How to measure AI referral traffic
Measure AI referral traffic from an analytics export, see where visitors land, and explain which visits you can attribute to AI tools.
Free app. Runs on your Mac with the Claude or Codex plan you already have.
What AI referral traffic is
AI referral traffic is the set of visits your analytics attributes to identifiable AI assistants. A session is a visit made up of related interactions. The referrer is the address a browser may pass along when someone follows a link. A campaign tag can also identify a source if the link carries one. Together, those signals let you describe the AI traffic in analytics and show where those visitors land.
Treat the count as the traffic you can attribute under a stated rule. A citation is a named or linked source in an answer, and people can read it without opening the site. A visit can also arrive without source information when someone copies a link or opens it from an app. Keep a separate record of where your site appears in sampled answers and whether your pages are accessible to assistants that fetch web pages. Those checks can explain what to investigate next.
What it does for the business
- Fix pages people already reachCompare the recorded entrances with the pages you expect people to see. If an assistant sends readers to a retired guide, a redirect to the closest live equivalent gives those readers a usable destination.
- Make the next step easy to findInspect the pages that draw recorded assistant referrals. An answer page can help someone solve a problem and still leave them looking for the relevant product, documentation or contact page. Give that reader a link that continues the task they came to do.
- Choose which content to updateWhen a page earns a citation and contains stale information, check the underlying facts and prioritize the updates a reader needs. Use the questions that brought you to that page to brief related content once the existing answer is accurate.
- Explain channel results to clientsA report with the source rules, dates and coverage lets a client repeat the measurement and understand a change. When a referral count is small or tracking coverage changed, that context helps the client decide how much work to fund based on the result.
AI referral traffic, step by step
Define the site, the periods and the AI sources
A single source rule makes two reporting periods comparable.
What to do
- Set the analytics property and the site's scope before exporting. Record the date range, time zone and session attribution settings. Note any consent or tagging changes that might affect which visits the property records.
- Export sessions by session source and medium. The source identifies where a session came from, and the medium describes the type of traffic. Add the landing page, the first page in a session, and include engagement or conversion columns if available.
- Build your channel classification from a dated list of assistant hostnames and known campaign source values. A hostname is the domain part of an address. Match known hosts and tags precisely, and inspect examples before adding an unfamiliar source.
- Apply the same source rule to both periods. Record what happens to rows without usable source information. Visits from answers shown within a search engine stay in search when the export supplies only the search source.
Why
The export is only comparable when the property, dates and source definitions are clear. A host can occur inside an unrelated address, so an overly broad text match can sweep other visits into the assistant group. A dated source list gives another analyst a way to reproduce the report and evaluate later changes.
Check how assistants can reach the site
Check what the server returns before proposing a change to crawler access.
What to do
- Open robots.txt, the file that gives crawlers site access instructions. Identify which AI crawler rules apply to relevant pages. Separate services that fetch for answers or assistant search from crawlers used to collect training text.
- Request the main pages without browser scripts and inspect the response. Record challenges, refusals or redirects, and check that the main answer appears in the returned page. See whether a clear answer and any relevant date are easy to find.
- Check for instructions that limit indexing or snippets, the short extracts a service can show from a page. Read the exact rule and verify which pages it covers before proposing a change.
- Sample questions that your main pages answer. Record the question, tool, date and any page you see cited. Describe the results as observations from those questions.
Why
Pages that open for you in a browser may return a challenge or very little content to a plain request. That difference gives you a concrete issue to investigate when an important page has few recorded visits. A successful access test and a sampled citation each answer a separate question from the session count.
Reconcile the export before comparing periods
Count every session once at the level the export supports.
What to do
- Identify what each row represents before adding anything. A row can be one session, one interaction within it, or a total for a source and page. Count repeated session identifiers once. In a summary export, check for overlapping subtotals and detail rows.
- Keep session, user and event measures separate. One person can have several sessions and a session can produce many events. Keep crawler or server log requests out of the visitor totals.
- Reconcile rows read, duplicates removed, crawler records excluded and the visit totals for AI, other and unknown sources. Retain unknown sessions in the all-session total when calculating the share of recorded visits from AI.
- Check that the comparison covers equal, complete periods. Use conversion data only when the export supplies a defined measure you can connect to those sessions. A page address or campaign tag by itself shows no completed action.
Why
An export can display a plausible channel total even when repeated rows or partial dates distort the comparison. A reconciliation lets you trace how the input becomes the reported number. With sparse referral data, give the counts and explain that a small change can be hard to interpret.
Read the results by tool and landing page
The landing page shows where a recorded referral begins.
What to do
- State the count of attributed AI sessions. Add their share of all recorded sessions and the direction of change when the data support those measures. Show counts by tool and include a monthly view when dates and coverage support it.
- List the leading entrance pages and inspect their current responses and content. For each page, choose a specific action such as redirecting a dead address, updating a stale fact or adding a relevant call to action.
- Compare recorded engagement and conversions with search traffic only when the definitions and coverage line up. Note what the export supports, especially when it contains sessions and events at different levels of detail.
- Look at high-priority pages with zero attributed AI sessions. Check for access blocks or weak answers and mark your explanation as a hypothesis unless you can verify it.
Why
An overall count can move while the page receiving most of the traffic stays the same. Reviewing tool and page together gives you an address to check and an action to assign. Differences in recorded outcomes describe the sessions that arrived, and the report can explain how those visitors were measured.
Check the evidence and hand off the actions
A reviewer should be able to reproduce the count and find the setting behind each proposed change.
What to do
- Recount the totals from the raw file using the recorded source rule. Check that the summary matches the breakdown and that rows without a source remain unknown.
- Reopen the address or exact crawler rule behind each proposed change. Name the live redirect target or write the revised call to action so the person who owns the page can act on it.
- State dates, coverage, source limitations and the main finding in the handoff. If you have no export, give the access findings, record the export needed and explain which traffic questions the file would answer.
Why
Checking the report against the source file catches arithmetic errors before they affect decisions. Reopening the pages ties each action to a condition the site owner can verify, and a short statement of coverage makes the result easier to repeat.
Common mistakes
- Assigning direct traffic to assistantsVisits without usable source information stay unknown even when you suspect people arrived through an assistant.
- Combining sessions and crawler requestsA crawler fetch records a machine request and belongs outside the count of human referral sessions.
- Adding an export total to its own detail rowsCheck whether the export includes a subtotal before aggregating the rows by tool or page.
- Treating every conversion event as a saleRead the conversion definition and report the action the analytics system actually counts.
- Claiming causation from a referral reportSource attribution describes recorded sessions and leaves open what caused a change in business outcomes.
What a good result looks like
- The property, site scope, dates and coverage are visible.
- The dated source rule can be applied to the original export.
- The attributed, other and unknown session counts reconcile with the source.
- Citations and access findings appear separately from recorded visits.
- Each proposed page action names the evidence and the change to make.
And then what? Using the result
- Send dead landing addresses and their verified destinations to the person who manages redirects.
- Give the page owner the exact next step to add for visitors who have reached the end of the answer.
- Update facts on cited pages that readers can use to make a decision.
- Add the dated source rule to the regular analytics review and annotate changes to tagging or consent.
- Give the content team related questions surfaced during the landing page review.
When to run it
- A client asks whether assistants send visitorsUse the recorded sessions to show which tools and URLs appear in the export. Include the scope and limits in the client report.
- A migration changes important URLsInspect old entrance addresses for errors or redirects. Referral reports can help identify retired pages still receiving visits.
- Pages appear in assistant answers and traffic is unclearKeep the sampled citations and traffic review side by side. Check access problems and request a session export to measure attributable visits.
Questions people ask
How do you measure visits from AI tools?
Export sessions by source and classify known assistant hosts or campaign tags under a dated rule. Add landing pages for a page breakdown and leave visits with missing source information in an unknown group.
How do you find ChatGPT referral traffic in Google Analytics?
In Traffic acquisition, select Session source / medium and look for recognized ChatGPT sources such as chatgpt.com. Add Landing page + query string to see where those sessions began and export the table for reconciliation.
Why do some AI visits appear as direct traffic?
Some browsers and apps pass no usable source information when someone opens or copies a link. Analytics may record those visits as direct, leaving their origin unknown.
Can you measure AI referral traffic from citations?
Citations alone cannot measure AI referral traffic. A citation shows that a page appeared as a source in an answer. Use analytics to count the visits you can attribute.
Can you measure conversions from AI referrals?
Yes, if the export has a defined conversion measure you can connect to the relevant sessions. Describe the action measured and keep any claim about business impact within the evidence.
What can you learn without an analytics export?
You can check whether relevant crawlers can retrieve your pages and see whether sampled answers cite them. Request a session export to measure recorded traffic by source and landing page.
More jobs in SEO agencies
Measure visits from AI tools, on your own Mac.Free app. Runs on your Mac with the Claude or Codex plan you already have. Your files stay on your computer.
Download for Mac