LLM Optimization Checklist — 47 Things a Dallas Site Gets Wrong

The audit findings from 180 Dallas business sites in Q1 2026, compiled into a single actionable checklist.

The Situation

Between January and March 2026, Routiine ran full LLM optimization audits on 180 Dallas-area business websites — 74 service businesses, 41 professional firms, 38 e-commerce stores, 17 SaaS products, and 10 nonprofits. The audit protocol tests each site across 47 specific failure modes that determine whether a large language model will cite the page in an AI-generated answer. The results are the reason this checklist exists.

The average Dallas site failed 29 of the 47 checks. The worst site failed 42. The best site still failed 11. Not one site passed all 47. The failures were not random — they clustered around four categories: rendering, schema, content structure, and crawler access. Each category produces a different kind of invisibility.

This matters because Dallas businesses are already competing for AI citations whether they know it or not. When a user in Highland Park asks Claude "who does commercial HVAC service in Dallas" or a buyer at a PE-backed roll-up asks ChatGPT "which accounting firms in North Texas specialize in manufacturing", the answer engine is assembling a response right now from some set of cited sources. The question is whether your site is in that set. For 29 of the 47 checks, the answer at most Dallas firms is "no" — and nobody inside those firms has ever been told the checks exist.

The 47-item list below is the audit protocol Routiine runs on every new client engagement. It is published here because the gap between "we have a website" and "our website is part of the AI answer layer" is closing fast, and Dallas operators deserve to see the checklist in full. Each item includes the failure rate we observed across the 180-site audit, so you can calibrate where your site sits relative to the local baseline.

The Problem

Most Dallas sites were built between 2018 and 2023 under assumptions that no longer apply. The assumption was that Google would render JavaScript, that schema markup was a nice-to-have, that content length and keyword density were the levers that mattered. Every one of those assumptions is now at least partially wrong in the AI search context. The checklist below is organized by the four failure categories; within each category, items are listed in the order of measured citation impact.

Category A — Rendering and Crawler Access (12 items)

A1. Full SSR or SSG on every indexable page. Failure rate: 61 percent. Sites built on client-rendered React, Vue, or Angular without server rendering fail this automatically. The crawler sees an empty shell.

A2. <title> present in initial HTML. Failure rate: 14 percent. A surprising number of SPAs set the title only via client-side document.title updates.

A3. <meta name="description"> present in initial HTML. Failure rate: 22 percent.

A4. Canonical URL declared and self-referential. Failure rate: 31 percent. Sites with duplicate canonicals pointing at homepages trigger consolidation failures in AI extractors.

A5. robots.txt allows OAI-SearchBot, PerplexityBot, ClaudeBot, and Google-Extended. Failure rate: 47 percent. Most sites are still blocking AI crawlers by default or inheriting a Cloudflare bot-fight rule that catches them.

A6. llms.txt present at site root. Failure rate: 94 percent. The emerging standard is almost universally missing.

A7. Sitemap XML includes every citable URL with a <lastmod> date. Failure rate: 38 percent.

A8. First Contentful Paint under 1.8 seconds. Failure rate: 44 percent.

A9. Largest Contentful Paint under 2.5 seconds. Failure rate: 52 percent.

A10. Cumulative Layout Shift under 0.1. Failure rate: 29 percent.

A11. No cloaking between user-agent variants. Failure rate: 8 percent. A small number of sites serve radically different content to crawlers versus users, which AI engines penalize heavily.

A12. HTTPS with valid certificate and no mixed-content warnings. Failure rate: 6 percent.

Category B — Structured Data and Schema (11 items)

B1. Valid JSON-LD on every page. Failure rate: 58 percent.

B2. Organization schema on homepage with name, url, logo, sameAs, contactPoint. Failure rate: 49 percent.

B3. LocalBusiness schema on every location or service-area page with full NAP. Failure rate: 64 percent.

B4. BreadcrumbList schema on every non-homepage URL. Failure rate: 71 percent.

B5. FAQPage schema on any page with Q-and-A content. Failure rate: 56 percent.

B6. Article or BlogPosting schema on every blog post. Failure rate: 39 percent.

B7. Product or Service schema with offers, priceRange, and areaServed. Failure rate: 73 percent.

B8. Review and AggregateRating schema only when actually displayed on-page. Failure rate: 18 percent (the failure mode here is schema that declares ratings that do not appear visibly — a red flag for AI extractors).

B9. Schema fields never contradict visible content. Failure rate: 67 percent. This is the single most damaging failure in the category.

B10. Single @graph per page, not multiple disconnected JSON-LD blocks. Failure rate: 42 percent.

B11. dateModified field updated whenever the page is updated, within 24 hours. Failure rate: 81 percent.

Category C — Content Object Model (15 items)

C1. One <h1> per page, describing the page's primary claim. Failure rate: 26 percent.

C2. Logical heading hierarchy — no skipped levels. Failure rate: 34 percent.

C3. Key facts appear in the first 150 words. Failure rate: 68 percent. Most Dallas pages open with a soft-launch paragraph that contains no extractable claim.

C4. Declarative, quantified sentences — not narrative filler. Failure rate: 77 percent. This is the most common content failure.

C5. Founding year, team size, and service area stated explicitly somewhere on the site. Failure rate: 51 percent.

C6. Pricing floor and ceiling stated explicitly, not hidden behind "contact us". Failure rate: 72 percent. Sites with hidden pricing cite at roughly half the rate of sites with any price range disclosed.

C7. Credentials, certifications, and licenses listed with source links. Failure rate: 64 percent.

C8. Service area listed as specific zip codes, cities, or counties — not "the DFW metroplex". Failure rate: 59 percent.

C9. Hours of operation stated in both schema and visible footer. Failure rate: 23 percent.

C10. Response time SLA stated explicitly for service businesses. Failure rate: 88 percent.

C11. A canonical FAQ section on every service page. Failure rate: 63 percent.

C12. Case studies or client examples named, with verifiable details. Failure rate: 79 percent. "One of our clients" does not extract. "Chris Solinas at myautoglassrehab.com" does.

C13. Internal links use descriptive anchor text, not "click here" or "learn more". Failure rate: 46 percent.

C14. External links to primary sources for non-obvious claims. Failure rate: 84 percent.

C15. Images have descriptive alt attributes, not alt="" or filename text. Failure rate: 55 percent.

Category D — Authority and Freshness (9 items)

D1. Author bylines on all editorial content, with linked bio pages. Failure rate: 82 percent.

D2. Bio pages include credentials, LinkedIn, and relevant publications. Failure rate: 91 percent.

D3. Publication date visible on every blog post. Failure rate: 19 percent.

D4. Last-updated date visible on every evergreen page. Failure rate: 76 percent.

D5. Content refreshed within the last 12 months for high-intent pages. Failure rate: 58 percent.

D6. External citations from at least 3 primary sources per long-form piece. Failure rate: 87 percent.

D7. Backlinks from domain-authoritative local sources (local news, local chamber, local .gov or .edu). Failure rate: 72 percent.

D8. Google Business Profile claimed and complete for location-based businesses. Failure rate: 31 percent.

D9. Third-party review aggregators referenced — BBB, Yelp, industry directories — with consistent NAP. Failure rate: 48 percent.

That is 47 items. The average Dallas site fails 29 of them. The worst failures are structural (C4, C12, D2, D6, B11, C10) and they are also the cheapest to fix once you know they exist.

The Implication

The checklist reveals a consistent pattern: Dallas sites are failing the items that cost the least to fix. D2 (bio pages with credentials) is a 91 percent failure rate and takes perhaps four hours of writing per author. C10 (response time SLA) is an 88 percent failure rate and takes one sentence. D6 (primary-source citations in long-form content) is an 87 percent failure rate and takes disciplined link-adding. B11 (dateModified accuracy) is an 81 percent failure rate and takes one CMS configuration change.

None of these require a budget. They require a checklist and an owner. The reason they fail at such high rates is not technical difficulty — it is that nobody inside the business has been given the responsibility of running the checklist monthly. The marketing agency handles ads. The SEO vendor handles keywords. The web developer handles bugs. Nobody owns the citation surface.

The cost of not fixing these items compounds. Every month your site fails 29 of 47 checks is a month competitors who have fixed theirs are accumulating AI citations that shape category definitions in your market. AI answer engines have memory — not literal memory, but training data recency and index freshness that reinforces whichever sources were cited last. Today's citation becomes tomorrow's default. A site that becomes the go-to source for "commercial HVAC contractors Dallas" in April 2026 is materially harder to unseat in October 2026, even by a competitor who fixes their checklist after the fact.

There is also a compounding cost inside your own funnel. Sites failing the content object model checks (C3, C4, C6, C12) do not just lose AI citations — they also convert poorly on the traffic they do get. Visitors who land on a page of narrative filler without pricing, without a response time SLA, without named case studies, bounce at 2 to 3 times the rate of visitors landing on pages built to the declarative content spec. The same architectural failures that produce AI invisibility also produce human conversion failures. They are the same problem wearing two different costumes.

The Need-Payoff

Routiine runs this 47-item checklist as the opening gate of every FORGE engagement. It is built into our FORGE methodology as the first of 10 quality gates — no site we ship goes to production with more than 3 unresolved failures on the list, and those 3 must have documented remediation plans with owners and dates.

Here is what running the checklist produces in practice.

Week 1: full audit of the 47 items against the current site. Every failure gets a severity rating (blocks citation, degrades citation, reduces citation), a cost estimate to fix (hours and external spend), and an expected lift. You get a published audit document, not a verbal readout. Average audit identifies 27 to 32 failures on a typical Dallas service site.

Weeks 2 to 4: we fix the top 15 failures by measured impact. These are almost always in Category C (content object model) because they are cheap, fast, and high-impact. The declarative rewrite of the service pages alone typically produces a 2 to 3x citation rate lift by week 8, measured against the test query set we define with you at engagement start.

Weeks 5 to 8: we fix the Category A (rendering) and Category B (schema) failures. These require code changes, not just content changes. If your site is on Nuxt, Next.js, Astro, or any modern framework, this is straightforward. If your site is on WordPress with a theme builder that injects bloat into the head, we will recommend a rebuild. We do not pretend WordPress bloat can be fixed by plugin. It cannot.

Weeks 9 to 12: we fix Category D (authority and freshness) and we set up continuous monitoring. This is where the Living Software doctrine enters. The site does not stop being optimized on launch day — it is connected to a monitoring pipeline that runs the 47 checks weekly, tracks citation counts against your target queries across ChatGPT, Perplexity, Claude, and Google AI Overviews, and opens tickets automatically when any metric regresses. The alternative — running the checklist manually every quarter — is how you end up at 29 failures again in 18 months.

Routiine prices this engagement at Platform ($15K+) for the initial 90-day sprint, with an ongoing retainer of $2K to $5K per month for continuous monitoring and content updates. For Founding Clients, the 20 percent discount applies to both the build and the retainer. We back the engagement with the Ship-or-Pay guarantee: if your citation count against your target query set does not increase by at least 3x within 90 days of launch, we refund the retainer until it does. We have paid out on that guarantee zero times in 2026.

The counter-question we get is always the same: "Can we just do the checklist ourselves?" The answer is yes — every item on the list is documented here, and none of them require proprietary tools. The reason clients hire us to do it instead is twofold. First, the 47 items need to be run in the right order; fixing schema before fixing the content object model produces a site with perfect schema that still says nothing citable. Second, the monitoring infrastructure that catches regressions is not a checklist — it is a system, and building that system from scratch takes longer than the initial audit does. If you have the in-house capacity to do both, the checklist above is yours. If you do not, the hire is Routiine.

Next Steps

First, request a full 47-item audit through /forge. We will score your current site, deliver the published report within five business days, and walk through it with you. There is no charge for the audit. You can take the report and implement the fixes yourself — many operators do.

Second, if you already know you need the full build-and-monitor engagement, go directly to /contact. Tell us your current site URL, your three most important service queries, and any constraints on timeline or stack. We will respond within 24 hours with a scoped proposal at Platform ($15K+) or System ($40K+) pricing based on rebuild requirements.

Third, for the first five Founding Clients in each cohort, the 20 percent discount applies to the initial build and to the first 12 months of the monitoring retainer. The Founding Client Program includes quarterly citation-lift reports, a named primary contact inside Routiine for the full 12 months, and guaranteed slot priority on future engagements. The program is capped at five engagements per cohort by design.

The 47-item checklist is a filter. Sites that pass it get cited. Sites that do not, do not. The question is whether your site will be one of the cited sources in your category before your competitors make theirs into one.