ChatGPT stopped adding utm_source=chatgpt.com to outbound links around October 8, hiding AI referrals in Direct traffic until tags returned by October 16.
Ecommerce Paradise readers who track AI-search traffic in GA4 are holding weekly reports with a gap of roughly eight days in them. Adobe’s analysis of about 500 pages found ChatGPT visits down about 31,000 a day while Direct visits rose about 26,000 a day, which points to mislabeled traffic more than lost demand. If you cut a budget, shelved a content project or told a partner “ChatGPT isn’t working” off those numbers, you made the call on a broken dashboard.
Black Friday is five weeks out, and Shopify keeps pushing the idea that AI-referred sessions convert well, a claim I covered in Shopify’s 80% conversion claim for AI-referred sessions. Below is what broke, how the sources disagree on the fix, what it does to a high-ticket store’s reporting, and what I’d do this week.
Your analytics just proved how much a shaky foundation costs, so fix the one you control: form your LLC with Northwest Registered Agent, the best LLC formation service for ecommerce owners. Northwest has operated for more than 25 years under stable ownership, which matters when your registered agent has to still be around at your fifth annual report, not just your first. Form your LLC with Northwest →
ChatGPT Dropped utm_source Tags From Oct. 8 Until Oct. 16
According to Search Engine Roundtable, published October 14 and updated October 15, ChatGPT links lost their UTM tag after the GPT-6 release. Cedric Huesler of Adobe wrote on LinkedIn that GPT-6 opens links with window.open using noopener and noreferrer, so analytics files the visit under Direct.
Huesler’s numbers, as reported by Search Engine Land on October 15, cover about 500 pages where ChatGPT drove at least 5% of visits. He compared October 9 to 13 with the prior week. ChatGPT visits fell about 31,000 a day, Direct rose about 26,000 a day and search fell 7%. Direct rose on two of every three pages that lost ChatGPT traffic, against roughly half the week before.
Other measurements, per Search Engine Journal, point the same way. Graphite CEO Ethan Smith reported that GA4 sessions attributed to ChatGPT fell 22% to 57% from October 9 on sites with meaningful AI assistant traffic, an estimated 45% overall. LLM Pulse counted ChatGPT sessions 46% lower on October 9 to 11 than on the same days in the previous two weeks, with total traffic flat. Tomek Rudzki of Peec AI found direct URLs in 42% of ChatGPT responses on October 9 to 13, up from 8% to 10% before. Those figures count links displayed, not clicks.
Profound measured the tag itself. The share of ChatGPT answers with at least one tagged source link fell from about 98% on October 8 to 66% on October 14, according to SEJ’s summary.
Huesler said OpenAI told him on October 14 that the missing tag was a known bug. Relevant Audience reported on October 16 that the rollout began around 2am Pacific on October 15, and that by 9am Pacific tagged ChatGPT clicks were back to roughly 95% to 100% of the same hour a week earlier across thousands of domains in Adobe’s data. OpenAI has made no public statement. Its release notes for October 1 to 15 do not mention the issue, so the confirmation is secondhand.
The sources disagree on the exact recovery. Profound’s own log shows the tagged share of cited answers dropping to about 10% at 19:00 UTC on October 14, sitting between 5% and 12% for roughly 27 hours, then climbing to about 99% by 06:30 UTC on October 16. Glenn Gabe, an SEO consultant, said he had not yet seen the fix on paid or free accounts as of the October 15 reports. Evertune, updating its write-up on October 20, says UTM parameters began reappearing October 16. Adobe measures clicks and Profound measures cited answers, so both can be right about different things.
Two losses are permanent. The referrer has not been restored on most ChatGPT clicks, per Huesler, and sessions already counted as Direct will not be backfilled.
How GPT-6 Broke the UTM Tagging OpenAI Added in June
OpenAI introduced the utm_source=chatgpt.com tag in June, according to Search Engine Land. Before that, AI referrals depended on whatever referrer the browser passed along. The tag made ChatGPT traffic countable in a way store owners could actually put in a weekly report.
The mechanics matter because GA4 is unforgiving about unlabeled traffic. Google’s default channel group definitions treat Direct as traffic arriving via a saved link or a typed URL, with source “(direct)” and medium “(not set)” or “(none)”. A click with no tag and no referrer looks exactly like someone typing your address.
Counterpoint, and it is a real one. SEJ notes that LLM Pulse did not observe a Direct increase, and that nobody can yet separate lost visits from relabeled visits. Huesler’s analysis rests on four days of sampled data. Graphite, LLM Pulse, Profound and Peec AI all sell SEO or AI search services, and Huesler works at Adobe, so every dataset comes from a party with a stake in AI traffic mattering. Search Engine Land also pointed out that OpenAI had not confirmed whether the change was intentional when it published.
This is the second attribution hole in a month. Safari 27 now blocks the email-to-ID matching many ad trackers lean on, which I broke down in Safari 27 blocks email-to-ID ad trackers. Add ChatGPT’s tag gap and the clean last-click report most owners trust is shakier than it was on October 1.
For context on why anyone cares, ChatGPT is still a small paid channel for stores. I covered the early ad numbers in ChatGPT ads get 0.73% CTR with ecommerce mostly absent. The organic referral line is where store owners have been watching ChatGPT, and that is the line that broke.
What the ChatGPT Tracking Gap Does to a High-Ticket Store’s Reports
My read is that most of the “drop” is mislabeling. Three independent datasets (Adobe’s page-level shift into Direct, Graphite’s GA4 sessions and Profound’s tag counts) line up with the same window, and tags came back within hours of OpenAI’s reported fix. That is a bug pattern, not a demand pattern. I would not bet that zero visits were lost, because LLM Pulse saw no Direct bump, and nobody has closed that gap.
Here is the hypothetical math, labeled as such. Say your store logs 200 ChatGPT sessions in a normal week. If Graphite’s 45% estimate applies, about 90 of them land in Direct and your report shows 110. At a made-up 0.8% conversion rate and a $2,500 average order, those 90 sessions are worth about 0.7 orders, or roughly $1,800. The revenue error is small. The decision error is not. A $1,500 monthly content or AI-visibility retainer suddenly looks like it lost 45% of its reach in one week, and an owner who panics cuts it.
Use thresholds so you do not overreact. If ChatGPT drives under 5% of your visits (Adobe’s cutoff for its sample), annotate the dates and move on, since the noise is smaller than your normal week-to-week swing. If it drives more than 5%, rebuild that week. Look at Direct landing pages: a Direct spike on a specific product or category URL looks like ChatGPT, while Direct on the homepage looks like normal Direct. Rudzki’s data showed 54% of the direct URLs ChatGPT displayed pointed to home pages, so some real homepage Direct is mixed in.
High-ticket stores have a second problem that the bug exposed. Buyers of $2,000 to $20,000 items research for weeks, call before they buy and often come back through a branded search. A ChatGPT visit is usually the first touch, not the last, and last-click reports undercount first touches in the best of weeks. That is why I keep pushing owners toward a “how did you hear about us” field and a clean CRM, as in my guide to why high-ticket Google Ads need a complete conversion system. Last-click GA4 alone was never going to tell the full story. A tight routine like the one in high-ticket sales follow-up turns that source field into closed sales.
If an agency, freelancer or VA sends you a monthly report, ask them one question: did you adjust for October 8 to 16? A report that shows ChatGPT down 45% with no note about the tag bug is either careless or hiding a weak month. Ask for the annotation in writing, and ask what they changed in the account because of it. If the answer is “we paused something,” you want to know before it compounds.
Benchmarks built on the broken window are also contaminated. If you saw a number comparing ChatGPT traffic quality or cost, including the measured estimates in ChatGPT ad clicks cost 6x Google’s, check whether its sample runs through October 8 to 16 before you quote it to anyone.
If you want incrementality instead of attribution guesses, Google’s own lift tests are the better tool now that they cover more campaign types. I wrote about that in Google opens conversion lift tests to Search and PMax. A lift test does not care which UTM tag survived.
Cleaning up attribution is one more job on a list that already includes feeds, supplier terms, ad accounts and phone sales. That pile is the reason owners hand the build and the running of the store to my team through the turnkey done-for-you service.
Not sure which of your numbers to trust going into Q4? Get 1-on-1 coaching to audit your attribution and fix the store behind it. Get the coaching details →
Fix Your GA4 ChatGPT Data Before Black Friday Budgets Lock
Five moves, in order. None of them require waiting for OpenAI to say anything.
- Annotate GA4 for October 8 through October 16, converted to your reporting timezone. Profound’s guidance is to mark the dates and note that comparisons across the gap are distorted, and that the percentages are not a multiplier for fixing old reports. Use the worst stretch, October 14 evening UTC to October 16 early UTC, as the outer bound.
- Build a GA4 exploration of Direct landing pages for October 9 to 15 against the prior two weeks. Flag the product and category pages where Direct rose and ChatGPT fell, which is Huesler’s pattern. Export it and have Claude rank the pages by the size of the shift. The workflow is in my guide to Claude for ecommerce.
- Freeze budget decisions that rest on October 8 to 16 AI-referral numbers. Treat the rebound as carefully as the drop, since restored tags reflect restored attribution, not new demand. Then run your holiday plan through the high-ticket Black Friday stress test with clean data.
- Put a “how did you hear about us” field on every quote form and phone intake, and log the answer in your CRM. HubSpot free works for this. My setup guide is how to set up HubSpot for a high-ticket store.
- Cross-check GA4 against Shopify‘s own referrer and order reports. Then confirm every email link carries UTMs so email does not pollute Direct either, since Klaviyo and Omnisend both have a UTM setting for campaign links.
Set a calendar reminder for the first full week of November to re-pull ChatGPT referrals with tags intact. That week becomes your clean baseline. If you want a second pair of eyes on the setup, book a call on my discovery page.
Frequently Asked Questions
Did ChatGPT really lose traffic, or was it only a tracking problem?
Probably mostly tracking, per Adobe’s page-level shift into Direct. Nobody has proved that no visits were lost, since LLM Pulse saw no Direct increase. Treat the window as unreliable either way.
Is the bug fixed?
Adobe’s Cedric Huesler says tagged clicks recovered on October 15, and Profound and Evertune put the return of UTM parameters at October 16. OpenAI has not published a statement, so keep watching your own GA4 for the first full week of November.
Can I recover the lost ChatGPT attribution in GA4?
No. Sessions already counted as Direct will not be backfilled, and the referrer is still missing on most clicks. Annotate the window and compare around it, not through it.
Does this change how I should think about AI shopping channels?
It changes how you measure them, not whether they matter. Shopify is opening its catalog to personal agents, which I covered in Shopify opens its catalog to personal AI agents. Google is testing ads inside AI Mode, covered in Google tests ads in AI Mode. Build measurement that does not hang on one UTM tag.
Should I pay for ChatGPT or AI visibility tools to track this?
Not because of this bug. ChatGPT itself is a fine research tool. A tracker such as Ahrefs and its Brand Radar can show where AI answers cite your domain, but none of that fixes Direct. A CRM source field does.
Where do I start if my store is new and I have no AI traffic yet?
Start with the basics in my guide to high-ticket dropshipping. Then grab the free high-ticket niches list. Attribution cleanup can wait until you have sales to attribute.
Want my team to scale the store you already have? See the scaling service →
I’ll keep watching whether OpenAI publishes anything official about the October 8 to 16 gap. Subscribe to the YouTube channel for daily breakdowns. More breaking news coming through the day.
Related Articles
If this was useful, these go deeper:
- Google Universal Cart Is Live: Audit Your Shopify Settings
- Google Tag Manager Can Now Skip Extra gtag.js
- How to Turn Google Shopping Clicks Into Sales for High Ticket Products
- High-Ticket Niches List: 150+ Best Dropshipping Niches for 2026
- Meta’s Shopping Agent Told to Rank Known Merchants First
Trevor Fenner is an ecommerce entrepreneur and the founder of Ecommerce Paradise, a platform focused on helping entrepreneurs build and scale profitable high-ticket ecommerce and dropshipping businesses. With over a decade of hands-on experience, Trevor specializes in high-ticket dropshipping strategy, niche and product selection, supplier recruiting and onboarding, Google & Bing Shopping ads, ecommerce SEO, and systems-driven automation and scaling. Through Ecommerce Paradise, he provides free education via in-depth guides like How to Start High-Ticket Dropshipping, advanced training through the High-Ticket Dropshipping Masterclass, and fully done-for-you turnkey ecommerce services for entrepreneurs who want a faster, more hands-off path to growth. Trevor is known for emphasizing sustainable, real-world ecommerce models over hype-driven tactics, helping store owners build scalable, sellable, and location-independent brands.
