build a button as a <div> styled with JavaScript and the agent has to infer from screenshots only, which is slower, more error-prone, and more likely to skip the action entirely.

The same audit that fixes accessibility for screen-reader users fixes the accessibility tree for agents. There is no separate “agent audit.”

The Chrome team’s specific markup recommendations, all of which apply equally to humans on assistive tech and to agentic browsers like Comet, Atlas, and Dia:

• Use semantic HTML for every interactive element. <button> over <div role="button">. <a> over <div onclick>. If you cannot use semantic HTML, set the appropriate role and tabindex on the element.
• Set cursor: pointer on every actionable element. Chrome’s guidance specifically calls this out as a strong actionability signal that agents pick up on.
• Add for attributes on every <label> to link them to their inputs. Agents use the label-to-input link to understand what each form field is for.
• Make sure every interactive element has at least 8 square pixels of visible area. Smaller elements get filtered out by the vision-model layer of agentic browsers and are effectively invisible to those agents.
• Avoid “ghost” elements: transparent overlays, decorative divs positioned over real content, or invisible click-catchers. Vision models discard nodes they cannot see, even when those nodes are functional.
• Keep the layout stable across page types. If the “Add to cart” button moves to a different position on each product category page, screenshot-based agents lose track of it.

You can preview your own accessibility tree in Chrome DevTools (Elements panel → Accessibility tab).

If the tree is shallow, the labels are missing, or the roles are wrong, that is what your agents are also seeing.

What We Recommend for SMBs on WordPress

• Move to a managed WordPress host with a dedicated performance team (our own managed hosting is built for this reason).
• Audit themes and plugins twice a year. Remove anything that adds JavaScript or CSS without earning its weight.
• Serve modern image formats (WebP or AVIF) with explicit width and height. Defer non-critical CSS and JavaScript.
• Track 28-day field data in PageSpeed Insights monthly. Field data is what Google actually uses.
• Run an automated accessibility scan monthly (axe DevTools, Pa11y, or WAVE) and fix what it finds before it ships.

Layer 1 Receipt

Here’s a recent example from our team:

Layer 2: Content Architecture

This is the layer where humans skim, search engines rank, and AI extracts.

All three audiences want the same thing: the answer first, the context second, structure they can navigate without rendering JavaScript.

Lead with the Answer

The most-extractable passage on any page is the first 40 to 60 words that follow an H2 heading phrased as a question or topic. AI systems pull these passages directly into AI Overviews and Perplexity citations. The Princeton GEO study documented that pages with clear, self-contained answer passages near the heading saw the largest visibility lifts.

Bury the answer four paragraphs in and you have written a page nobody (human or model) wants to extract.

Use the right block for the job

For each common query intent, there is a content block pattern that performs best:

• Definition blocks for “what is X” queries. One paragraph, 40 to 60 words, leading with the definition.
• Comparison tables for “X vs Y” queries. Tables outperform prose for comparison content because both humans and AI agents can extract a row at a time.
• Numbered step lists for “how to” queries. Tag with HowTo schema (covered in Layer 3).
• FAQ blocks for “how much / how long / can I” queries. Tag with FAQPage schema (covered in Layer 3).
• Statistic blocks with cited sources for “what percentage” queries.

If you sell a service that buyers compare against alternatives and you do not have a “you vs the alternatives” page that is fair, balanced, and updated, you are leaving the most-cited content type on the table.

Headings should match how people phrase the query

A header that reads “Solutions” tells the AI nothing. A header that reads “How to Update Your robots.txt for ChatGPT, Perplexity, and Claude” tells it exactly what it is about, matches the query phrasing, and gets the section extracted as a unit.

Use H2 for top-level sections, H3 for sub-points, and never skip a level. Skipped heading levels confuse both screen readers and content extractors.

Layer 2 receipt

The Brand Brain pillar we published in April is structured this way on purpose.

Every H2 reads like a query. Every section leads with a 40 to 60 word answer block. Every list of “what assets do you need” is rendered as a real list, not as a prose paragraph.

We track AI citations on that pillar monthly. The structure is the reason it earns them.

What We Recommend

• Audit your top 20 pages. For each one, identify the primary query the page is trying to answer. Rewrite the H2 and the first paragraph to lead with the answer.
• Convert any prose comparison into a table.
• Convert any prose process into a numbered list.
• Add an FAQ block with 6 to 10 natural-language questions to every pillar-length page.
• Set a “definition first” rule for every new piece of content.

Layer 3: The Machine-Readable Layer

This is the layer most SMBs are completely missing. It is also where the cheapest, fastest wins live.

The machine-readable layer is everything that the AI reads but the human never sees:

• The schema graph in the page source
• The directives in robots.txt, the structured files at the site root (sitemap.xml, llms.txt, pricing.md, AGENTS.md)
• The JSON-LD blocks that describe your business as an entity

Schema Markup is the Foundation

JSON-LD schema in the <head> of every page tells the AI agent what it is looking at.

Pages with valid structured data are roughly 2.3x more likely to appear in Google AI Overviews than equivalent pages without it. (Cited via Duane Forrester’s machine-readable content stack analysis.)

For an SMB, the schema graph that should be on every page includes at minimum:

• Organization (your business as an entity, with sameAs references to your social profiles)
• WebSite and WebPage (the site and the specific page)
• BreadcrumbList (the path the visitor took to get here)
• Page-type-specific schema: Article or BlogPosting for blog posts, Service for service pages, Product for product pages, LocalBusiness if you have a physical location, FAQPage for any page with an FAQ block, HowTo for any step-by-step page

You can verify your current schema with Google’s Rich Results Test and Schema.org’s validator.

Receipt: Our Website’s Current Schema Graph

We checked the schema graph on our M10 pillar from a few weeks ago.

Here is the actual list of @type declarations on the page:

• Article
• AggregateRating
• BreadcrumbList
• ContactPoint
• EntryPoint
• ImageObject
• ListItem
• Offer
• OfferCatalog
• Organization
• Person
• PostalAddress
• PropertyValueSpecification
• QuantitativeValue
• ReadAction
• SearchAction
• Service
• WebPage
• WebSite

That is the level we recommend for an SMB blog post.

robots.txt for AI

Every AI platform crawls with its own bot. If you block (or fail to allow) the bot, that platform cannot cite you. The bots that matter in 2026:

• GPTBot and OAI-SearchBot (OpenAI / ChatGPT)
• ChatGPT-User (the on-demand fetcher when a ChatGPT user asks about a URL)
• PerplexityBot (Perplexity)
• ClaudeBot and anthropic-ai (Anthropic / Claude)
• Google-Extended (Google Gemini, separate from regular Googlebot)
• Bingbot (Microsoft Copilot via Bing)
• Applebot-Extended (Apple Intelligence)

Most WordPress sites ship a default robots.txt (or one through a popular SEO plugin, like the one from Yoast SEO) that does not mention any of these. That is not the same as blocking them, but it does mean the file is doing nothing useful for AI visibility.

Receipt: Our Website’s robots.txt Update That We Shipped with This Article

When we audited our own robots.txt, it was the default Yoast block: standard Disallow rules for ?s=, /wp-json/, and the search and pagination URLs. Zero AI bot directives.

We shipped this update alongside this pillar:

# Allow all major AI crawlers explicitly
User-agent: GPTBot
Allow: /

User-agent: OAI-SearchBot
Allow: /

User-agent: ChatGPT-User
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: ClaudeBot
Allow: /

User-agent: anthropic-ai
Allow: /

User-agent: Google-Extended
Allow: /

User-agent: Applebot-Extended
Allow: /

If you want to allow citation-grade crawling but disallow training-only crawls, the middle-ground move is to allow the bots above and explicitly disallow CCBot (the Common Crawl bot used as training data for many open models).

llms.txt: Ship It but Do Not Bet On It

llms.txt, proposed by Jeremy Howard at llmstxt.org in 2024, is a Markdown file at your site root that gives AI systems a curated index of your most important pages.

The proposal is sound. The execution is cheap.

Every SMB site should have one.

Here is the part most “AI-ready” guides will not tell you: there is currently no measurable evidence that having an llms.txt increases AI citations.

SE Ranking analyzed 300,000 domains in November 2025 and found:

• Only 10.13% of domains have an llms.txt at all.
• Adoption is not concentrated among high-traffic sites; mid-traffic and high-traffic sites adopt at similar rates.
• An XGBoost model trained to predict AI citation frequency performed better when the llms.txt variable was removed. The file’s presence introduced noise, not signal.

A separate August 2025 audit of 1,000 Adobe Experience Manager domain CDN logs (Longato.ch) found that LLM-specific bots are essentially absent from llms.txt requests, and Google’s own crawler accounts for the vast majority of file fetches that do happen. Google has publicly stated that AI Overviews and AI Mode rely on traditional SEO signals, not llms.txt.

We still recommend shipping one for these reasons:

1. The file is trivially cheap. A 50-line Markdown file you generate from your sitemap.
2. It is a public signal of seriousness. Buyers, vendors, and prospects who check it form a positive impression.
3. Standards form around early adopters. The brands that shipped Schema.org in 2012 shaped how Google parsed structured data for the next decade.

For reference, our own llms.txt lives at the standard path and is generated automatically from our sitemap with a hand-curated header.

pricing.md: Machine-Readable Pricing for Buyer Agents

If your pricing lives behind a “talk to sales” form, a buyer’s AI agent cannot evaluate you.

The agent compares the vendors whose pricing it can read and recommends from that subset. The opaque vendors are filtered out before the human ever sees the comparison.

pricing.md (or pricing.txt) is a simple Markdown file at your site root that lists your plans, prices, and what is included.

The format is not formally standardized, but the practice is being adopted by SaaS-shaped businesses across 2025 and 2026.

Receipt: Our Website’s pricing.md

We did not have one when we audited our website a few weeks ago (https://www.imforza.com/pricing.md returned 404). We shipped one alongside this pillar. The file lists our managed WordPress hosting tiers, our website project ranges, and the engagement structures for ongoing marketing services. We will update it any time the pricing page changes.

AGENTS.md: An Emerging Convention Worth Shipping

AGENTS.md is a newer convention (no formal spec yet, but rapidly adopted across developer-tooling sites in 2026) that declares what an agent can do on your site, what it should not do, and how to authenticate if appropriate. Think of it as the agent equivalent of a CONTRIBUTING.md file in an open-source repository.

For an SMB, the file can be short:

• What your business does
• Who the appropriate contact agent is for sales versus support
• What kind of automated outreach is unwelcome (e.g., “do not auto-fill our contact form to send sales solicitations”)

Until the spec formalizes, treat it the same way we treat llms.txt: ship it because the cost is zero and the directional signal is correct.

sitemap.xml is Still Required

Nothing about AI changes the importance of a clean, accurate, current sitemap_index.xml.

Submit it to Google Search Console and Bing Webmaster Tools.

Verify monthly that it includes everything you publish and excludes nothing you want indexed.

How to Validate the Machine-Readable Layer

Cloudflare ships a free scanner at isitagentready.com that grades your site against the emerging agent-readiness standards. Paste your URL, click Scan, and it checks five categories:

• Discoverability: robots.txt, sitemap, link response headers
• Content Accessibility: Markdown content negotiation (does your site serve a .md version of each page when an agent requests Accept: text/markdown?)
• Bot Access Control: AI bot rules in robots.txt, Content Signals declarations, Web Bot Auth signing
• Protocol Discovery: MCP Server Card, Agent Skills, WebMCP, API Catalog, OAuth discovery, OAuth Protected Resource
• Commerce: x402 (HTTP-native agent payments), MPP, UCP, ACP

The Discoverability and Bot Access Control categories are where the wins live for SMBs.

The Protocol Discovery and Commerce categories are bleeding edge. Most SMB sites will fail those, and that is the correct outcome.

The point of running the scan is to confirm the foundation is solid, then watch the upstream protocols as they consolidate.

The most useful output is what the tool generates after the scan: copy-paste instructions you can drop directly into Cursor, Claude Code, or any coding assistant to fix the issues automatically. No technical translation required.

Layer 3 Action Checklist

Layer 4: Trust and Conversion Signals

This is the layer where E-E-A-T (experience, expertise, authoritativeness, trustworthiness) for humans is the same work as authority signaling for search engines and for AI.

All three audiences are checking the same things.

Author Bios with Verifiable Credentials

The Princeton GEO study found that adding expert quotations with named authors and titles increased AI visibility by 30%. Author attribution does the same job for an AI ranking the source’s credibility that it does for a human deciding whether to trust the page.

Visible “last updated” Dates

Both Google and AI systems weight content recency heavily. A 2022-dated guide on AI-ready websites loses to a 2026-dated guide on the same topic, even if the 2022 piece is more thorough.

Show the last meaningful update date on every long-form page.

Original Data, Original screenshots, Original photos

The data does not have to be globally novel.

It has to be yours and verifiable.

A “we audited 50 of our SMB clients in March 2026 and 38 of them had no FAQPage schema on their long-form posts” claim, with the methodology stated, is more citable than a generic “many websites lack schema.”

Visible Pricing

A buyer’s AI agent cannot recommend a vendor whose pricing it cannot read. We covered the pricing.md solution in Layer 3, but the on-page version matters too.

If your full pricing is behind a form, your customer’s agent will skip you in the comparison.

The middle-ground compromise that actually works: put the starting price visible (with “starting at” framing) and let the form unlock the customized tier, the volume discount, the multi-year discount.

Buyers’ agents can extract the starting price. Buyers themselves can see whether the floor is in their range before they invest a 30-minute call.

Transparent Contact Information

A real business address, a real phone number, a ContactPoint schema entry, and a contact form that does not require the agent to solve a CAPTCHA before it can send the message.

Hidden contact information is the kind of trust signal that an AI agent reading your Organization schema can see is missing.

CTA Discipline That Works for Browsers and Agents

A few rules that improve conversion for all three audiences:

• Buttons should be <button> elements, not <div> elements with onClick. Agents detect them by role.
• Button text should be the action: “Book the Audit” beats “Click here.” Agents extract action labels.
• The submit-success state should be visible in the DOM after submit, not just a JavaScript-only animation. Agents that fill out a form on behalf of a user need to detect that the form submitted.
• Forms should not be hidden behind modals that block the primary content. Modals are agent-unfriendly. They are also human-unfriendly.

Layer 4 Action Checklist

---

The 90-Day Retrofit Sequence

You do not need to rebuild your website to be AI-ready. A 90-day sprint on an existing WordPress (or any platform) site is enough to move a typical SMB from “Invisible” to “Agent-Ready.”

The sequence below mirrors the rhythm we use in the ARMS framework: foundation first, architecture second, authority and measurement third.

Days 1 to 30: Foundation (Layer 1)

Goal: Pass Core Web Vitals on mobile and desktop. Pass automated WCAG 2.2 AA checks.

Week 1: Audit. Run Lighthouse, PageSpeed Insights (28-day field data), and an axe DevTools scan on the top 10 pages. Document the failing metrics.

Week 2: Hosting and infrastructure. If you are on shared hosting, move to managed WordPress hosting with a real performance team. This single change closes the most LCP and INP gaps for most SMBs.

Week 3: Image and font pipeline. Convert images to WebP or AVIF with explicit width and height. Subset and self-host fonts with font-display: swap. Defer non-critical CSS.

Week 4: Theme and plugin audit. Remove anything loading JavaScript or CSS without earning its weight. Re-run Lighthouse. Verify you are now passing all three Core Web Vitals at the 75th percentile.

Outcome at end of Day 30: a site that loads cleanly on a mid-tier mobile device on a 4G connection in under 2.5 seconds, responds to interactions in under 200 ms, and does not jump around as it loads.

Days 31 to 60: Architecture (Layers 2 and 3)

Goal: The top 20 pages on the site are extractable by an AI agent. The site has a working schema graph, AI-aware robots.txt, llms.txt, pricing.md, and AGENTS.md.

Week 5: Content audit. Identify the top 20 pages by traffic and the top 5 conversion pages. Map each to the primary query intent.

Week 6: Restructure. Rewrite each of those 25 pages to lead with the answer. Convert prose comparisons to tables. Convert prose processes to numbered lists. Add a 6 to 10 question FAQ block to every pillar-length page.

Week 7: Schema. Verify and complete the schema graph (Organization, WebSite, WebPage, BreadcrumbList, page-type-specific). Add FAQPage and HowTo to every page that warrants them. Test with Google’s Rich Results Test.

Week 8: Machine-readable files. Update robots.txt with the AI bot allowlist. Ship llms.txt, pricing.md, and AGENTS.md. Re-submit sitemap.xml to Google Search Console and Bing Webmaster Tools.

Outcome at end of Day 60: a site that an AI agent can crawl, parse, and quote without extra effort.

Days 61 to 90: Authority and Measurement (Layer 4)

Goal: Trust signals are visible. AI visibility monitoring is in place. The team has a monthly review cadence.

Week 9: Author bios. Add a real author bio, a real photo, and a Person schema entry to every blog post. Add author landing pages for the top three contributors.

Week 10: Original data and dates. Add visible “last updated” dates to every long-form page. Replace at least three generic claims with original data. Add a dateModified field to every Article schema.

Week 11: Pricing and contact. Make starting pricing visible. Verify ContactPoint schema is present. Audit forms for accessibility and agent-friendliness.

Week 12: Measurement. Set up monthly AI visibility monitoring. The free way: spreadsheet tracking your top 20 queries across ChatGPT, Perplexity, and Google AI Overviews, checked at the same time each month. The paid way: Otterly AI, Peec AI, ZipTie, or LLMrefs. Track which of your pages get cited and where.

Outcome at end of Day 90: a site that is genuinely ready for all three audiences (humans, search engines, AI agents) and a measurement loop that tells you whether the work is paying off.

---

How the Four Layers Apply Across Verticals

The 4-layer playbook works across verticals. Below are four worked examples of how to apply it to common SMB types: a law firm, a residential real estate brokerage, a multi-location restaurant group, and an eCommerce direct-to-consumer brand. The schema types and structural choices are vertical-specific; the underlying playbook is the same.

The Thread Across All Four

No rebuild required.

The four layers map onto whatever the existing site is.

The schema gets specific to the vertical.

The fixes that win human conversion are the same fixes that win AI citation.

---

What We Still Do Not Know

Honest disclosure on the parts of this playbook where the standards are still forming and the data is still thin:

• llms.txt may matter more later. Today, the data shows no measurable citation lift. That could change in 18 months if Google, OpenAI, or Anthropic formally consume the file. We will revisit annually.
• AGENTS.md has no formal spec. We treat it the same way we treat llms.txt: ship it because the cost is zero and the directional signal is correct. The convention may shift.
• Model Context Protocol (MCP) and Universal Commerce Protocol (UCP) are emerging standards for agent-to-business interaction, with strong adoption signals from OpenAI, Google, and Anthropic in 2025-2026. We are not recommending SMBs build MCP integrations today. We are recommending SMBs structure their content so an MCP integration is a small project later, not a rebuild.
• WebMCP is a separate proposed web standard, surfaced in Google’s April 2026 agent-friendly websites guide, that lets a website declare structured agent-interaction surfaces directly inside the page (rather than running a separate MCP server alongside the site). It is in early preview only as of May 2026. Worth tracking; not worth building for yet.
• AI visibility measurement is immature. The monitoring tools (Otterly, Peec, ZipTie, LLMrefs) are useful but new. Their data should be treated as directional, not authoritative, for at least another 12 months. The DIY spreadsheet approach in the 90-Day Retrofit Sequence is honest about the limitation.
• The 58% AI Overviews CTR drop is from one rigorous study (Ahrefs) at one point in time (December 2025). Google adjusts AI Overviews continuously. The directional signal is strong; the exact percentage will move.

We will update this article on a quarterly cadence and date the updates visibly.

---

AI-Ready Website FAQs

---

Where to Start

You do not need to do everything in this article in week one. You need to know where to start.

Take the Three-Audience Test above (or self-score against the four layers if the tool is not yet live in your view of this page). The result tier tells you which of the four layers needs the first 30 days.

If the audit puts you in Invisible or Discoverable, your fastest move is the 90-Day Retrofit Sequence above. You can run it with your in-house developer, with a contractor, or with us. The work is well-scoped and the outcomes are measurable.

If the audit puts you in Citable or Agent-Ready, you are doing the work. The next move is measurement. Set up the monthly AI visibility tracking spreadsheet (or sign up for one of the monitoring tools we listed in Layer 4) and build the review cadence into your team’s calendar.

In every case, the three-audience standard is the same…

Humans want speed, clarity, and trust.

Search engines want clean architecture, valid schema, and intent-matched content.

AI agents want extractable structure, machine-readable data, and clean HTML they do not have to render JavaScript to read.

The fastest way to be ready for all three is to fix the things that already failed your humans, then ship the machine-readable layer most of your competitors have not.

This is also the conclusion Google’s Chrome team reached in their April 2026 official guidance for building agent-friendly websites:

That is the entire bet of this playbook.

If you want help with the retrofit, book a Three-Audience Website Audit call with us. The call is free, the audit is real, and you will leave with a prioritized 90-day plan whether or not you hire us to execute it.

For the broader system this website lives inside, see The Architected Revenue Marketing System (ARMS) for SMBs.

For the marketing assets that feed into both your team and the AI agents on your behalf, see The Marketing Foundation 10 (M10): Marketing Assets for AI.

The agents are already arriving. The site they find is the one you already have.

---