Schema, citations, and source trust: what makes Claude cite you

Claude citations depend on crawlability and mentions on sites Claude already trusts. Pages also need a quotable sentence near the top. The evidence for these factors is stronger than most of the "optimize for AI" advice floating around.

The stakes for a Head of Growth are simple. Forrester's January 2026 Buyers' Journey Survey of roughly 18,000 business buyers found 94% used AI somewhere in their buying process. Most SaaS teams are absent from the conversation their buyers are having.

How Claude finds a page

According to Anthropic's web search documentation, Claude can access real-time web content through a web search tool when a request benefits from current information; the API runs the searches and provides Claude with the results. It queries a search backend, gets structured result blocks, and picks passages from those results. Anthropic has not published its ranking formula.

Per result, Claude sees url, title, page_age, and encrypted_content rather than raw HTML. When it cites you, the cited_text field carries up to 150 characters of your page. The practical unit of "getting cited" is therefore a single sentence Claude can lift and stand behind.

Anthropic runs three crawlers, and its Anthropic crawler guidance draws a line most SaaS teams miss. ClaudeBot collects training data, while Claude-SearchBot indexes pages for search quality and Claude-User fetches pages when a user asks a question.

Site owners can block ClaudeBot while leaving the other two accessible. Blocking the retrieval bots to protect training data can quietly remove a site from Claude's answers.

Does schema help? The evidence says no

Anthropic says nothing about schema in its site-owner guidance. Its documented controls focus on crawler access through robots.txt, including Crawl-delay. Nothing in that guidance connects schema to Claude citation behavior.

A 2026 SIGIR citation study ran 252,000 trials across 18 content factors and found topical relevance and list position were the primary citation drivers. Schema was not among the factors tested. The study leaves schema's effect untested and provides no evidence that JSON-LD improves citation rates.

The available evidence supports what Anthropic's documentation implies: Claude reads search results as text. Keep your schema for Google rich results if it earns them. Don't ship a schema project expecting Claude to notice.

What blocks you before trust even enters the picture

Confirm that Claude can reach and read the page before working on authority or content. Start with four checks:

• Rendering: Serve navigation, product details, comparison copy, and integration information in crawlable HTML. A React or Vue marketing site that leaves essential content to client-side JavaScript can hide that content from retrieval crawlers.
• Edge controls: Audit Cloudflare and any other WAF rules before relying on robots.txt. Confirm that Claude-SearchBot and Claude-User can get through the edge and reach the pages you want cited.
• Search access: Check whether the same pages are available to mainstream search crawlers. If access rules, authentication, or lead-generation modals prevent retrieval, stronger copy will not solve the problem.
• Optional files: Do not let llms.txt distract the team from rendering and crawler access. Indexable HTML is another control tied directly to whether Claude can retrieve a usable passage.

These checks belong at the start of the work, before a content team has rewritten dozens of pages. A page Claude cannot retrieve is unavailable.

The trust signals that correlate with Claude citations

For Claude, prioritize earned media placements on relevant third-party sites over a plan built around social threads alone. A strong placement gives Claude an independent source that can mention your company, explain the category, and point readers toward a page it can retrieve.

Branded mentions and awareness signals help Claude verify your company. Backlinks still have value. The objective should also include surrounding copy that names the company and explains what it does in relation to the buyer's problem.

For B2B queries, product pages are especially useful because they answer the questions buyers ask near a decision: what the product does, which use case it fits, how it compares, what it integrates with, and which option someone should choose. Your pricing tiers and feature matrix should be available in crawlable HTML. The same applies to your integration docs.

Make your company easy to verify across your own product pages and the relevant third-party sites a buyer already trusts.

Writing a page Claude can lift a sentence from

Treat an H2 as a prompt and the next paragraph as the answer. Put the clearest sentence near the top instead of making the reader work through a long introduction before reaching the point.

Claude's cited_text field makes brevity practical. Aim for an answer capsule that can stand on its own in roughly 150 characters. The sentence should name the product or category and state the answer. Include the most useful number or distinction when one is available.

Specific language gives Claude more to work with. Name the tool and the metric. Identify the competitor as well. If the page explains implementation, include the relevant code. If it compares options, state the difference directly rather than burying it under generic benefits.

The surrounding section still matters. Follow the short answer with the evidence, context, limitations, and detail a sophisticated buyer needs. The capsule makes the passage easy to lift; the rest of the page makes it trustworthy.

Keyword stuffing can backfire because repetition differs from relevance. Write the answer, put it under the heading, back it with a number when you have one, and stop.

Why this requires coordination

The work requires four owners:

• Engineering: Ship server-side rendering where needed, audit Cloudflare, review robots.txt, and confirm that Claude-SearchBot and Claude-User can reach the intended pages.
• Content and creative: Rewrite product and comparison pages so each section answers a clear question near the top, then support that answer with useful detail.
• PR: Earn relevant third-party mentions on sites buyers already trust and make sure each placement points to a page Claude can read.
• Measurement: Use marketing measurement frameworks alongside tools such as Profound, Peec AI, Ahrefs Brand Radar, or OtterlyAI to monitor whether the right pages and messages appear.

The result depends on coordination across disciplines. Those four responsibilities need shared priorities and a shared page list. They also need consistent positioning.

Most growth teams hand those jobs to four vendors who never talk. That separation can leave server-side rendering without rewritten H2s, and comparison pages behind a lead-generation modal. It can also send traffic from a PR placement to a JavaScript-rendered landing page. By month three, you're the only person who knows the whole picture, and you're exhausted.

The engines also use different source mixes, so a plan built around one engine's preferred sources can miss how another retrieves and cites information. Claude, Google's AI Overviews, and ChatGPT's browsing mode do not necessarily pull from the same pages or weigh the same signals.

A single-engine plan leaves coverage gaps on the others. Make the underlying assets accessible regardless of engine, and keep the language specific enough to quote.

A pipeline line item from Claude citations this quarter is too uncertain to promise. Buyers who ask Claude about your category are building their shortlist right now. The fixes that make you citable in Claude start with crawlable HTML and quotable product pages. Third-party mentions extend that visibility. These same fixes make you citable elsewhere.

Earn Claude citations with Understory

Understory runs paid media programs and outbound marketing sequences for B2B SaaS under one team, which also handles creative production. Our engineers review your rendering and Cloudflare configuration, then check robots.txt for retrieval bot access while our content team rewrites your comparison pages to answer-first.

If you already have a PR partner landing tier-one coverage, we coordinate alongside them so every placement points at a page Claude can read. It's the same coordinated model that took Rivial Security's paid media spend from $20K to $70K monthly without losing message consistency across channels, so nothing here requires stitching together another vendor.

Schedule a demo to review which of your pages Claude can already read and close the gaps.