If you’re still only chasing classic backlinks for AI search, you may be missing the bigger lever. Semrush’s latest walkthrough shows how to find the pages AI models actually cite—then get your brand represented correctly on those pages.
Why AI citations decide visibility
Large language models retrieve and synthesize information from multiple sources. Your website matters, but third-party pages often carry more weight when models decide which brands to name.
Semrush’s research with Kevin Indig found that 62% of AI citations are ghost citations—the model cites your page but never mentions your brand. In Google AI Overviews, visibility can hinge on a small number of cited pages.
Take the prompt ‘What are the best AI meeting notetakers?’ Semrush notes that Claude pulled real user reviews from Medium and G2 ratings, then recommended Otter, Fathom, Granola, and Fireflies because they dominated almost every comparison list. But Claude also repeated an outdated claim that Granola doesn’t separate speakers, even though Granola’s documentation shows speaker tags on Zoom and Google Meet.
That means AI citation outreach has two jobs: close brand gaps and correct inaccurate product information before it scales into bad AI answers.
A repeatable outreach workflow
Semrush’s process starts in Prompt Tracking. You can monitor ChatGPT, Gemini, and Google AI Mode, then export the Sources report to see which pages AI cites—and whether your brand is mentioned. From there, the workflow moves into Claude Code for analysis at scale.
The setup uses three project files:
- CLAUDE.md explains the CSV data so Claude reads it correctly.
- brand-facts.md holds positioning, pricing, competitors, and any incorrect claims to flag.
- analyze-sources.md defines exclusion rules and how to categorize each page.
Claude then labels opportunities as Not mentioned, Underrepresented, Outdated or inaccurate, Negative, No clear opportunity, or Manual review. It also routes review platforms to a separate file instead of skipping them.
Human review is non-negotiable. In the H&M example, Claude labeled an affordable office-clothes roundup as Underrepresented because H&M appeared only once, but a manual check found H&M was actually linked twice. Publishers will ignore pitches with factual errors, so treat the AI output as a draft, not a final list.
Make the pitch easy to accept
For outreach, provide enough context for a tailored email: noting that H&M appears as a $50 blazer among 50 brands and suggesting pants and tops with current prices beats a generic add us to this listicle.
Sean Markey, who does outreach for Semrush, recommends opening with a single correction—thanks for the mention, fix a small detail—then asking permission to send more. Eight of his last 14 responses agreed to accept changes without asking for anything in return. If a publisher asks for value exchange, align on offers like a backlink, product access, a social shoutout, or a one-time payment before outreach begins.
Track what changes
Re-export the Sources report monthly and compare: Is the page still cited? Does your brand now appear in AI responses? Are new pages surfacing? The North Star metric is overall AI visibility and sentiment. If you don’t see growth within a quarter, review your positioning, messaging consistency, and whether your own content explains what you do.
AI citation outreach isn’t a replacement for link building. Semrush’s study with Kevin Indig and Growth Memo found high-quality backlinks still influence AI visibility. Run both.
AI visibility is dynamic, but the loop is clear: monitor prompts, analyze citations, pitch low-lift fixes, verify changes, repeat. Make it a monthly habit and the ghost citations start turning into brand presence.
Source: Semrush Blog
