ChatGPT Seems to Have Dropped UTM Tags: How to Measure AI Traffic Now
ChatGPT appears to have stopped adding UTM tags to its outbound links, so AI-driven visits are blending into direct traffic. It's time for a new way to measure your SEO and GEO returns.
Visits from ChatGPT now show up as “direct” traffic in my analytics. How do I work out the return on what I spend on SEO and GEO?
Short answerStop relying only on the “chatgpt.com” source in GA4. Search Engine Land reports that ChatGPT links appear to no longer carry UTM parameters, so many AI visits get logged as direct traffic. Measure in three layers instead: AI citation share (how often AI answers to your core questions mention you), direct traffic to deep pages plus branded search, and orders plus post-purchase surveys. Cross-check all three.
Key takeaways
- According to Search Engine Land, ChatGPT links appear to no longer carry UTM parameters, so many AI visits now show up as “direct” traffic in analytics tools.
- Adobe's Cedric Huesler found that across about 500 pages, ChatGPT-referred visits fell by about 31,000 a day while direct traffic rose by about 26,000 a day.
- Measure the AI channel in three layers: AI citation share; direct traffic to deep pages and branded search; and orders plus post-purchase surveys.
- Don't cut your GEO budget based on one or two weeks of source data; how often you appear in AI answers is the core metric that tracking parameters can't distort.
Visits from ChatGPT suddenly show up as "direct" traffic in your analytics. So how do you work out the return on SEO and GEO? Bottom line: stop treating the "chatgpt.com" source in GA4 (Google Analytics 4) as your only evidence, and cross-check three layers instead: AI citation share, changes in direct traffic and order data. Any one number on its own can push you into the wrong budget call.
It started with news this week. Search Engine Land reported on October 15 that ChatGPT's outbound links appear to no longer carry the "utm_source=chatgpt.com" parameter OpenAI had been adding, and that the way the links open also hides referrer information. Adobe's Cedric Huesler analyzed about 500 pages where ChatGPT accounted for at least 5% of visits, comparing October 9–13 with the week before. ChatGPT-referred visits fell by about 31,000 a day, while direct traffic to the same pages rose by about 26,000 a day. OpenAI hasn't confirmed whether the change is intentional.
Why did my ChatGPT traffic suddenly "disappear"?
The traffic didn't disappear. It got a different label. Analytics tools decide where a visit came from mainly by looking at two things: UTM parameters (tracking tags added to a link) and the referring URL the browser passes along. When both are missing, the visit is logged as "direct," as if the shopper had typed your address into the browser bar.
According to the report, the change was spotted after GPT-6 launched in October. Huesler's data also showed that two-thirds of the pages that lost ChatGPT referrals saw direct traffic rise at the same time, and the new "direct" visits behaved a lot like previously identified ChatGPT visits. The report also cautions, though, that the analysis can't prove every missing referrer has the same cause.
If you run an independent store and your store dashboard or GA4 shows ChatGPT referrals plunging and direct traffic jumping after mid-October, don't jump to conclusions. Most likely the label changed, not AI's willingness to recommend you.
Why can't click data alone tell you whether GEO is paying off?
Performance measurement theory, from management research, makes a simple point: whatever you measure is what your team optimizes, and the wrong metric leads to the wrong decisions. If you only watch "visits from ChatGPT," this week's report will say your GEO results collapsed. You might cut the budget when, in reality, shoppers are still coming, just under a different label.
Go one level deeper, and much of the AI channel's value never shows up as a click in the first place. A shopper sees ChatGPT recommend your brand, remembers the name, and three days later types in your URL or searches your brand on Google. No attribution model will credit that order to AI. So measuring the AI channel means asking whether you've earned a place in shoppers' minds, not just reading one source line in a report.
A three-layer framework for measuring the AI channel
Split your metrics into three layers, each answering one question. Did AI mention you? Did shoppers visit because of it? Did those visits turn into sales?
| Layer | What to track | Where to look | Why it matters |
|---|---|---|---|
| Exposure | AI citation share; whether AI describes your brand accurately | A fixed set of questions, checked regularly in ChatGPT, Perplexity and Google AI | Doesn't depend on UTM tags; the most direct read on GEO results |
| Traffic | Direct traffic to deep pages; branded search volume; AI referrals you can still identify | GA4 landing page reports, Google Search Console | Few people type a long product-page URL by hand, so a rise in this kind of direct traffic deserves attention |
| Business | New-customer orders, average order value, share of post-purchase survey answers that pick "AI assistant" | Store dashboard, post-checkout survey | Answers the final question: is it worth it? |
Read the three layers together. If exposure rises but traffic doesn't move, AI is mentioning you without giving shoppers a reason to click, so add buying information such as price ranges and use cases to the pages being cited. If traffic rises but sales don't, the problem is most likely your landing pages and checkout, not GEO itself.
How should I set up GA4 so I lose less AI traffic?
- Create an "AI assistants" custom channel group. Group the referral domains of AI products such as chatgpt.com and perplexity.ai together, using the domains that actually show up in your account, so they don't get scattered across "Referral" where nobody looks.
- Break down direct traffic by landing page. Focus on direct traffic to deep pages like product detail pages and blog posts, and analyze it separately from homepage direct traffic. Direct visits to your homepage mostly come from returning customers; an unusual rise on deep pages is more likely to be AI.
- Annotate the date in your reports. Log the mid-October change as a note, so later year-over-year and month-over-month comparisons don't fool you into thinking GEO stopped working.
- Add a post-checkout survey. Ask just one question, "How did you hear about us?", and include "ChatGPT or another AI assistant" as an option. You don't need a big sample; the trend alone is valuable.
- Spot-check the same questions every month. Pick 30 to 50 core buyer questions and record whether AI answers mention you and link to you. That's the raw data behind your AI citation share.
When my boss asks whether SEO and GEO are worth it, what do I say?
Don't promise precise attribution. Make the case with trends and cross-checks. Here's a hypothetical example: an independent store selling outdoor gear sees its AI citation share climb from one in ten to one in four over three months. In the same period, direct traffic to deep pages and branded searches both grow, and more new customers pick "AI assistant" in the post-purchase survey. Three lines moving in the same direction are more convincing than any single source number.
For the math, use a simplified formula: the return on SEO and GEO roughly equals the added gross profit from organic channels divided by the total cost of content, tools and people over the same period. Estimate the top number conservatively from order data and surveys, and count your team's time in the bottom number.
Then compare those organic orders with your customer acquisition cost from ads over the same period, and you have a benchmark your boss can follow. Even if you can't calculate it to two decimal places, the direction and the scale are clear, and that's enough to support a budget decision.
What should I watch for next?
OpenAI might bring UTM tags back or keep changing how its links work, and other AI platforms could change things without warning, too. So don't make budget decisions on one or two weeks of swings, and don't bet your whole measurement system on any one platform's tracking parameters.
In the short term, archive four weeks of data on either side of mid-October as a comparison set. If ChatGPT referrals come back later, you can use it to estimate roughly how much AI traffic was "hidden" in the meantime.
The truly stable metric is how often you appear in AI answers. If you'd rather not spot-check by hand every month, ToucHigh's GEO Autopilot tracks AI citation share directly, whether or not UTM tags are there.
How ToucHigh helps
UTM tags change and referrer data gets lost, but whether AI answers include you is something you can measure directly. ToucHigh's GEO Autopilot is built around exactly that:
- It audits which questions you're invisible on across ChatGPT, Perplexity, Google AI Overviews / AI Mode and Gemini.
- It writes and ships fixes, including pages, answers and brand entity data, and builds GEO articles around search questions as they trend.
- It tracks your AI citation share continuously, giving you a results curve that doesn't depend on UTM tags.
If you'd like to see your baseline in US AI search first, book a 30-minute strategy call.
Frequently asked questions
Does a sudden jump in direct traffic always mean it came from ChatGPT?
Not necessarily. Search Engine Land's report also cautions that Huesler's analysis can't prove every missing referrer has the same cause. Check three things: Did the rise start in mid-October? Is it concentrated on deep pages like product pages and articles? Did ChatGPT referrals drop at the same time? Only when all three line up is it likely to be AI traffic that has been relabeled.
What about traffic from AI platforms other than ChatGPT?
First check the referral report in GA4 to see which AI product domains actually appear, and put them into one “AI assistants” channel group. But this ChatGPT change shows that platforms can change how their links work at any time, so treat referral data as just one input, read alongside AI citation share and order data.
How do I calculate AI citation share?
Fix a set of core questions, say the 40 category questions US shoppers ask most. Each month, ask every one on a few major AI platforms, count the answers that mention or link to your brand, and divide by the total number of answers. If 8 of 40 questions cite you, your share is 20%. Keep the question list unchanged so the trend is real.
We're a small team with no data analyst. What's the minimum we should track?
Three numbers are enough: monthly AI citation share, direct traffic to deep pages, and the share of new customers who pick “AI assistant” in your post-purchase survey. The first reflects exposure, the second visits and the third sales. Spend half a day a month logging them in one spreadsheet, and after three months you'll see whether the AI channel is driving growth.
Sources
- ChatGPT appears to drop UTM tracking, making AI referral traffic harder to measure — Search Engine Land, 2026-10-15
Why we wrote thisWe wrote this because “How do I calculate the return on SEO and GEO?” is where sellers most often get stuck before setting a budget, and this week's report that ChatGPT may have dropped UTM tags makes the question even harder.
