To track links from ChatGPT, Gemini, and Perplexity, use UTM parameters on links you control, then examine untagged AI referrals separately in Google Analytics 4 (GA4). A UTM can identify a platform only when the visitor actually clicks a tagged URL. It cannot make an AI assistant add your preferred tags to a citation it generates independently.

For a home improvement company, that distinction matters. A homeowner might click a cited roofing page, copy the address into a browser, or read an AI recommendation and call without visiting the website. Each path leaves a different trail. Our approach at Rank For AI Search is to separate confirmed tagged visits from identifiable referrals and from leads whose AI influence is reported by the customer. That gives you a useful picture without pretending every lead has a perfect source label.

Start with a consistent naming convention, test every tagged link, review session sources in GA4, and record calls and form inquiries alongside AI recommendations. The sections below show how to do each part.

What UTM tracking can and cannot tell you

UTM parameters are values appended to a destination URL. When a visitor opens that URL, an analytics platform can use the parameters to describe the visit’s source, medium, and campaign. The core fields are:

utm_source: The source named in the link, such as chatgpt, gemini, or perplexity.
utm_medium: Your classification of the link, such as ai_referral for a controlled AI-related placement.
utm_campaign: The initiative you want to compare, such as roofing_guides.
utm_content: An optional distinction between placements or destination pages, such as roof_cost_guide.
utm_term: A field generally used for paid-search terms. Most organic AI citation tracking does not need it.

A tagged URL could take this form:

https://rankforaisearch.com/?utm_source=chatgpt&utm_medium=ai_referral&utm_campaign=ai_visibility_guide

That URL is an illustration of the format, not evidence that ChatGPT will choose it as a citation. The source value is a label set by whoever creates the link. It does not independently prove where a visitor saw the link. If you place the same URL in an email, GA4 may still report chatgpt because the tag says so. Placement records and sensible naming are therefore part of accurate attribution.

The governing rule: Tag a link when you control its publication or distribution. When an assistant creates its own link to your site, inspect the incoming referrer and other evidence instead of assuming your UTM scheme will be attached.

Why AI-referred visits are difficult to identify

A browser can pass a referrer when someone clicks from an AI answer to your site. Depending on the platform and browsing context, GA4 may then show a recognizable source associated with ChatGPT, Gemini, or Perplexity. But referrer information is not guaranteed. Apps, browser privacy behavior, redirects, and copied URLs can all affect what reaches your analytics property.

There are also recommendations that produce no website session at all. A homeowner may ask which basement waterproofing contractor to call, see your business in an answer, and dial the phone number directly. GA4 cannot attribute a session that never happened. Conversely, a visitor classified as direct might have arrived after copying a link from an AI answer, or might have heard of you somewhere entirely different.

It helps to keep three evidence categories separate:

Tagged visit: The landing URL contains a UTM that identifies a controlled placement. Check where that link was actually published.
Recognizable referral: The session source or referrer points to an AI platform, even though the destination URL has no UTM.
Reported AI influence: A prospect tells you they found or evaluated your company through an assistant, with or without a recorded click.

These categories answer different questions. Tagged and recognizable visits describe observable website acquisition. A lead-source answer describes what the customer recalls influencing the decision. Combining them is valuable; treating them as interchangeable is not.

Build a UTM naming convention before sharing links

Decide on values once and keep them in a shared link register. Lowercase names are easier to filter consistently. The particular words matter less than whether everyone uses the same ones.

Field	Example value	What to record
`utm_source`	`chatgpt`, `gemini`, `perplexity`	The platform only if the link’s actual placement justifies that label.
`utm_medium`	`ai_referral`	A consistent label for a controlled AI-related link. GA4’s channel grouping may require separate review.
`utm_campaign`	`roofing_guides`	The campaign or content initiative, not an assumed buyer prompt.
`utm_content`	`replacement_cost_page`	The particular link placement or destination variation, if needed.

For a contractor operating across multiple services or cities, record the destination page, service, location, placement owner, publication date, and final tagged URL in the same register. Keep the source field focused on the source. Put a city or service distinction in the campaign or content field when you need to compare it later.

A controlled link intended for a roofing guide might look like this:

https://rankforaisearch.com/seo-vs-ai/?utm_source=perplexity&utm_medium=ai_referral&utm_campaign=roofing_guides&utm_content=educational_link

Use a platform-specific value only when you can identify the placement. If you post a resource on your own social profile and hope an assistant eventually cites it, that profile link is a social placement, not a confirmed Perplexity referral. Naming the source perplexity would make the report look more precise than the evidence allows.

Do not add campaign UTMs to your website’s ordinary internal navigation. Someone who arrives through one source and then clicks a tagged internal link can have their session attribution disrupted. Keep the tags on inbound links you distribute, and let visitors move through the site without a second acquisition label.

A practical setup for AI link tracking in GA4

1. Choose a page that matches the homeowner’s question

A homeowner comparing window replacement options should land on useful window content, not a generic page about every service you offer. A person asking about a leaking basement needs a waterproofing page that explains the problem, service area, and next step. A clear destination makes a tagged click easier to interpret and gives the visitor a better chance of finding what they came for.

Before tagging any URL, check that the page loads on mobile, presents the business identity clearly, and offers a relevant way to contact you. A UTM can reveal that a visit occurred. It cannot repair a page that fails to answer the buyer’s question.

2. Create and log only the links you control

Add the UTM values to the destination URL and save that exact URL with its intended placement. If the page already has a query string, add further parameters with & rather than starting a second query string. In published HTML, encode ampersands as & inside link attributes; the browser will still request the URL correctly.

Then open the link yourself and check that it resolves to the intended page. If it redirects, confirm that the final URL retains the UTM values. A redirect that drops parameters can turn a carefully tagged campaign into an untagged visit.

3. Check session-level acquisition fields

In GA4, open the traffic acquisition reporting area and inspect Session source / medium and Session campaign. Look first for the exact UTM values you chose. Then inspect untagged sources with recognizable platform names, including variations that may refer to ChatGPT, Gemini, and Perplexity. Labels and channel classifications can change, so do not rely on one default channel label alone.

Where reporting needs more detail, use an exploration to compare session source, campaign, landing page, and key events. Keep tagged sessions distinguishable from recognizable but untagged referrals. A custom grouping can make regular reviews easier, but test its rules against actual source values before relying on it for a management report.

4. Verify that a lead action is recorded

For home improvement businesses, a form submission or scheduled estimate is more informative than a page view. A phone-link click is worth tracking too, but it does not by itself confirm that a call connected or resulted in an appointment. Review the tracking on your forms, scheduling flow, and phone links, and check whether your lead records can retain the original campaign or source when a customer contacts you.

Once your events are reliable, compare AI-related sessions with the inquiries they produce. Record which inquiries are in your service area and which involve work you actually sell. That distinction keeps a surge in curiosity clicks from masquerading as business growth.

How to assess ChatGPT, Gemini, and Perplexity traffic without forcing one label

The setup process is the same for each assistant. The observed source is what varies. Use your UTM scheme for controlled links, and review actual session-source data for independently generated citations. Do not create three tagged URLs, place them nowhere, and expect those URLs to reveal how the assistants link to you.

ChatGPT

Look for recognizable ChatGPT-related sources in GA4 and inspect the landing pages receiving those sessions. A link someone copied out of a conversation may not carry a usable referrer. If you distribute a tagged URL in a controlled ChatGPT-related context, keep the placement recorded so its utm_source value has a clear meaning.

Gemini

Inspect Gemini-related source values where they appear, but distinguish Gemini referrals from broader Google traffic. A visit from a traditional Google search result is not automatically a Gemini visit. Likewise, seeing your business in a Gemini answer does not prove that a later direct visit came from that answer. Keep the observed referral, the answer observation, and the lead record separate until there is evidence connecting them.

Perplexity

Perplexity answers often present linked sources that users can open. Review recognizable Perplexity referrals and the pages they reached, then compare those pages with the sources you observe in relevant answers. Even when a citation is visible, a click still needs to reach your site and appear in analytics before you can count it as a session. For more on the visibility side, see our guide to how sources earn citations in Perplexity.

Across all three platforms, the important reporting distinction is what happened on your site versus what the assistant displayed. Both matter, but they are not the same metric.

Measure recommendations that never produce a click

UTMs stop at the edge of a website visit. To understand the business impact of AI answers, ask new callers and form leads how they heard about you. Give staff a way to record an answer such as ChatGPT, Gemini, Perplexity, another AI assistant, or unsure, rather than forcing every customer into a generic internet category. Allow the customer to describe the discovery path in their own words.

Next, keep a repeatable set of questions that resemble actual buying decisions. For a home improvement company, those might concern the service offered, the city served, project cost, emergency availability, and which contractor to hire. Record the platform, prompt wording, date, location context, brands recommended, and linked sources. Repeat the same tests periodically so changes have a meaningful comparison point.

At Rank For AI Search, we use AI visibility audits, monthly monitoring and reporting, citation work, and, in applicable programs, prompt and competitor tracking to examine the recommendation side of this equation. That work complements the visit and lead data in your analytics. It does not turn an observed mention into a proven sale. Our guide to tracking AI search leads without a perfect referrer goes further into that distinction.

Common mistakes that make AI attribution misleading

Tagging a page and assuming an assistant will cite the tagged version. You can publish a URL with parameters, but you cannot require an independently generated citation to reproduce them.
Calling every direct visit AI traffic. Direct is an ambiguous acquisition label. Use it as a reason to investigate, not as proof.
Using different names for the same source. Values such as ChatGPT, chat_gpt, and chatgpt can split what should be one report. Pick a convention and maintain it.
Mixing an owned placement with an independent citation. A tagged link you control and a link generated by an assistant can both produce visits, but only the first follows your planned UTM scheme.
Counting crawler requests as customer visits. AI crawlers reading a page are not homeowners requesting an estimate. Review crawler activity separately from human referral and lead reporting.
Reporting clicks without lead quality. A visit outside your service area has a different business value from a homeowner requesting a relevant estimate.
Overwriting acquisition with internal UTMs. Keep navigation links free of campaign tags so the original arrival source remains useful.

When we assess an AI search program, we put these checks alongside the fundamentals: can assistants access the important pages, does the content answer homeowner questions, and is the business identity consistent across the site and off-site profiles? Measurement is more useful when the underlying presence is sound. Our AI search lead attribution setup guide explains how to bring those signals together.

Best practices for a contractor’s monthly report

A useful report should let an owner see what is known, what is inferred, and what still needs investigation. Keep the first view focused on decisions rather than on every available analytics dimension.

Separate confirmed sources: Show controlled UTM sessions and recognizable untagged AI referrals in distinct lines.
Report landing pages: Identify which service or location pages receive the visits. That points to the homeowner questions and offers worth improving.
Include meaningful actions: Track completed forms, appointment requests, and phone-related activity without treating a click as a completed job.
Review lead quality: Note service fit and location fit in your lead records, then compare qualified inquiries rather than raw sessions alone.
Show recommendation observations separately: List prompts where your company was recommended or cited, and prompts where competitors appeared instead.
Record customer-reported discovery: Keep what callers say in the report, even when no matching website session exists.
Investigate changes: If referrals rise but inquiries do not, inspect landing-page clarity and contact paths. If recommendations rise without referral traffic, check whether people are calling directly or whether cited links reach other sites.

This separation also makes investment decisions easier. A service page that earns citations but few relevant inquiries calls for a different response than a page that converts well but is seldom surfaced in answers. UTM tracking identifies part of that pattern; prompt monitoring and lead records supply the rest.

Frequently asked questions

Can I add UTMs to links that ChatGPT, Gemini, or Perplexity generate?

You can add UTMs to a link you create and distribute. You cannot assume an assistant will attach or preserve those parameters when it independently links to your site. For independently generated links, inspect the source information that actually reaches your analytics property.

What should I use for utm_source on an AI link?

Use a stable lowercase value that reflects the link’s real placement, such as chatgpt when you control a link used in that context. Do not label a general website or social link chatgpt merely because you hope it will be cited there later.

Does a recognizable AI referrer need a UTM?

No. A visit can appear with a recognizable source even if the URL contains no campaign parameters. Review session source and landing page together. UTMs are especially useful when you control the inbound link and want consistent campaign detail.

Why does an AI-influenced lead sometimes appear as direct?

The visitor may have copied a link, used a browsing context that did not pass a referrer, or contacted you without clicking at all. Direct traffic alone does not establish AI influence. Ask the customer, and retain their answer alongside the analytics record.

Can UTM data tell me which prompt produced a customer?

Not reliably for an independently generated AI citation. A UTM records the values attached to the URL someone clicked. If you need to understand which questions trigger recommendations, track those prompts separately and avoid claiming an exact prompt-to-lead connection unless you have direct evidence, such as the customer’s own account.

Should I track AI crawlers in the same report as referral leads?

No. A crawler requesting a page and a homeowner clicking a citation are different activities. Crawler logs help you assess access to your content; GA4 acquisition and lead records help you assess visitor behavior.

What should a home improvement owner review first?

Start with identifiable AI-related sessions, the pages they reached, completed inquiries, and the lead sources customers report. Then compare those findings with a consistent set of homeowner prompts. That gives you a clearer answer than traffic totals alone.

Turn AI link data into a decision you can act on

UTM tracking works well for controlled links. It does not cover every citation, copied URL, direct call, or recommendation that shapes a homeowner’s choice. Treat it as one layer of attribution: keep your tags consistent, verify them in GA4, separate recognizable referrals, and connect both to qualified inquiries and ongoing prompt observations.

If your company needs a clearer view of where it appears in AI answers and how its website and off-site signals support those recommendations, book a free consultation with Rank For AI Search. We work with home service businesses on AI visibility audits, content and technical improvements, brand-signal alignment, and ongoing monitoring so the measurement is tied to a stronger presence where homeowners make decisions.