How Can a B2B Brand Check Which Sources AI Assistants Actually Cite?

November 17, 2026

By Steve Merrill · November 17, 2026

How Can a B2B Brand Check Which Sources AI Assistants Actually Cite?

Run your buyers' real questions through ChatGPT, Perplexity, and Gemini, then log every domain each answer cites. Compare that log against your server logs and referral traffic to see who gets cited and who gets ignored.

That's the whole method. The rest of this post is how to do it without fooling yourself.

Why Does It Matter Which Sources AI Cites?

Buyers ask an assistant for a shortlist before they ever open your site. Whoever the assistant cites shapes that shortlist.

Your rankings in Google don't tell you this. The sources an assistant picks often differ from page one of search. Google's own Search Central documentation on AI features says its AI experiences surface links from a wide set of supporting pages. That set isn't the same as your top ten blue links.

So you have to measure citations directly. Nobody does it for you.

How Do You Run a Manual Citation Audit?

Use five steps: build a prompt list, run it across assistants, log citations, check your logs, and rank the gaps. A spreadsheet is all you need.

Step 1: What prompts should you test?

Write 15 to 25 questions your buyers ask. Cover the problem stage ("how do we reduce onboarding time"), the comparison stage ("X vs Y for mid-market teams"), and the vendor stage ("best tools for…").

Pull the wording from sales calls and support tickets. Buyers don't phrase things the way your marketing team does.

Step 2: Which assistants should you run them in?

Start with ChatGPT (search turned on), Perplexity, and Gemini. Use a clean or logged-out session so your own history doesn't bend the answer.

Run each prompt at least twice. Answers shift between runs. One run is an anecdote.

Step 3: What should you log?

Record five columns: prompt, assistant, brands named, URLs cited, and whether your brand appears as mentioned, cited, both, or neither.

Perplexity makes this easy because its answers carry numbered citations. For Gemini, the API's Grounding with Google Search returns metadata listing the web sources behind an answer, so you can pull citations programmatically if you'd rather not copy by hand.

Mentioned and cited are different outcomes. An assistant can name you without linking you, and it can link you without naming you. Track both.

Step 4: How do you confirm with your own data?

Open your server or CDN logs and search for AI user agents. OpenAI lists GPTBot, OAI-SearchBot, and ChatGPT-User in its crawler documentation. Perplexity lists PerplexityBot and Perplexity-User in its bot guide.

Then check analytics for referrals from chatgpt.com and perplexity.ai. ChatGPT search links typically carry a UTM source tag naming chatgpt.com, which makes those visits easy to segment.

Keep the limits in mind. A crawler hit proves access. It doesn't prove a citation. And many AI-driven visits show up as direct traffic, so your referral count will undercount.

Step 5: How do you turn the log into a plan?

Sort cited domains by frequency. The top ten tell you what the assistants trust for your category.

Then sort them by type: review sites, industry publications, forums, vendor blogs, documentation, analyst pages. The type matters more than the name, because it tells you what kind of page you need to publish or earn.

What Patterns Show Up in B2B Citation Logs?

Expect surprises. The first time I logged citations by hand, I figured vendor sites would dominate. Many answers leaned on third-party pages instead: comparison articles, community threads, directories.

Three patterns are worth looking for:

  • Third-party dominance. If review and listicle sites fill most citation slots, your fix is getting accurately listed there.
  • Stale sources. If an assistant cites an old comparison that misdescribes your product, that page is shaping your pipeline right now.
  • Absent entirely. If you never appear in any run, your pages probably don't answer the question in a form a model can lift.

Structured data helps with the last one. Marking up articles, FAQs, and organization details with schema.org types gives crawlers clean facts to read. It won't force a citation, but it removes guesswork.

What Should You Fix First?

Fix the page that answers your highest-value prompt. Put a direct answer in the first two sentences, name specifics, and cite your own sources.

Next, correct any third-party page that gets your product wrong. Reach out to the publisher with facts and a link to your current documentation.

Then rerun the audit in 30 days. Same prompts, same assistants. If the citation log doesn't move, you learned something. If it does, you know what worked.

That feedback loop is the point. Most brands guess at AI visibility. A monthly log replaces the guess with a record.

Frequently Asked Questions

Do AI assistants always show their sources?

No. Perplexity shows numbered citations by default, and ChatGPT and Gemini show links when web search or grounding is active. Without search, a model answers from training data and may name no sources at all.

How many prompts do I need to test?

Fifteen to twenty-five prompts is enough to see patterns. Run each one more than once, because answers and citations change from run to run.

Can I see AI crawler visits to my site?

Yes. Check server or CDN logs for user agents such as GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot, and Perplexity-User. A crawler visit shows access, but it doesn't guarantee a citation.

How often should I repeat the audit?

Monthly works for most B2B brands. Rerun it after you publish or update a major page so you can see whether the citation pattern moved.

Want Help Building This Into a Routine?

WRKNG Digital builds AI visibility systems for brands that sell online. If you want the audit run for you, with the gaps ranked and a fix plan attached, see how we do it here.

Back to Blog