ChatGPT Started Linking to Brands Inside Its Answers. Here's How to Capture the Traffic.

On May 7, 2026, ChatGPT changed one small thing in how it presents answers, and it moved a lot of traffic. Brand names that used to sit in footnote-style citation chips became clickable links placed inside the answer itself, most of them pointing at brand homepages and carrying a utm_source=chatgpt.com tag [1][2]. The share of ChatGPT responses containing a URL jumped from about 4.5% to 20–24% on the same day, roughly a 5x increase, and referral traffic rose to match [2]. For brands that had spent a year asking whether AI visibility converts into real sessions, this update turned an abstract question into a measurable one.

This article explains what actually changed, what the data shows, why the links mostly land on homepages, and the concrete steps to track and capture this traffic in your analytics. The short version: AI visibility is now a trackable acquisition channel, and the brands that instrument it correctly will understand their AI performance months before competitors relying on guesswork.

---

What Changed on May 7, 2026

Before the update, ChatGPT typically attributed sources through citation chips: small numbered or footnote-style references sitting beside or below an answer. Users had to notice them, hover, and click to leave. Most did not. After the update, brand mentions inside the answer became direct, visible hyperlinks, so the link to a brand now sits in the sentence a user is already reading [1][5].

Two technical details matter for measurement.

• The links carry a tracking parameter. Outbound links append utm_source=chatgpt.com, which means the traffic is identifiable in standard analytics rather than hiding inside direct or unattributed sessions [3].
• The links mostly point at homepages. The update favored brand homepage URLs rather than the specific supporting page a claim came from [1][2].

Both facts shape strategy. The tracking parameter makes attribution possible for the first time at scale. The homepage bias changes which page you need to optimize for conversion once the visitor lands.

Citation chips versus inline links

---

What the Data Shows

Multiple independent analyses converged on the same conclusion within days, which is unusual and worth noting. When several trackers measuring different samples report the same direction and rough magnitude, the finding is more credible than any single vendor claim.

• Profound measured the share of ChatGPT responses containing a URL rising from about 4.5% to roughly 20–24% on the same day, a roughly 5x increase that mapped cleanly onto the referral lift [2].
• Similarweb reported that ChatGPT referral traffic nearly tripled, attributing the jump to brand names becoming clickable callouts placed front and center in responses rather than buried in footnotes [5].
• Qwairy, in a study of more than 140,000 ChatGPT answers, found the share of answers containing brand links jumped from 0.43% to 6.20% overnight [3].

The exact percentages differ because each study measured a different sample, a different prompt mix, and a slightly different definition of "link." That is expected. The signal is the agreement in direction and scale: a step change, on a single day, in how often ChatGPT sends users to brand sites.

Reported magnitude by source

A note on reading this table: do not average these numbers into a single figure. They measure different things. Treat them as three independent confirmations that the change was real and large, not as competing estimates of one value.

Ruling out confounding variables

Before attributing a traffic jump to this one change, a rigorous team should rule out the other things happening in ChatGPT around the same period. ChatGPT launched ads in February 2026, and model upgrades shipped through the spring, both of which can independently move citation and referral behavior [6]. So how do we know the May 7 lift came from the link format rather than a coincident model change?

Three characteristics make the attribution credible:

• The change was same-day and step-shaped, not gradual. Model quality improvements tend to move metrics over weeks. A jump concentrated on a single date is the signature of a UI or formatting change, not a slow capability shift [1].
• Multiple trackers with different samples saw it on the same date. Independent methodologies converging on one day is hard to explain with a gradual model rollout [2][5].
• The mechanism is directly observable. You can see the links in the answers, so the cause is not inferred from traffic alone; the traffic simply confirms the visible format change [1].

For your own reporting, apply the same discipline: annotate your analytics timeline with known AI platform changes, and be cautious about crediting any single one until the date, shape, and mechanism line up.

---

Why the Links Point at Homepages

The homepage bias is the single most important strategic detail, and it reverses a habit that content teams built over a decade of SEO.

In classic search, you want the deep page to rank: the specific article, product, or comparison that answers the query. In this ChatGPT behavior, the link often points at the brand's homepage instead of the page that supplied the underlying claim [1]. So the user reads an answer assembled partly from a deep page, then arrives on your homepage with the context of that answer already in their head.

This creates a mismatch worth planning around. The visitor's intent was formed by a specific answer, but they land on a general page. If the homepage does not quickly reconnect to that intent, the session is wasted even though the click was earned.

Three implications follow:

• The homepage becomes an entry point for specific intents, not just a brand overview. It should surface clear paths to major product categories and high-intent destinations.
• Entity clarity on the homepage matters more. The homepage should state plainly what the company is, what it sells, and who it is for, because that is the page ChatGPT is most likely to send people to.
• Deep pages still do the persuasion work inside the answer. They supply the facts ChatGPT synthesizes, even when they are not the link destination, so they remain essential for being cited in the first place.

---

How to Track ChatGPT Referral Traffic

Because the links carry utm_source=chatgpt.com, this traffic is measurable with tools you already run. The goal of this section is a setup you can complete in an afternoon.

Step 1: Confirm the source is arriving

In Google Analytics 4, check traffic acquisition and look for chatgpt.com as a session source. Because the parameter is a standard UTM value, GA4 attributes it to the referral or the campaign source dimension automatically rather than lumping it into direct traffic [3].

Step 2: Build a dedicated segment or report

Create an exploration or segment filtered to session source = chatgpt.com. Track it separately from organic search and generic referrals so the channel does not disappear into a blended bucket. Useful dimensions to add:

• Landing page (expect homepage to dominate)
• Device category
• New versus returning users
• Geography

Step 3: Define the conversions that matter

Homepage-heavy traffic behaves differently from deep-page organic traffic, so measure it on its own terms. Rather than judging it only by immediate purchase, track intermediate signals that fit a top-of-journey visitor.

Step 4: Build a custom "AI Referral" channel group

By default, GA4 does not have an AI channel. Traffic from chatgpt.com usually lands in Referral, and traffic without a clean UTM or referrer can fall into Unassigned, so AI visits get scattered and undercounted. The fix is a custom channel group that isolates AI sources into one bucket.

In GA4, go to Admin, then Channel groups, create a new group, and add a channel named "AI Referral" defined by a rule on the source dimension. A practical matching pattern covering the major answer engines looks like this:

Session source matches regex:
.*chatgpt\.com|.*openai\.com|.*perplexity\.ai|.*gemini\.google\.com|
.*copilot\.microsoft\.com|.*bing\.com/chat|.*claude\.ai

Place this channel above Referral and Organic in the group's priority order, so an AI source is claimed by the AI channel before a lower rule catches it. Once saved, the group applies to reports going forward and gives you a single, clean "AI Referral" line to trend over time.

Step 5: Recover the visits that arrive untagged

Not every AI visit carries utm_source=chatgpt.com. Some arrive with only a referrer, and some as direct traffic. Two habits reduce the blind spot:

• Match on referrer as well as UTM. The regex above keys on the source host, so a visit referred by chatgpt.com without a UTM is still captured.
• Watch for correlated direct-traffic spikes. A rise in direct traffic to deep pages that coincides with higher AI citation frequency is a soft signal of untracked AI influence. Treat it as directional, not exact.

Step 6: Separate the channel in attribution

Because ChatGPT often opens a research journey rather than closing a sale, last-click models will undercount it. Review assisted conversions and data-driven attribution so an early ChatGPT touch is not written off as low value.

A caution on trackability

Not all AI referral traffic is tagged, and behavior varies by model version and surface. One analysis found trackable brand citation traffic rising from 8% in one model generation to 49% in the next, a more than sixfold increase across a single upgrade [6]. The practical reading: what you can measure with utm_source=chatgpt.com is a floor, not the full picture. Some genuine AI-influenced visits will still arrive as direct traffic.

---

Beyond ChatGPT: Identifying Every AI Referral Source

Once you have a tracking setup, the obvious next question is what the other AI platforms look like in your data. ChatGPT is the loudest source, but Perplexity, Gemini, and Copilot also send referral traffic, and each shows up under its own host. Building the full picture prevents you from optimizing for one engine while ignoring the others.

The table below lists the referral hosts to watch. Treat it as a starting map; platforms adjust domains and parameters over time, so verify against your own logs.

Two practices make this table actionable:

• Trend each source separately, then as a group. A per-platform view tells you which engine drives your AI traffic; the grouped "AI Referral" view tells you how the whole channel is growing.
• Compare landing-page behavior across platforms. ChatGPT may favor your homepage while Perplexity sends visitors to deep pages. Those visitors have different intent and need different conversion paths.

What to Optimize Now

The update rewards a specific combination: being cited often enough to appear, and being ready to convert homepage-first visitors once the link is clicked. That splits into two workstreams.

Get cited in the first place

You cannot earn an inline link if ChatGPT does not use your content in the answer. Being cited depends on the same fundamentals that govern AI visibility generally.

• Publish clear, self-contained passages. Answer engines lift individual paragraphs, so each should carry one complete idea with an explicit subject.
• State verifiable facts. Specific claims, figures, and defined outcomes are easier to synthesize than vague marketing language.
• Keep entities consistent. Use one stable name for your brand, products, and category across the site so the system connects mentions reliably.
• Make content technically readable. Ensure critical content is in server-rendered HTML and that AI search crawlers are not blocked, because a page that cannot be read cannot be cited.

Prepare for homepage-first arrivals

Once the click happens, the homepage carries the burden that a targeted landing page used to.

• Lead with a plain-language statement of what you are and who you serve.
• Expose high-intent paths to major categories and decision pages within the first screen.
• Match the vocabulary shoppers and ChatGPT use to describe your category, so the arriving visitor sees continuity with the answer they just read.
• Instrument the homepage with events that capture onward navigation, not just purchases.

#### Before and after: a homepage that reconnects intent

Consider a company that sells lightweight travel hair dryers. ChatGPT recommends it in an answer about "compact dryers for frequent travelers," then links to the homepage.

• Before (answer-blind homepage): the hero reads "Beauty, Reimagined," with a lifestyle image and a single "Shop Now" button. The visitor who just read about travel dryers sees nothing that echoes their intent and has to hunt through navigation. The click is earned but the thread is dropped.
• After (answer-ready homepage): the hero reads "Ultra-Light Travel Hair Dryers — Under 1 Pound, Dual Voltage," with immediate links to "Travel Dryers," "Compare Models," and "Reviews." The visitor sees the same language and category they encountered in the answer, and a clear next step.

The after version does not require a redesign. It requires stating the concrete category and use case in plain language, and surfacing the two or three paths a high-intent visitor is most likely to want. That is the difference between capturing an answer-driven click and wasting it.

Two-workstream summary

---

FAQ

What exactly changed in ChatGPT on May 7, 2026?

ChatGPT shifted from footnote-style citation chips to inline, clickable brand links placed inside the answer text. The links mostly point to brand homepages and carry a utm_source=chatgpt.com parameter, which made the resulting traffic measurable in standard analytics [1][3].

How big was the traffic increase?

Independent trackers reported different exact figures because they measured different samples, but they agreed on direction and scale. Profound saw responses containing a URL rise from about 4.5% to 20–24%, and Similarweb reported ChatGPT referral traffic roughly tripling [2][5].

How do I track this traffic?

Filter your analytics for a session source of chatgpt.com. Because the links use a standard UTM parameter, GA4 attributes them automatically instead of counting them as direct traffic. Build a dedicated segment and track landing pages, engagement, and assisted conversions [3].

Why does ChatGPT link to my homepage instead of the relevant page?

The update favored brand homepage URLs over the specific supporting page a claim came from [1]. Plan for visitors who formed intent from a specific answer but land on a general page, and make sure the homepage reconnects them to that intent quickly.

Is all my AI referral traffic trackable?

No. The utm_source=chatgpt.com traffic is a measurable floor, not the full total. Trackable citation traffic has varied significantly by model version, rising from 8% to 49% across one model generation in one analysis, so some AI-influenced visits still arrive as direct traffic [6].

How do I track Perplexity, Gemini, and Copilot too?

Extend the same custom channel group to match their hosts: perplexity.ai, gemini.google.com, copilot.microsoft.com and bing.com/chat, and claude.ai. Trend each source separately to see which engine drives your AI traffic, and as a group to see how the whole channel grows. Note that platforms differ in landing-page behavior, so their visitors may need different conversion paths.

Does this replace SEO?

No. Being cited by ChatGPT depends on the same fundamentals as strong search performance: readable, factual, well-structured content and clear entities. The update changes how the resulting visit is delivered and where it lands, not the underlying work of earning the citation.

---

Key Takeaways

• On May 7, 2026, ChatGPT moved brand links into the answer text, and referral traffic rose sharply as a result [1][5].
• The links carry utm_source=chatgpt.com, so this is a trackable acquisition channel, not an invisible one [3].
• Independent trackers agree on direction and scale, even though exact percentages differ by sample [2][3][5].
• Links favor homepages, so the homepage must reconnect visitors to specific intent formed by the answer [1].
• What you can track is a floor, not the ceiling, because trackability varies by model version [6].
• Optimize two workstreams: get cited with clear, factual, readable content, and convert homepage-first arrivals with strong entity clarity and onward routing.

AI visibility used to be hard to connect to revenue. As of May 2026, one of the largest answer engines is sending identifiable, taggable traffic straight to brand sites. The brands that set up tracking now, and prepare their homepages for answer-driven arrivals, will read this channel clearly while others are still guessing whether AI search matters.

---

References

[1] - https://www.33sticks.com/articles/chatgpt-may-7

[2] - https://www.tryprofound.com/blog/chatgpt-referrals-branded-links

[3] - https://theplusaddons.com/blog/chatgpt-utm-source-tracking-wordpress/

[4] - https://nicklafferty.com/blog/best-generative-engine-optimization-tools-2025/

[5] - https://similarweb.com/blog/insights/ai-news/chatgpt-referral-traffic-triples

[6] - https://neilpatel.com/marketing-stats/how-much-chatgpt-citation-traffic-is-actually-trackable/