AI citation outreach: a Semrush & Claude workflow

Optimizing for AI hunt is the two an on-page and off-page effort. Your website have to be accessible and optimized for AI models. And it additionally needs to appear on third-party websites that these models citation whenever answering person prompts.

This guide breaks downward the AI citation outreach procedure stage by step. 

You’ll study how to discover citation opportunities in Semrush and how to automate the procedure in Claude Code.

What is AI citation outreach?

AI citation outreach is the procedure of finding pages AI answers citation and getting your brand included on those pages.

Securing AI citations isn’t fair concerning collecting backlinks. 

When an AI answer cites a particular webpage in an answer to a prompt, it method that AI identified that webpage as authoritative, helpful, and applicable to that prompt.

The brands mentioned on the leaf additionally advantage from that authority. Because it boosts their visibility in AI search. And provides crucial brand environment that AI can reference.

Why third-party sources power AI visibility

Third-party websites power AI visibility since ample tongue models (LLMs) recover and synthesize data from multiple sources to answer a prompt.

AI additionally looks at your brand’s website.

But the chances of AI mentioning your brand are higher whenever you appear in applicable online sources additional often. 

In fact, our study alongside Kevin Indig established that 62% of AI citations don’t outcome in a brand mention, additionally known as “ghost citations.” So AI power citation your website in a response, but never citation you. The identical energetic shows up in Google AI Overviews, anywhere a brand's visibility frequently rests on a small figure of cited pages.

This is particularly crucial for evaluation-stage prompts: questions buyers ask whenever weighing solutions.

For example, if you ask Claude “What are the finest AI gathering notetakers?”, it runs a web hunt and says multiple sources.

Not all of these websites are AI notetaking companies. For example, Claude used two genuine person reviews published on Medium. The reply additionally references G2 ratings.

Claude notes in its reply that four brands dominate nearly all difference list: Otter, Fathom, Granola, and Fireflies. 

So equal although the merchandise reviews Claude peruse citation many another notetakers, it recommends the four alongside the general largest presence.

However, the reviews Claude pulled additionally merge outdated information, which impacts the norm of a brand’s AI visibility. 

For example, the reply claims that Granola doesn’t distinct speakers. But its product documentation says speaker attribution is accessible through speaker tags on Zoom and Google Meet.

So AI citation outreach isn’t fair a scheme to close AI visibility gaps. You should additionally use it to justify online sources provision exact data concerning your products.

How to prioritize AI citation opportunities

When prioritizing opportunities for AI citation outreach, obtain into document the following factors:

Your final catalog of top opportunities have to be feasible to execute and contain frequently cited domains.

How to find, win, and track AI citation outreach opportunities

To discover AI citation outreach opportunities, extract the pages AI cites for your precedence prompts, analyze them, afterward scheme your outreach. 

If you’re lone analyzing a fistful of prompts, you can do this manually in your AI visibility tracking tool. 

But to do it at scale, you can use Claude Code. We’ll display you how to do it stage by stage below.

Step 1: Pull citation data

The archetypal stage is to export citation data from the AI visibility tracking tool you use to detect prompts.

If you aren’t monitoring prompts yet, you can do it in Semrush’s Prompt Tracking tool. Enter your domain and click “Set up tracking.”

Select the AI example to track prompts in. You can choose from ChatGPT, Gemini, and Google AI Mode. Add your location, afterward click “Continue To Prompts.”

Then, upload prompts to your campaign. 

After you paste the prompts, click “Add prompts to campaign” > “Start tracking.” Wait up to 24 hours for the data to populate.

In the Prompt Tracking report, go to “Sources.” Here, you’ll discover the pages AI cites in responses to your tracked prompts. And whether your brand is mentioned.

Download the catalog by clicking “Export.” Name it “sources.”

You’ll use this CSV document to run the remainder of the inspection in Claude Code.

Step 2: Set up your Claude Code workflow

Claude Code is a type of Claude that plant alongside records on your computer. 

Although it’s a developer tool, you don't need to have coding cognition to use it. You compose your instructions in plain English, preserve them in a file, and Claude follows them.

Our AI citation outreach workflow in Claude Code allows you to analyze hundreds of webpages at a time. And discover the finest opportunities to get additional AI visibility.

The workflow automatically checks whether your brand appears on particular pages and in what context, preservation you hours of manual work.

Here’s how to set it up.

Install the Claude desktop app

Download the desktop app from claude.com/download and instal it. Then initiate it, sign in, and appearance for the “Code” tab.

You'll need a paid Claude scheme (Pro, Max, Team, or Enterprise). Claude Code isn't accessible on the liberated tier.

Create your project folder

Make a new directory on your device and name it “ai-citation-outreach.” Move your immediate tracking CSV export into it.

Then, create three plain content records in the “ai-citation-outreach” directory that inform Claude Code how to execute the workflow.

You can create these records manually using any content publishing company or create them alongside Claude’s help.

We’ll shield how to create these records and what they need to contain in the next section.

Create the project files

The “CLAUDE.md” document is a project environment document. It needs to explain the CSV data so Claude says it correctly. And understands what to do.

Here’s an example using the Prompt Tracking CSV export. 

Then, prepared the “brand-facts.md” file. It should contain data concerning your:

• Product positioning
• Pricing
• Available products and services
• Direct competitors
• Any incorrect data or claims that third-party pages frequently publish

And another key data that you desire third-party websites to contain or get right. Keep the document succinct, as overloading Claude alongside too much environment can create disturbance and dilatory downward the process.

The document "analyze-sources.md" contains detailed inspection instructions.

Before you compose it, archetypal decide which pages Claude should skip. For example, exclude:

• Your domain
• Social media
• Knowledge bases that don't obtain pitches
• Directories and marketplaces
• Vendor assistance centers, akin help.openai.com

Handle assessment platforms separately. Instead of skipping them, ask Claude to compose them to a second file.

And don't exclude competitor domains. A competitor's blog or difference leaf power motionless be value pitching.

Then, inform Claude how to categorize the pages. You can use this category of framework:

• Not mentioned: The leaf covers your category and lists comparable brands, but you're absent.
• Underrepresented: You appear, but alongside small degree or prominence than comparable brands on the identical page.
• Outdated or inaccurate: Your brand is represented, but the leaf contradicts your brand facts.
• Negative: You appear in a crucial or unfavorable framing.
• No apparent opportunity: The leaf is off-topic, doesn't shield brands akin yours, or already covers you well.
• Manual review: Pages Claude couldn't open or assess.

Write the document manually or alongside Claude’s help. Here's an example.

Now have Claude preserve the document to your project folder.

Open Claude Code > “Select folder” > “ai citation outreach.” 

Paste this prompt, alongside your own instructions in location of the placeholder:

Create a directory called .claude/commands/ in this project, and preserve the following as analyze-sources.md inner it. Use the content exactly as written:

[paste your instructions here]

Claude volition verify formerly it creates the command:

To inspect the command saved correctly, category "/" in the immediate box. If "analyze-sources" shows up in the list, Claude established it and you're prepared to run the workflow.

The document should additionally appear in your project folder.

This is what the completed directory looks like:

Now, you’re prepared to test the AI citation outreach inspection workflow.

Step 3: Run the AI citation outreach analysis

To run the AI citation outreach analysis, open Claude Code, choose your project directory and category “/” in the immediate box. 

Select “/analyze-sources.” Use a additional advanced example akin Opus for the analysis.

Hit enter.

Claude Code volition run the workflow and update you on its progress:

The amount of period and tokens Claude spends on the workflow depends on how many URLs are in the Sources file. (Tokens are data units that AI processes. The additional data that Claude analyzes, the additional tokens it uses.)

If the inspection is taking too long, you can continually inform Claude to halt the workflow and resume it afterward at your command.

When Claude finishes, or whenever you halt it early, you'll get a summary of what it produced:

The workflow produces opportunities.csv (assessed pages), manual-review.csv (pages Claude couldn't open or finish), and routed-reviews.csv (review-platform pages from your elimination rules). Open all CSV document to commencement the assessment process.

Step 4: Review and validate Claude's output

Review the output in the “opportunities.csv” file.

Human assessment is non-negotiable since Claude Code can motionless create mistakes.

Claude is additionally non-deterministic. So if you run the inspection of the identical document twice, you power get slightly distinct classifications on borderline pages.

Using our H&M example, among the top opportunities is this roundup of the most affordable agency apparel for women. 

The leaf links to H&M twice, but Claude tagged it as “Underrepresented” and provided this explanation:

“H&M appears formerly (a ~$50 blazer) among ~50 brands. Expand its workwear presence, tailored trousers, blazers, and knitwear, alongside current cost points so it says as a fuller affordable-workwear choice next to Uniqlo, Mango and Zara.”

A manual assessment of the leaf confirms Claude’s chance categorization. There are additional categories anywhere the webpage could citation H&M products, but doesn’t.

Claude additionally missed that the part features H&M twice, in the blazer and tops sections, not once.

This is exactly why a individual needs to assessment all opportunity. If a publishing company finds an error in your pitch, they power disregard your email.

The archetypal period you run the workflow, you power discover akin errors or areas for improvement. Ask Claude to diagnose the issues in the current workflow and update the project records so forthcoming runs don't reiterate the identical errors.

Step 5: Prepare the outreach handoff

Create a distinct spreadsheet alongside a catalog of your human-reviewed precedence opportunities. It should include:

• The leaf URL
• Associated prompts
• Number of citations
• Claude’s classification
• Claude’s advice alongside any of your additions
• Outreach owner
• Date of outreach
• Outcome

In the “Recommendation” column, provision adequate environment to the outreach owner so they write a tailored pitch.

• Insufficient context: “Add H&M to this listicle.”
• Sufficient context: “H&M appears as a $50 blazer among 50 brands. Suggest adding pants and tops alongside current prices.”

When example the pitch, propose a low-lift alter alternatively of a lengthy catalog of additions to the article.

Sean Markey, who does outreach for Semrush, established that beginning alongside a sole correction and afterward asking approval to dispatch additional gets a improved reply than foremost alongside the complete request.

Here’s an example of his pitch:

Hi Andi,

My name is Sean and I'm operating alongside Semrush.

I'm penning concerning your article on how to inspect your Google ranking. We're in there as "SEMrush" (thanks for the mention!), but I'm expecting we can fix that to fair "Semrush," affirmative a few another small things.

Is it chill if I dispatch complete a quick write-up from the satisfied squad you can use to update it?

Thanks, Sean

Of his final 14 responses, 8 accepted to obtain the changes without asking for item in return. 

You can modify the identical construction to any chance type. Here’s an example using H&M:

Hi Kat,

My name is [Name] and I activity alongside H&M.

I'm penning concerning your roundup of the most affordable agency apparel for women. We're in there alongside a $50 blazer (thanks for including us!). Our prices and inventory move about a fair bit, so I'd be blessed to dispatch you the current particulars for that piece.

We additionally have a few another workwear items that would fit the article well. Is it chill if I put together a abbreviated write-up alongside current prices you can use to update it?

Thanks, [Name]

Send a follow-up email if you don’t get a reply in 3-4 days.

Some publishers power arrive rear alongside a value toggle request. Before starting the outreach, align on a negotiation method so your outreach squad has a apparent procedure to follow.

For example, you power offer:

• A backlink, as lengthy as your website contains pages anywhere you can contain links organically, akin a blog
• Free or discounted merchandise access
• A social media shoutout
• A one-time payment. Give the outreach owner a apparent budget, so they have area to negotiate

Log all ask in your outreach spreadsheet next to the opportunity.

Step 6: Track results

Track your AI citation outreach results in the identical spreadsheet you built in stage 5.

Note publishing company responses, implemented changes, and whenever they went live. Then observe what happens in your immediate tracking.

Re-export the Sources study all duration and difference it to your former export. Monitor the following:

• Is the leaf motionless cited? 
• Does your brand now appear in the immediate responses anywhere AI cites these pages?
• Are new webpages appearing as citations? Do they citation your brand?

The North Star metric to track is your general AI visibility and sentiment. Within a quarter, you should see growth in your AI reply presence. 

If this doesn’t happen, measure another aspects of your AI brand visibility strategy. Perhaps your merchandise positioning and messaging is inconsistent. Or you power not have any integrated satisfied on your website that talks concerning what you do.

AI citation outreach vs. traditional nexus building

The chief differences between link building and AI citation outreach are:

Even although there are differences between nexus construction and AI citation outreach, our study alongside Kevin Indig and Growth Memo established that high-quality backlinks do power AI visibility. 

So investing in the two is beneficial for your hunt and AI presence.

Make AI citation outreach an ongoing part of your strategy

AI visibility is energetic and requires ongoing monitoring. Now that you have a Claude Code workflow set up, it’ll be easier to execute AI citation outreach on a monthly basis.

Semrush Prompt Tracking shows you the pages AI cites for your prompts. And whether your brand appears in the answers. 

Claude Code says those pages and drafts the recommendations against your rules, which saves you hours of manual review.

But individual involvement is motionless an essential step, particularly whenever example the pitch, choosing opportunities, and building relationships alongside publishers.