Skip to main content
AI Visibility7 min read

Measure AI Agent Referrals Across Seven Engines

Measure AI agent referrals by testing the same buyer prompts across ChatGPT, Perplexity, Gemini, Claude, Grok, Google AI Overviews and Google AI Mode, then connect answer visibility with referral traffic and the next content fix.

Published

Run a free AI visibility scan

Define an AI referral before measuring it

An AI referral is a visit or other measurable action that follows an AI answer, while a brand mention or citation is visibility without a recorded visit. Keep these outcomes separate because a page can be cited often without sending traffic, and an answer can send a visitor without naming your brand prominently.

Create three fields for every prompt and engine:

  • Mentioned: whether the brand appears in the answer.
  • Cited: whether the answer links to a tracked page or names a source page.
  • Referred: whether the site records a visit that can reasonably be attributed to that engine.

Add a fourth field for the commercial outcome you care about, such as a product-page visit, enquiry, demo request or purchase. Do not treat a citation as a referral, and do not treat an unattributed visit as proof that an AI engine sent it. This distinction prevents a high visibility score from hiding weak traffic quality.

Build a buyer-prompt set that represents real decisions

A useful referral measurement program starts with the questions buyers ask before they choose a provider, product or approach. Group prompts by decision stage rather than writing a list of brand-name searches.

  1. Collect questions from sales calls, support conversations, site search, Search Console and product documentation.

  2. Add comparison, recommendation, problem-solving and local or industry-specific prompts.

  3. Record the exact wording, location, language, date, category and intended buyer stage for each prompt.

  4. Mark prompts that could lead to a commercial visit, rather than tracking only broad educational questions.

Check whether the set includes questions where the brand should be considered, not only questions that already contain its name. For an industry example, AI Visibility for Real estate agents shows how a specific audience changes the prompt set and the pages worth checking. A prompt set that is too generic will produce impressive-looking visibility data with little connection to referrals.

Run identical prompts across all seven engines

Run the same prompt set across ChatGPT, Perplexity, Gemini, Claude, Grok, Google AI Overviews and Google AI Mode, while recording the conditions that can change an answer. Engine comparisons are meaningful only when the prompt, market, language, device context and observation period are comparable.

Capture the full answer, not just whether the brand appears. Record the brand's position in the answer, linked or cited pages, competitors named, answer type, and whether the result includes a direct route to the site. Save the prompt, engine, date and market with each observation.

Repeat the run on a consistent schedule because generated answers can change without a website edit. A manual sample can reveal patterns, but a recurring collection is better for referral measurement because it shows whether a change persists across engines rather than appearing in one answer.

How do I separate visibility from referral traffic?

Separate visibility from referral traffic by joining answer observations to web analytics without assuming that every AI-influenced visit carries a clean source label. A citation proves that a page was available to the answer, while an identifiable referral shows that a visitor arrived through a measurable path.

Use a source or referral field in analytics when an engine passes one, and compare it with landing-page, campaign and conversion data. Also keep an assisted-visibility record for answers that cite a page but produce no directly attributable session. This catches the important case where a buyer reads an AI answer, later returns through search or direct navigation, and the analytics record does not preserve the original influence.

Check each engine separately rather than combining them into one AI channel. Then compare three views:

  • Direct engine referrals and their landing pages.
  • Cited pages with no recorded engine referral.
  • Commercial actions from sessions associated with each engine.

If source data is missing, label the visit unknown instead of assigning it to an engine. The measurement is still useful, but its limitation must remain visible.

Record position, citation and landing-page quality

Referral measurement improves when every answer observation includes the quality of the route, not just a yes or no result. A first-position recommendation with a relevant product page is different from a passing mention linked to an outdated article.

For each brand appearance, record:

  • Answer position or order of appearance.
  • Whether the engine names the brand, links it, or does both.
  • The exact page cited or opened.
  • Whether that page matches the prompt's intent.
  • The competitor pages shown instead.
  • Whether the destination loads, answers the question and offers a clear next action.

Illustrative example: suppose the prompt is "best payroll software for a 40-person company" and Perplexity names the brand third while linking to a general homepage. Record the position, the homepage and the prompt mismatch. Compare referral visits and commercial actions from that route with a later run after creating a payroll page, then check whether the answer cites the new page and whether visitors reach the intended action. The example is a measurement design, not a claimed test result.

Connect engine referrals to business outcomes

Connect AI referrals to business outcomes by measuring what visitors do after arriving, not by treating traffic volume as success. A referral is more useful when its landing page, engagement and next action match the buyer's original question.

Create a consistent reporting view with engine, prompt category, destination page, referral sessions where identifiable, qualified actions and later conversion signals. Keep a separate count for citations with no attributable visit. This lets a small team decide whether to repair a missing citation, improve a destination page or investigate a tracking gap.

Check the setup before trusting the report:

  • The landing page is captured consistently.
  • Engine sources are not merged with ordinary organic search.
  • Redirects and consent settings do not erase the useful source fields.
  • Conversions are assigned the same way across engines.
  • Unknown or direct traffic is not presented as confirmed AI traffic.

AI Visibility Metrics can help structure the measurement vocabulary, while referral reporting adds the operational question of what happened after the answer was shown.

Diagnose the reason a referral is missing

A missing AI referral usually points to one of four different problems: the engine did not select the brand, the answer selected the brand but not a useful page, the page was cited but the visitor could not act, or the visit could not be attributed. Each problem needs a different response.

Use the observation record to classify the failure:

  • No mention: inspect relevance, authority, structured information and competing pages.
  • Mention without citation: inspect supporting pages, factual clarity and crawl access.
  • Citation to the wrong page: improve internal routing and the page's answer to the prompt.
  • Citation with visits but weak action rate: improve the destination's clarity and conversion path.
  • Visitor with no reliable source: repair analytics interpretation before editing content.

Do not rewrite content when the real issue is attribution. Do not add schema when the cited page already answers the prompt but offers a poor next step. This diagnosis is the decision point that stops teams from applying the same fix to every visibility gap.

Choose the next fix and retest its referral effect

Choose the next fix from the largest confirmed gap between buyer intent, answer visibility, useful citation and measurable business action. Prioritise a prompt category with commercial value and a page that can be improved without changing several variables at once.

  1. Select one prompt group and document its baseline across the seven engines.

  2. Choose one intervention, such as a clearer page section, FAQ markup, schema, llms.txt or a better destination page.

  3. Record the exact change, target prompts, target page and expected observation.

  4. Run the same prompts again and compare mention, citation, position, destination quality and referrals.

  5. Check business actions separately and keep the change only if the new visibility is relevant and the measurement is trustworthy.

Cituna is an AI visibility platform that runs these checks across all seven engines, records who each answer names and cites, and generates fixes such as schema, FAQ markup, llms.txt and page changes. Its built-in Google Search Console connection can be used alongside referral data to see whether related search clicks moved. A free AI visibility scan is a practical first check before deciding whether recurring measurement or implementation support is needed.

Official sources to check

Run a free AI visibility scan

Drafted with AI assistance from our own research and Search Console data, and reviewed by Rahul A before publishing. Rules and prices change; check the linked official source before you act.

Frequently asked questions

What is the difference between an AI citation and an AI referral?

An AI citation means an answer names or links to your brand or page. An AI referral means a visitor reaches your site through a path attributable to that answer or engine. Citations can influence later direct or search visits, so report them separately instead of counting every citation as traffic.

Which AI engines should a referral measurement program include?

Include ChatGPT, Perplexity, Gemini, Claude, Grok, Google AI Overviews and Google AI Mode. These engines can present different answers, citations and routes to a site. Run the same buyer prompts across all seven, and record the engine and conditions so comparisons remain meaningful.

Can analytics prove that an AI answer caused a visit?

Analytics can identify some engine referrals, but it cannot always prove that an answer caused every later visit. A buyer may return through search, direct navigation or a bookmark. Report identifiable referrals, cited pages without visits and unknown sources as separate categories.

What should I fix first when an AI engine names a competitor?

First determine whether the problem is selection, citation, destination quality or attribution. Check the exact prompt, competitor pages, cited sources and landing-page intent. If the engine never mentions your brand, improve relevant evidence and page coverage. If it cites the wrong page, fix routing before rewriting the whole site.

How does Cituna help measure AI agent referrals?

Cituna asks ChatGPT, Perplexity, Gemini, Claude, Grok, Google AI Overviews and Google AI Mode buyer questions every day, then records names, citations, positions, competitors and pages. It also generates fixes and connects Google Search Console data, so teams can compare visibility changes with search clicks and their own referral reporting.

Find your next AI visibility fix with Cituna

Cituna asks ChatGPT, Perplexity, Gemini, Claude, Grok, Google AI Overviews and Google AI Mode your buyers' questions every day, writes the fix for every answer you are missing from, and publishes new articles to your site. Run all of it from Claude or any AI agent.

Start free trial

3-day free trial · Card required, cancel anytime · Plans from $39 a month

Check crawler readiness free