AI assistants may send visitors to your site, but a visit is hard to evaluate if you cannot tell which assistant, link, or campaign brought it in. To track AI referrals with UTM parameters, start with the landing URL the visitor actually opens, then use a consistent naming scheme and verify what your analytics platform records.
This distinction matters because you control some AI-related links, such as an ad destination or a link you distribute yourself, but you do not control every link an assistant generates or cites. ChatGPT also adds its own source parameter to referral URLs, which can make some visits identifiable without a tag you created. The goal is therefore not to label every AI visit perfectly; it is to make identifiable traffic useful while keeping the gaps visible.
How UTM parameters help track AI referrals
UTM parameters are values appended to a URL. In Google Analytics 4 (GA4), parameters including utm_source, utm_medium, and utm_campaign can populate session-level source, medium, and campaign dimensions. They let you ask different questions about the same landing visit: where did it come from, what kind of traffic was it, and was it part of a named initiative?
Direct answer: To track AI referrals with UTM parameters, put a consistent utm_source, utm_medium, and, when useful, utm_campaign on links you can control. For ChatGPT referrals, also check for the utm_source=chatgpt.com parameter that OpenAI says ChatGPT includes. In GA4, review session source, medium, and campaign, then test whether the tagged landing URL survives redirects and is recorded as expected.
Consider a controlled link to an article: https://example.com/guide?utm_source=assistant_partner&utm_medium=referral&utm_campaign=ai_content_test. The source names the particular origin chosen for that link, the medium describes the traffic type, and the campaign groups related activity. These are example values, not a claim that an assistant will automatically create that URL or that GA4 will classify it in any particular default channel.
OpenAI’s Help Center says ChatGPT automatically includes utm_source=chatgpt.com in referral URLs. That gives site owners a recognizable value to look for in analytics. It does not mean every visit influenced by ChatGPT will carry that value: a person can read an answer and later type your address, use a bookmark, or arrive through another route.
OpenAI also says publishers who allow OAI-SearchBot can measure ChatGPT search referrals with analytics platforms such as Google Analytics. That is a reason to check your existing reports before building a new tagging process. First determine whether identifiable ChatGPT traffic is already appearing; then decide which additional, operator-controlled links need their own parameters.
Decide which AI links you can actually tag
The first practical decision is about control, not syntax. You can append parameters to a landing page URL that you supply for an ad, partner placement, or other link you manage. You generally cannot guarantee that an independently generated answer or citation will preserve campaign parameters you wish it had used.
Separate your measurement opportunities into three groups before making a naming sheet:
- Links you publish or provide: You can create and test the exact tagged destination. Examples might include a link in a placement you manage or a landing URL supplied to a partner.
- Platform-added parameters: ChatGPT’s documented utm_source=chatgpt.com belongs here. Observe what arrives rather than assuming you chose or can change the value.
- Uncontrolled citations and indirect influence: An assistant may show an untagged URL, summarize your page without a click, or prompt someone to search for your brand later. A UTM report cannot reconstruct all of those paths.
This classification prevents a common mistake: describing all AI-related sessions as one campaign. A tagged ad click, an organic citation click, and a visit after someone copies a URL may reflect different actions. Combining them under a single medium can make channel comparisons less useful even if the resulting session count looks tidy.
Check the destination, not just the link you submitted
A URL can change between placement and landing. A redirect may remove the query string, a canonical destination may be different from the one you expected, or a browser and consent setup may affect what analytics receives. OpenAI notes that redirects, consent, browser blocking, and UTM handling can affect ad session counting; the same issues matter when you inspect AI referral sessions more generally.
For each link type, keep a record of who controls the destination, which parameters are present when the visitor lands, and which analytics property collects the visit. That short inventory is more useful than generating dozens of UTM combinations before you know whether the first one survives the click.
If you cannot control a URL, use the evidence you do have: platform-added parameters where present, available referral information, and your site analytics. Describe the result as identifiable AI referral traffic, not as a complete count of all readers who encountered your business in an assistant.
Choose a UTM naming convention for AI traffic
A naming convention turns individual tagged links into a report you can read later. GA4 can record the values supplied by UTM parameters, but it cannot infer that two spellings are meant to describe the same source. If one team uses assistant_partner, another uses partner_assistant, and a third uses an abbreviated label, the resulting source breakdown will need manual cleanup.
Start by assigning a single job to each field:
- utm_source identifies the specific origin named by the link. For ChatGPT traffic that carries OpenAI’s built-in parameter, preserve chatgpt.com as the observed source rather than relabeling it in reports as though you created the link.
- utm_medium identifies the kind of placement or relationship. Use distinct values for paid placements and other links when that distinction is important to your reporting.
- utm_campaign groups a defined initiative. Use it when multiple links belong to the same effort, rather than treating every general AI referral as a campaign you launched.
- Optional parameters add detail only when someone will use that detail. GA4 includes fields for utm_creative_format and utm_marketing_tactic, which can support a richer scheme for controlled campaigns.
For example, a team supplying the destination for a paid AI placement might choose https://example.com/demo?utm_source=ai_ad_placement&utm_medium=paid&utm_campaign=product_launch. A separate link it manages for an editorial partnership could use the same campaign if it genuinely belongs to that initiative, while using a different source and medium. The chosen names are internal conventions; they should describe real placements rather than imply endorsement by a platform.
Keep a small shared record with each approved value, its meaning, the link owner, and the final landing URL. Decide in advance how to handle capitalization, spaces, abbreviations, and renamed campaigns. Consistency matters more than an elaborate taxonomy: a short list that people actually follow produces clearer reports than a detailed list that no one checks.
It is tempting to label every AI-related medium as referral because the visit arrived from elsewhere. That may be reasonable for some controlled, unpaid links, but it hides paid activity if you apply it to ads. Treat source as the answer to who or where, and medium as the answer to what kind of placement; keep the distinction meaningful for the people who will compare results.
Build and test an AI referral tracking URL
Creating a URL is straightforward; ensuring that it records the intended session takes more care. GA4 documentation shows that session-level manual source and medium dimensions can be populated from URL parameters such as utm_source and utm_medium. The parameters therefore need to arrive with the relevant landing URL, not merely appear in a planning document or be added by a script after the visit has begun.
- Choose the final landing page. Confirm that it is the page visitors should reach, rather than an old address that immediately forwards them elsewhere.
- Add the essential fields to a link you control. Begin the query string with a question mark, separate parameters with ampersands, and make each value follow your naming convention.
- Include a campaign value if the link belongs to a defined initiative. Add more fields only when they answer a reporting question you expect to revisit.
- Open the complete link and follow any redirects. Check that the final address retains the intended parameters and that the page loads normally.
- Make a test visit under conditions you can recognize, then inspect your analytics to see whether the session dimensions reflect the expected source, medium, and campaign.
Suppose a controlled link is https://example.com/resources?utm_source=assistant_partner&utm_medium=referral&utm_campaign=resource_distribution. If the visitor instead lands on https://example.com/resources with the query string removed, you should investigate the redirect or URL handling before judging the campaign’s performance. A correctly written starting URL is not evidence that the final visit was attributed correctly.
Watch for competing campaign settings
GA4’s configuration reference says campaign_source, campaign_medium, and campaign_name settings can override the corresponding UTM query parameters. If a link appears to contain the right values but the report shows a different source or campaign, inspect your analytics configuration as well as the URL. Fixing a link will not resolve a conflicting setting that replaces its values.
Test with a clearly identifiable campaign name so you can distinguish your visit from routine traffic. Then compare what you entered, what appeared in the browser at landing, and what GA4 recorded. If those three views disagree, investigate that discrepancy before distributing the link widely.
Finally, check the user experience. A tagged URL should still load the intended content, work on the devices your audience uses, and avoid unnecessary redirects. Reliable measurement is useful only if the link remains a good way for a person to reach the page.
Read AI referral data in GA4 without overclaiming
Once tested links are in use, read the report at the same level as the question you are asking. GA4’s session source, medium, and campaign dimensions help you inspect the acquisition context of sessions. Its manual source and manual medium dimensions are relevant when the values came from URL parameters. The field you select matters: a session-level question should be evaluated with session-level dimensions rather than assumed from a differently scoped view.
Begin with source values you can substantiate. Look for chatgpt.com when reviewing identifiable ChatGPT referrals, and look for the exact source values you assigned to your own links. Check the associated medium and campaign before combining rows. A platform-added source value and a team-created paid campaign may both appear in AI-related analysis, but they do not necessarily represent the same kind of visit.
Then move from identification to evaluation. Compare the landing pages those sessions reached and the actions that matter on your site, using only the events or conversions your analytics setup actually measures. If one tagged placement leads to a useful page but visitors do not take the next intended step, the next decision may concern the page or the offer, not the UTM string.
A simple review sequence keeps interpretation disciplined:
- Confirm the exact source, medium, and campaign values that reached GA4.
- Inspect landing pages to see where identifiable AI visitors entered.
- Review engagement or conversion measures that your site has configured, and compare like with like.
- Separate controlled campaigns from untagged or platform-tagged referrals in any summary you share.
- Record known measurement gaps next to the finding that depends on them.
GA4 also supports utm_creative_format and utm_marketing_tactic in its schema. Those can help when you genuinely have multiple creative approaches or tactics to compare across controlled links. They do not repair a missing source on an untagged arrival, so add them only after the core fields are working.
Be careful with conclusions such as “AI generated all of these customers” or “this assistant never sends traffic.” A UTM analysis describes observed sessions with usable attribution, not everyone who read an answer or was influenced before returning through another channel. It is most effective as a repeatable way to compare identifiable links and landing experiences over time.
Keep ChatGPT Ads measurement separate from organic referrals
ChatGPT Ads introduce a different tagging decision because advertisers supply landing page URLs. OpenAI explicitly recommends adding static tracking parameters, including UTM parameters, to those URLs. OpenAI says the parameters persist on ad clicks and can be used with existing analytics tools to understand traffic from ChatGPT Ads.
That makes the destination URL a practical place to distinguish an ad campaign from other ChatGPT referrals. Give the ad a source and medium convention that your team can recognize, and use utm_campaign for the specific initiative you intend to evaluate. Do not assume that the built-in utm_source=chatgpt.com used for ChatGPT referrals is a substitute for a deliberate paid-campaign scheme on an advertiser-controlled landing URL.
For example, an advertiser-controlled destination could be https://example.com/offer?utm_source=chatgpt_ads&utm_medium=paid&utm_campaign=offer_test. Those example values are a naming choice, not a required OpenAI format. The important practices are to keep the values consistent, verify the final landing URL, and avoid presenting paid visits as ordinary editorial referrals.
Use platform reporting and site analytics for different questions
OpenAI says ChatGPT Ads Manager Beta reports impressions, clicks, spend, click-through rate, average cost per click, average cost per thousand impressions, and conversions. That reporting helps answer questions about delivery and ad performance inside the platform. Your web analytics can help answer what happened after a visitor reached a tagged destination, according to the tracking and consent conditions on your site.
Do not expect the click count in an ad interface to match the session count in GA4 exactly. OpenAI notes that redirects, consent, browser blocking, and UTM handling can affect session measurement. When the numbers differ, investigate the landing flow and collection setup before treating either report as erroneous.
For a useful paid-campaign review, keep the ad’s submitted URL, its final destination, and the associated UTM convention together. Examine both the platform’s click and spend measures and the site’s observed landing sessions and outcomes. That approach preserves the different roles of each data source without forcing them into an artificial one-to-one match.
Understand what AI referral UTMs cannot measure
UTM tags make known links more legible; they are not an AI influence detector. They cannot prove that a person read a particular answer, saw a citation but did not click, or came back later because of an earlier AI interaction. They also cannot guarantee that an assistant will reuse a URL you previously distributed with its parameters intact.
The most consequential limits arise at different points in the journey:
- Before the click: You may not control the link an assistant displays, and some AI-assisted discovery never results in a direct visit.
- During navigation: Redirects and URL handling can remove or change parameters before the final page loads.
- During collection: Consent choices and browser blocking can affect whether an analytics session is recorded.
- During attribution: Configuration values that override UTMs can make a recorded session show a different campaign from the one written in the URL.
- During interpretation: A source value identifies what analytics observed, not the visitor’s complete path to making a decision.
These limits suggest a sensible alternative when link-level attribution is unavailable: compare the evidence that does exist rather than assigning a made-up source. Review identifiable ChatGPT referrals, examine your site’s landing-page and conversion data, and use campaign reports for links you control. Keep observations about broader AI visibility separate from measured referral sessions.
When sharing results, name the population precisely. “Sessions carrying our tagged AI partner link” is a narrower, more defensible statement than “all traffic from AI.” Likewise, “sessions showing chatgpt.com as a source” describes an observable analytics value without claiming that every ChatGPT-influenced visitor is present.
This precision also makes improvements easier. If the problem is a lost query string, fix the navigation path. If source values are inconsistent, fix the naming process. If most AI-related visits are outside your control, focus on understanding the identifiable portion and the pages it reaches rather than manufacturing a complete attribution story.
Turn AI referral tracking into a repeatable process
A small operating routine is more durable than a one-time tagging exercise. Assign responsibility for approving source and medium values, require a final-URL check before a controlled link goes live, and keep a record of the campaign name used. That reduces the chance that each new placement creates a fresh spelling for the same origin.
After launch, revisit the same questions in order. Are expected values appearing in GA4? Do they match the final landing URLs? Are paid and nonpaid visits distinguishable? Are the landing pages helping visitors take the next useful step? Each answer points to a different action, from repairing a redirect to revising the page.
Keep the convention modest enough to maintain. A small site may need only a source, medium, and campaign for controlled links, plus a check for ChatGPT’s documented source value. A larger team with multiple creative formats may benefit from the additional UTM fields GA4 supports, provided those fields drive a decision rather than merely lengthening URLs.
Document changes when you alter a value or a landing path. Otherwise, a later report may show two campaign names or sources that appear unrelated even though they describe the same effort. The purpose of the naming record is not administrative perfection; it is to preserve the meaning of the data when the person reviewing it did not build the original link.
Start with one link you control and one existing referral source you can inspect. Test the controlled link end to end, check whether chatgpt.com appears in your GA4 data, and write down what each result does and does not tell you. Once those basics are trustworthy, expand the scheme to the additional placements you can actually manage.
Tracking AI referrals with UTM parameters works best when the URL, analytics configuration, and reporting language agree. Use tags for links you control, recognize platform-added values where they arrive, and keep paid campaigns distinct from other referrals.
The useful outcome is not a claim to measure every AI-influenced visit. It is a dependable view of identifiable traffic that helps you choose which links to improve, which landing pages to revisit, and which measurement problems to fix next.
