We Audited 5 Enterprise Websites Built by Top Web Agencies. Not One Was Ready for AI.

Two law firms with nothing in common landed on identical AI visibility scores last week. Both 7 out of 12. Both Blurry.

One is a national class-action firm with offices in multiple U.S. cities and a roster of partners listed across the major peer-review directories. The other is a personal injury firm with a coast-to-coast presence and one of the most-recognized brand names in the category.

The only thing they share is their vendor. Both sites are built on the same WordPress theme directory: PaperStreet.

A third PaperStreet site we audited the same morning came in lower, at 5 out of 12. A Scorpion HVAC client scored 9 out of 12 but failed on the signals that actually drive AI citation. A Ford dealership on Dealer.com scored 1 out of 12. Ghost.

Five websites. Three industries. Three of the biggest names in agency web. Not one of them was ready for AI.

This article shows you what we found, why it’s structural rather than incidental, and why every month a business waits to rebuild for AI is a month its competitors compound advantage that gets harder to close.

“I Just Rebuilt My Website”

This is the most common objection we hear, and it is killing more businesses than any other single belief in the market right now.

Most of the SMBs we audit have rebuilt their website in the last one to three years. They paid an agency anywhere from a few thousand to well into six figures. They rebuilt for branding. They rebuilt for mobile. They rebuilt because their old CRM didn’t sync with their new lead forms. They rebuilt because their last designer abandoned them.

They did not rebuild for AI.

“Rebuilt for AI” is a specific technical bar: the one AI platforms now use to decide which businesses to cite, recommend, and transact through. Hitting that bar requires structural changes template agencies do not deliver, even at premium pricing. The five audits below are the evidence.

What “Built for AI” Actually Means

AI search engines do not read web pages the way search crawlers did in 2018. They parse pages into entities: structured representations of businesses, services, people, places, ratings. Then they decide which entities to surface based on how confidently each one is described and how heavily it is corroborated by independent sources.

A website built for AI does five things at the structural level:

1. Every service is its own entity. Not a paragraph in a wall of brand prose. A typed Service or LegalService node, with description, area served, and provider reference, attached to a dedicated page with the depth AI uses to extract specifics.
2. Every key person is its own entity. Attorneys, doctors, founders, technicians: each represented as a Person node with role, credentials, and verifiable identifiers, not just a headshot and a bio paragraph.
3. Reviews are structured. AggregateRating and Review nodes tied to the business entity, sourced from verifiable platforms. Star counts on a graphic do nothing for AI. The data has to be machine-readable and entity-linked.
4. The business is connected to itself across the web. A sameAs property linking the business entity to its Google Business Profile, LinkedIn, social profiles, professional directories, and any knowledge base entries. This is the signal AI uses to resolve identity when the same name exists in multiple markets.
5. Depth matches the category leaders. Practice or service pages need the explanatory depth that the top-cited businesses in the category already publish. A 200-word service page cannot compete for citation against a 1,400-word version of the same page on a competitor’s site.

Schema markup is one piece of the stack. A site that bolts JSON-LD onto a template that wasn’t built around entities is the equivalent of writing a contract in English and stapling a French translation to the back. The translation is there. The structure is wrong.

Exhibit 1: A Scorpion HVAC Client: 9/12 Visible (and Still Failing)

We audited the website of a U.S. HVAC contractor whose digital marketing is managed by Scorpion. The site scored 9 out of 12 on our Free AI Audit, placing it in the Visible tier (8–12).

The headline score reads like a pass. The detail rows do not. The site failed on the three signals that matter most for commercial AI queries:

• Services defined. AI does not know what services you provide. An HVAC contractor that AI cannot describe as an HVAC contractor cannot be cited when a homeowner asks an AI assistant for “the best HVAC company near me.” That is the entire commercial point of being on the internet in this category.
• Reviews or ratings. AI finds no reviews or ratings. The business has plenty of reviews. AI cannot read them because the entity linking between the business node and the review platform is broken.
• Social profiles linked. AI cannot verify you across other platforms. The sameAs property is missing or empty. The site fails the basic identity-resolution check.

Compounding this: Google PageSpeed Insights returned a Failed Core Web Vitals assessment for mobile, with Largest Contentful Paint at 2.8 seconds, First Contentful Paint at 2.2 seconds, and Time to First Byte at 1.5 seconds. Heavy template overhead shows up exactly this way.

A 9/12 Visible score that fails on services, reviews, and identity verification is functionally Ghost for AI citation in the HVAC category. The site is technically present, commercially invisible.

Exhibit 2: A Dealer.com Ford Dealership: 1/12 Ghost

A Ford dealer built on Dealer.com. That’s Cox Automotive’s dealer platform. The audit gave the site 1 out of 12. Ghost.

Dealer.com is the dominant website platform in U.S. automotive retail. A Ford franchise pays for the Dealer.com platform itself, then layers digital marketing, inventory feed integrations, and dealer-portal compliance fees on top. The total monthly spend on a dealer’s web presence routinely runs in the five figures. The output of that spend, measured against the standard AI platforms now use to decide which businesses to recommend: 1 out of 12.

The dealer is not free to leave. Franchise programs enforced through FordDirect require dealers to operate on an approved-vendor list. Co-op funds are tied to platform compliance. Switching platforms means losing program participation. The dealer is structurally locked in to a platform that does not deliver AI readiness.

This is the largest tier of consumer purchase intent in the U.S., served by the dominant platform, scoring Ghost.

Exhibit 3: Three PaperStreet Law Firm Sites: The Template Ceiling, Measured

PaperStreet is one of the most established website agencies in the legal vertical. They state they have built over 2,500 websites over the past 25 years for law firm clients spanning class action, personal injury, intellectual property, family, and corporate practice. We audited three of their most expensive Enterprise websites for large law firms. All three returned scores in the Blurry tier (4–7).

Inspecting the raw HTML revealed why. All three sites are built on the same WordPress theme directory: /wp-content/themes/paperstreet. Every site shipped exactly one JSON-LD block. Every block contained the same schema scaffold: Organization, WebSite, WebPage, BreadcrumbList, SearchAction. No site contained any of the following:

• LegalService schema for any practice area
• Person schema for any attorney, including, on one site, the founder whose name is on the door
• AggregateRating or Review schema
• A populated sameAs property linking the firm to its profiles across the web. On one of the three sites, the sameAs array was literally empty ("sameAs": []), meaning the template generates the field and the agency’s onboarding process never fills it.

The 7/7 tie between the two largest firms is the most important finding in this article. Two firms with nothing in common except their vendor landed on identical scores on identical dates. That is a measurement of template ceiling. PaperStreet’s WordPress theme delivers approximately 7 out of 12 worth of AI signal out of the box, and the per-firm work required to push past Blurry into Visible does not happen as part of the agency’s standard delivery.

Why This Is Structural, Not Incidental

These agencies are constrained by their own business model.

Template agencies operate on margin economics that depend on shared scaffolding. The unit economics of a $5,000-to-$15,000-per-month managed service work only when 70-to-80% of the build is reused across clients. Each new client gets the same theme, the same module library, the same content templates, the same schema scaffold. The agency’s content team writes practice-area copy, populates contact information, and ships.

The per-firm work required to be cited by AI cannot be templated: LegalService schema scoped to each practice area, Person schema for each attorney, AggregateRating tied to verified review platforms, hand-curated sameAs linking, page-depth matching the cited competitors in the category. It is per-client custom engineering. Delivering it across hundreds or thousands of accounts at template-agency pricing would break the model.

The failure replicates across firms because the template is the same. Two firms with nothing else in common share an audit score because they share a vendor. And no schema plugin, no dashboard checkbox, and no “AI optimization add-on” can close the gap. The architecture itself is the bottleneck.

The “We Added Schema” Trap

Every major template agency now sells an AI optimization service. The pitch is usually the same: we will add schema to your pages, and AI platforms will start citing you.

The single best controlled study on this question was published by Ahrefs on May 11, 2026. They tracked 1,885 pages that added JSON-LD schema between August 2025 and March 2026, matched against 4,000 control pages over the same window. The result: Google AI Mode citations moved +2.4%, ChatGPT moved +2.2%, both within statistical noise. Google AI Overviews citations declined 4.6%, a result Ahrefs reports as statistically significant.

The conclusion: adding schema markup to a page that wasn’t built for AI citation does not move the citation outcome. Schema is necessary infrastructure for AI to parse a business cleanly. By itself, it doesn’t move citations. The lever is the full signal stack: entity clarity, third-party authority, machine-readable depth, and freshness, with schema embedded inside it.

A bolted-on schema plugin on top of a template that wasn’t built for entity organization is exactly the scenario Ahrefs tested. It does not work. Our PaperStreet audits show the same thing from the other direction: even with a baseline schema scaffold shipping, three sites tied or near-tied at Blurry.

Why Waiting Makes the Gap Wider

AI citation isn’t a one-shot contest you can join whenever you’re ready. It’s a feedback system. Whichever businesses get cited build a track record on the platforms doing the citing. Whichever ones get skipped fall further out of the pattern every month.

Stacker’s March 2026 study ran 87 pieces of content across 30 clients through 2,600+ prompts on 8 AI platforms. Brands that pushed content through third-party editorial channels saw a median 239% lift in AI citations versus brand-owned content alone. Citations create more citations.

Google’s June 2025 Knowledge Graph cleanup removed over 3 billion ambiguous entities, a 6.26% contraction in the world’s largest structured business database. The system simultaneously promoted entities with high-confidence classifications. Google is rewarding businesses with established entity confidence and cutting the ones it can’t pin down.

Every month a business stays Blurry or Ghost, its competitors compound citations on the platforms that will cite them again next quarter, next year, and in every model retrain after that. There is no neutral month. Waiting is choosing to lose ground.

What an AI-Ready Website Actually Looks Like

A site rebuilt for AI is built around the entities AI extracts. Page architecture, schema graph, and content depth are designed for citation from day one.

At the page level, each commercial intent gets its own entity: a service, a practice area, a product line, with the depth, structured properties, and supporting content needed for AI to confidently cite that page when the matching query is asked. At the firm level, every key person is its own entity. Reviews are pulled from verified platforms and tied to the business entity through AggregateRating. The business is connected to its own footprint across the web through populated sameAs linking. Page hierarchy is sequential and parseable. None of these are exotic technical asks. They are the table stakes of a build done for AI rather than for branding.

This is what EntitySeal builds. Our Website Design service delivers AI-ready websites engineered around the entity model AI platforms now use to decide who gets cited. Each service has its own page and its own schema node. Each key person is its own entity. Reviews, ratings, and identity links are wired into the build. The result clears the 12-point audit at the Visible tier (8-12), and the structural decisions hold up under future model retrains.

The site is half the work. AI citation also requires a third-party authority layer: structured editorial profiles on independent platforms AI already cites. EntitySeal deploys this through the EntitySeal Editorial Network as part of the same engagement. The site work and the editorial layer compound on each other. One without the other leaves citations on the table.

See Where Your Site Stands: Free, 60 Seconds, No Signup

Before you decide whether your last rebuild was for AI or for something else, see your score. Our Free AI Audit runs the same 12-point diagnostic against your site that we ran against the five sites above. You will see exactly where your business falls (Ghost 0-3, Blurry 4-7, or Visible 8-12) and which signals are failing.

If you score Blurry or Ghost, the next decision is whether the next dollar you spend goes to the same template economy that produced the result, or to the kind of rebuild the score actually requires.

The Cost of Waiting

The five sites we audited represent the dominant template agencies in three of the largest service categories in the U.S. economy. None produced an AI-ready build at the level the audit measures. None will, until the underlying business model changes, and the business model has every incentive to keep things as they are.

If your last rebuild produced a Blurry or Ghost result, the most expensive option available to you is the option most business owners pick by default: wait. Wait for the agency to figure it out. Wait until you have more budget. Wait until you see what competitors do. Every month that passes, the AI citation infrastructure becomes more locked in, the entity-density gap between you and your competitors widens, and the cost of closing it rises.

The audit takes 60 seconds. The conversation that comes after it is the one that matters.

Frequently Asked Questions

I just rebuilt my website. Why isn’t that enough for AI?

Most website rebuilds in the last three years were done for branding, mobile responsiveness, lead capture, or CMS migration. Building for AI is a structural target: entity-organized pages, per-service and per-person schema, AggregateRating tied to verified review platforms, populated sameAs linking, and category-matching content depth. Most agency rebuilds, including those at premium price points, do not deliver any of these. The Free AI Audit shows you which signals your current site is passing and failing.

Does adding schema markup make my site AI-ready?

On its own, no. Ahrefs published a controlled study on May 11, 2026 tracking 1,885 pages that added JSON-LD schema between August 2025 and March 2026, against 4,000 control pages. Citation deltas were +2.4% on Google AI Mode and +2.2% on ChatGPT (both within statistical noise), and Google AI Overviews citations actually declined 4.6%, a statistically significant result. Schema is necessary for AI to parse a business cleanly, but adding it to a template not built around entities does not produce a citation lift on its own.

Can my web agency build an AI-ready website?

Most cannot, at their current service model. Template agencies depend on shared scaffolding across hundreds or thousands of accounts. The per-client custom engineering required for AI-grade entity infrastructure is incompatible with template-margin economics: per-service schema, per-person schema, hand-curated entity linking, category-matching page depth. The replicating pattern we observed across our audits is the structural ceiling of the business model.

How do AI platforms decide which businesses to recommend?

AI platforms evaluate a stack of signals before treating a business as a citable candidate: entity clarity (consistent identification across the web), third-party editorial authority (structured profiles on independent sources AI cites), machine-readable content depth, freshness, and structured data that lets AI parse the business as a typed entity. Stacker’s March 2026 study of 2,600+ prompts across 8 AI platforms found brands cited by third-party editorial channels got a median 239% lift in AI citations versus brand-owned content alone. Citation density compounds: the businesses already cited often become the businesses cited next.

How can I tell if my website is invisible to AI?

Run the Free AI Audit. The 12-point diagnostic checks the signals AI platforms actually use to decide whether to cite a business: structured entity data, services definition, identity verification across platforms, reviews and ratings, and basic trust signals. You will be scored Ghost (0-3), Blurry (4-7), or Visible (8-12). The audit takes 60 seconds, costs nothing, and requires no signup.

How soon do I need to rebuild my website for AI?

The AI citation environment is a feedback system. Every month a business stays Blurry or Ghost, its competitors keep accumulating citations on the platforms that will cite them again next quarter. Stacker’s March 2026 study (87 pieces of content, 30 clients, 2,600+ prompts across 8 AI platforms) found brands distributing through third-party editorial channels got a median 239% lift in AI citations versus brand-owned content alone. Citation density itself becomes a moat. The longer a business waits to fix its AI signal, the bigger the gap.