Hey there 👋

For about a week in October, ChatGPT seemed to stop sending traffic.

Starting October 8, many ChatGPT links lost their usual tracking tag. Adobe's Cedric Huesler looked at about 500 pages where ChatGPT drove at least 5% of visits: ChatGPT traffic fell by about 31,000 visits a day, and Direct rose by about 26,000.

Huesler says OpenAI confirmed it was a bug, and he saw tagged clicks recover on October 15. Profound's monitoring had tagging back near normal on October 16. Until then, plenty of marketing teams were staring at a collapsing AI channel in GA4 with no way to tell how much of it was real.

The tags are back, but the bigger issue was there before the bug. A buyer can shortlist you in ChatGPT, come back a week later through Google or by typing your URL, and open an account. Click-based reporting credits the later visit and never sees the ChatGPT part.

Last week was about tracking visibility across markets. This week is the question your finance team asks next: how much revenue is any of it bringing in? I'll cover where those sales go missing and five ways to put a number on them that you can defend.

Let's go. 🚀

Alongside this newsletter, I help financial brands get found on Google, ChatGPT, Perplexity, and Google AI Mode, either directly or through my agency, Rise Up Media.

If you'd like help with yours, tell me what you're working on. It takes two minutes.

Work with me →

How the ChatGPT tagging outage looked from the outside. Profound tracked the share of ChatGPT answers with citations that included at least one tagged link. It sat around 98% on October 8, slid to about two-thirds by October 14, then dropped to between 5% and 12% for about a day before returning to roughly 99% by 06:30 UTC on October 16. Even within one answer, a link shown directly could lose its tag while one inside a product panel kept it. Separately, LLM Pulse saw ChatGPT sessions in GA4 fall about 46% on October 9 to 11 across the sites it monitors, while total traffic barely moved. LLM Pulse adds that the new interface may also send fewer clicks, which analytics can't show. Much of the drop looks like lost attribution, so check Direct, total conversions, and your citation visibility before treating it as a performance problem. [Source: Profound, LLM Pulse]

AI referrals convert a bit better than organic, but there's far less of it. The SEO Works, a UK agency, tracked 95 GA4 properties from September 2025 to August 2026. AI assistants sent 72,095 sessions against 2.8 million from organic search, so AI was 2.6% of organic, rising to 3.9% by August. On the 63 sites with enough data, AI referrals converted at 7.18% against 6.08% for organic, and at 2.48x the rate of Direct. But AI converted worse than organic on 23 of the 63 sites, and B2B product sites converted at 0.62x organic. They're observed rates for tracked referrals and say nothing about cause, so check your own data before you quote anyone's "AI converts better" stat. [Source: The SEO Works]

No single tool fixes this. You'll stack a few imperfect signals and keep a clear line between what you measured and what you estimated. Start with where the credit goes.

Standard attribution assumes the buyer clicks something and the click carries a label. AI search breaks that assumption in four ways.

Stripped clicks are the most common. Mobile apps, in-app browsers, and copied links often arrive with no referrer, so unless the platform adds its own UTM tag, the visit lands in Direct. Graphite CEO Ethan Smith told Search Engine Journal that 95% of mobile ChatGPT visits to Graphite's sites relied on the tag alone, which is why the October bug hit app traffic so hard.

Then, there's the delayed brand search, where the buyer never clicks at all. They remember your name and Google it, or type your URL, a day or a week later. Your branded organic and brand ads take the credit, which makes brand spend look stronger, and AI work look weaker than it is.

Some conversions happen entirely off your site. A finance director shortlists three treasury platforms in ChatGPT and books a demo with one. Your CRM logs it as "Website" or "Inbound call."

And sometimes the error runs the other way. Because utm_source=chatgpt.com lives in the URL, it travels wherever someone copies the link. If a tagged URL ends up in a Slack thread, a forum post, or a blog and someone else clicks it, GA4 can keep the ChatGPT source label even though that person arrived through another channel. Your AI numbers can be too low and too high at the same time.

On May 13, Google added an AI Assistant channel to GA4's Default Channel Group. Visits whose referrer matches a recognized AI assistant get the medium ai-assistant and their own row in acquisition reports, with no setup.

Two gaps to know about: Google names examples (ChatGPT, Gemini, DeepSeek, Copilot and Grok) rather than a full list, and its own channel definitions put AI Overviews and AI Mode clicks in Organic Search.

So keep your old custom AI channel running alongside it, with a regex like chatgpt|openai|perplexity|claude|gemini|copilot. Gemini was spotted adding UTM parameters to some of its links at the start of October, so check your source report for new values.

Then, annotate October 8 to 16 in your property's timezone, so nobody reads that dip as a straightforward loss of visibility. Report AI revenue by landing page as well as site-wide, because the pages AI sends buyers to tell you which content is doing the work.

If you can, check landing URLs that carry a ChatGPT UTM against their recorded referrer. A ChatGPT tag on a visit whose referrer is a forum or another site is likely the copied-link false positive from the table above.

It's the most direct way to catch AI influence that never produced a click.

Add two questions after the conversion, on the confirmation page after someone opens an account, in the demo booking form, or in onboarding:

List "ChatGPT or another AI assistant" as an option next to Google, social, and word of mouth, because nobody volunteers it when "Search engine" is the closest choice. Add a free-text "Somewhere else" field too. Ask after the conversion rather than before, so you're not adding friction to an application form. And store the answers on the same lead or order record as the tracked source, not in a separate survey spreadsheet, so you can compare the two on every sale.

For B2B, have your sales team ask "How did you first hear about us?" on the first call too, and log the answer in the CRM. It catches buyers who skipped the form.

How big is the gap? Omniscient Digital says its own lead data shows roughly a 4x to 5x difference between click-based and self-reported AI attribution, varying by client, with no sample size published. SegmentStream, which sells attribution software, says AI search goes from 2% of credit under last-click to 16% when first-click and self-reported answers are combined. Those are the same sales, credited differently.

Where you put the survey matters. In an Arcalea case study, 72% to 84% of one B2B distributor's ChatGPT-assisted customers checked out as guests each month, with no lead record. A survey on a lead form would have missed them. One on the order confirmation page wouldn't.

And keep the two answers apart. "I found you through ChatGPT" and "I used ChatGPT while researching" aren't the same claim, and a yes to the second doesn't mean ChatGPT recommended you.

Someone who types your homepage URL probably knew you already. Someone who lands directly on your FX fees page or a comparison page might have come from an AI answer that lost its tag. Or from a newsletter or a WhatsApp link, which produce exactly the same pattern.

To narrow it down, segment Direct sessions by landing page and exclude the homepage, login, and account pages. Cross-reference what's left with two things: the pages your AI visibility tool shows getting cited, and your campaign calendar (email sends, newsletters, social pushes) that could explain the same spikes.

Match those sessions against your survey answers from step 2 wherever you can. What you end up with is a pool of Direct revenue that might include AI influence. Treat it as the input for step 5 and nothing more.

When people find you in an AI answer and Google you later, it can show up as branded search growth.

Pull weekly branded impressions and clicks from Google Search Console, plus brand-term volume from Google Ads if you run brand campaigns. Plot them next to your AI share of voice for your priority prompts.

You're looking for a repeating pattern: AI visibility climbs, and a few weeks later branded search climbs with it, with no campaign or PR push to explain it. If you see that more than once, the pattern is consistent with AI contributing to your brand demand. Read it alongside your survey data, since a correlation on its own won't settle anything.

What you're building is a scenario range. It shows leadership what you measured and what customers told you, then adds your assumptions on top, clearly labeled. It doesn't claim to measure incremental revenue, and you should say so when you present it.

Deduplicate your orders or accounts first, then build three tiers, so no sale is counted twice:

The gap between the first two tiers is the most defensible part of the story, because customers told you about it. If it holds steady quarter after quarter, last-click is undervaluing your AI search work.

Put it on one scorecard and review it on a fixed schedule:

After a quarter, you'll have a monthly range with its assumptions written down, so any debate with leadership is about those assumptions, plus your own ratio between measured and self-reported AI revenue to sanity-check GA4 between survey cycles.

This turns an order export with survey answers into the three-tier range from step 5.

The Scenario: You added the two survey questions a few months ago. You've exported the last 90 days of orders or new accounts with the GA4 channel, source, landing page, revenue, a brand or nonbrand flag for search sessions, and both survey answers. Leadership wants to know how much revenue is associated with AI search.

The Prompt:

Run it on an anonymized export, with names, emails, and account numbers removed before you paste anything in.

These are the most popular AI tools we use at Rise Up Media. If you're not using them already, they're worth a look.

Full disclosure: some links above are affiliate links. If you sign up, we'll earn a small commission at no extra cost to you.

Count the AI revenue you can see properly, ask buyers about what you can't, treat Direct and branded search as clues, and report the whole thing as a scenario range with the assumptions attached.

The number won't be exact. It'll still be closer to the truth than a last-click report that hands all the credit to whatever came last.

And the next time your AI channel halves overnight, check whether Direct went up by the same amount before anyone touches the budget.

Until next time, keep exploring the horizon. 🌅

Alex Lielacher

P.S. If you want your brand to show up in Google AI Mode, ChatGPT, and Perplexity, reach out to me.