Why ChatGPT Doesn't Cite Your Site (And the Five Fixes)

Your site shows up for your brand name, ranks for your keywords, and still gets zero ChatGPT mentions. Here is why, and five ways to change it.

The Situation

A Dallas business owner opened ChatGPT in March 2026 and typed "who are the best custom software developers in Dallas Texas". The answer came back with four company names, a short description of each, and a footer with three cited sources. His company was not in the answer. His company ranks #4 in Google for that exact query. His company has 120 blog posts, a full portfolio, a Google Business Profile with 87 five-star reviews, and 14 years in business. None of it appeared.

He asked us why. The audit took 90 minutes and produced five specific failures — five structural reasons ChatGPT does not cite his site. Every one of them is fixable. None of them is about SEO in the traditional sense. This is the same audit pattern we have now run on 180 Dallas sites, and the same five failures appear on 89 percent of them.

ChatGPT's citation mechanic is not a ranking algorithm. It does not rank pages against each other and pick the top result. It extracts claims from multiple sources, synthesizes an answer, and cites the sources whose claims it used. A site can rank #1 in Google and still contribute zero claims to the synthesis — because the claims on the page, or the structure containing them, or the access path to the page, is broken in a way that blocks extraction.

The five reasons below are ranked by how often they appear in our audits and by how much each contributes to the null citation outcome. If your site is not being cited, you are almost certainly failing at least three of these five. The good news is that each fix is concrete and measurable.

The Problem

Reason 1: Your pages are walls of prose with no extractable claims.

A 1,200-word "About Us" page that talks about passion, family values, and commitment to quality is not a page ChatGPT can cite. The model cannot extract a claim from "we've been serving the community for decades" because "decades" is not a number, "the community" is not a place, and "serving" is not a quantifiable action. When the user asks "who has been doing this the longest in Dallas", the model needs a founding year. Your page does not have one. The page citing the competitor does. The competitor gets cited.

This is the single most common failure. Across our 180-site audit, 77 percent of pages produced zero extractable claims in the first 500 words. Dallas owners tend to write pages that sound the way they would talk at a chamber mixer — warm, relationship-forward, light on numbers. That voice does not extract.

Reason 2: Your schema markup is missing, broken, or contradicts your content.

If the page says your business was founded in 2011 and the schema markup (if any) says foundingDate: "2014-03-01", the model sees a conflict and downweights both facts. If your schema declares priceRange: "$$$" but the body says "affordable windshield repair," the conflict triggers the same downweight. In our audit, 67 percent of sites had at least one schema-to-body contradiction. The most common contradictions were in openingHoursSpecification (hours in schema disagreeing with the visible footer), areaServed (schema listing different cities than the service page), and aggregateRating (schema declaring ratings that do not appear visibly on the page).

Broken schema is worse than no schema. No schema leaves the model to extract from body text alone. Broken schema causes the model to distrust the body text too.

Reason 3: OpenAI's crawler cannot fully load your site.

OAI-SearchBot is the crawler ChatGPT uses to index the live web for citations. It renders JavaScript inconsistently. Sites built on client-rendered frameworks without server-side rendering deliver an empty HTML shell to the crawler. Pages that depend on lazy loading, infinite scroll, or click-to-reveal content do not expose that content to the crawler. Sites behind Cloudflare with aggressive bot-fight mode block OAI-SearchBot entirely. Sites with robots.txt rules that disallow /* or that explicitly disallow OAI-SearchBot block it by design.

In our audit, 47 percent of sites were blocking at least one major AI crawler via robots.txt or Cloudflare settings, and most of those owners had no idea. Cloudflare enabled AI crawler blocking by default for many customers in 2024 and 2025 as a "content protection" feature. The cost of that default is invisibility to ChatGPT.

Reason 4: Your content is not fresh, not dated, and not attributed.

ChatGPT's training data has a cutoff. For citations pulled from live web search, the model prefers pages with visible recency signals: a publication date, a last-updated date, and a named author with a linked bio. Pages with none of these signals get weighted lower than pages that have them, even when the content is otherwise identical.

82 percent of the Dallas sites we audited had blog posts with no visible author byline. 91 percent of the bylines that did exist linked to no bio page. 76 percent of evergreen pages had no last-updated date. A page that reads as if it was written by "the team" and posted in no particular year gets extracted less often than a page written by a named person with credentials, dated within the last 12 months.

Reason 5: You have no citable primary-source signal.

ChatGPT weights claims more heavily when the citation path leads to a primary source — a government database, an industry certification body, a peer-reviewed publication, a first-party case study with verifiable details. A page that asserts expertise and links to nothing carries less weight than a page that asserts the same expertise and links to a state licensing record. A case study that names the client, the project, the date, and the outcome carries more weight than "one of our clients saw great results."

87 percent of the long-form content we audited had zero external citations. 79 percent of case studies used anonymous stand-ins ("a Dallas manufacturer," "one of our biggest accounts") instead of named clients. This is a cultural pattern in Dallas business writing — an instinct to protect client privacy that has the unintended effect of making every claim unverifiable.

The Implication

These five failures interact. The compounding effect is worse than any one of them in isolation.

A site that has narrative content (failure 1) and no schema (failure 2) and is partially crawl-blocked (failure 3) presents the AI extractor with three distinct problems simultaneously. The extractor cannot reach much of the page. What it can reach has no schema scaffolding. What it can parse is narrative prose with no extractable claims. The result is not partial citation — it is zero citation. The page does not appear in the AI answer at all.

Now consider the business cost. A Dallas custom software firm ranks #4 in Google for its primary keyword. Three years ago, that #4 rank produced roughly 200 qualified leads per year. In 2026, with 38 percent of the upstream demand being served by AI answers the firm does not appear in, that same #4 rank produces 124 qualified leads. The 38 percent of users who asked ChatGPT instead of Google never saw the firm's listing. They called one of the four named companies in the AI answer. The firm's Google rank did not change. Its lead volume dropped by 38 percent anyway.

Scale this across the Dallas service economy and the numbers get uncomfortable. A residential HVAC company doing $6M in revenue and pulling 45 percent of leads from organic channels is looking at roughly $1M in annual revenue now flowing through an AI discovery layer it is absent from. A professional services firm billing $400 per hour is losing 8 to 12 qualified consultations per month that used to flow through organic search and are now resolved inside Perplexity or Claude without the firm's name ever appearing.

The five reasons are also why Dallas firms keep telling us "we did SEO and it didn't work." It did work, for the Google ranking layer it was targeting. It did nothing for the AI citation layer, because nobody explained there was one.

The Need-Payoff

Here are the five fixes, each tied to the reason above, each runnable under our FORGE methodology.

Fix 1 — Rewrite your pages as declarative claim lists. Every service page gets restructured so the first 150 words contain at least five extractable claims: founding year, team size, service area (as specific cities or zip codes), pricing floor, and response time SLA. Every claim appears three times on the page — once in the body, once in a canonical FAQ block, once in structured data. This is the content object model rewrite. It takes one to two weeks per site depending on page count. Measured lift: 2 to 3x citation rate within 60 days.

Fix 2 — Install a clean, validated, contradiction-free schema graph. Every page gets a single JSON-LD @graph block that declares the page type, the organization, and all extractable claims from fix 1. The schema is generated from the same data layer the visible page reads from, so there is no drift. We add BreadcrumbList, FAQPage, LocalBusiness, and Service schemas as appropriate. Validation runs in CI so broken schema cannot ship to production. Measured lift: additional 1.5 to 2x citation rate on top of fix 1.

Fix 3 — Unblock the AI crawlers and ship server-rendered HTML. We audit robots.txt, Cloudflare bot-fight settings, and any WAF rules. We ensure OAI-SearchBot, PerplexityBot, ClaudeBot, and Google-Extended are all allowed. If the site is on a client-rendered framework, we migrate to SSR (usually Nuxt 3 or Next.js) or to static generation. We ship an llms.txt file at the site root with the canonical URL and a one-paragraph declarative summary for every major topic. Measured lift: this is a threshold fix — without it, fixes 1 and 2 cannot reach the model at all.

Fix 4 — Add bylines, dates, and visible author bios to everything. Every editorial page gets a named author, a linked bio with credentials and external references (LinkedIn, publications, certifications), a visible publication date, and a visible last-updated date whenever the page is changed. The dateModified schema field updates within 24 hours of any content change. This is CMS configuration plus one round of writing. Measured lift: 1.3 to 1.8x citation rate within 45 days, compounding with fixes 1 through 3.

Fix 5 — Cite primary sources, and name your clients. Every long-form piece links to at least three primary sources for non-obvious claims. Every case study names the client (with the client's written consent), the project scope, the dates, and the outcome. The case studies become standalone pages with their own schema (CaseStudy or Article) and their own citation surface. Measured lift: harder to isolate but consistently adds 20 to 40 percent citation lift on pages where primary-source density increases.

These five fixes run as a connected sprint under the FORGE methodology. The first pass takes 90 days for a typical Dallas service site. After the first pass, continuous monitoring under the Living Software doctrine keeps the fixes in place and catches regressions — a page where schema quietly drifts, a crawler rule that gets re-enabled after a Cloudflare update, a blog post that ships without an author. Static sites decay. Living Software holds position.

Routiine prices this five-fix engagement at Platform ($15K+) for the 90-day sprint with a continuous monitoring retainer from $2K to $5K per month. For the first five Founding Clients in each cohort, the 20 percent discount applies to both the sprint and the first 12 months of retainer. The engagement is backed by the Ship-or-Pay guarantee — if the citation count on your target query set does not increase by at least 3x within 90 days of launch, we refund the retainer until it does.

The decision here is not whether to fix these five things. It is whether to fix them in 2026 while the AI search transition is still reshuffling positions, or in 2027 after the defaults have calcified. Sites that become the cited sources in their categories in 2026 are materially harder to displace 12 to 18 months later, because the language models reinforce prior citation patterns with every new index pass. The window to establish position is the window we are in right now.

Next Steps

First, request a five-point citation audit at /forge. We will test your site against 15 target queries across ChatGPT, Perplexity, and Claude, score every page against the five failures above, and deliver a published report within five business days. The audit is free. You can take the report and fix the issues yourself.

Second, if you know you want the full sprint, go directly to /contact. Share your site URL, your three most important target queries, and your launch timeline. We respond within 24 hours with a scoped proposal at Launch ($5K+), Platform ($15K+), or System ($40K+) based on site complexity.

Third, the Founding Client Program at /work is capped at five engagements per cohort with 20 percent off the standard pricing for the first 12 months. The program includes quarterly citation reports against your top 20 target queries and guaranteed slot priority on future engagements.

The AI citation layer is being built right now. The sources it trusts in April 2026 are the sources it will keep trusting as the defaults lock in. The fix list is short. The clock is not.