Your holiday AI traffic report will be wrong in both directions. Decide now how you will read it

Every January, someone on a CPG team builds the holiday recap slide. This year it will have a line for AI traffic, and that line will go up. We think most of those slides will be wrong, and not in one direction. Some will overstate AI's contribution because a tag came back. Some will understate it because a tag went away or because the shopper bought at Target. The rare slide that gets it right will be the one whose author decided in November how to read it.

That is our argument in one paragraph. The rest of this piece is the evidence and what we would write down before Black Friday.

The growth is real

Start with what is not in doubt. Adobe, which measures over 1 trillion visits to US retail sites, reported on October 29 that AI traffic to US retail sites grew 71% year over year in September 2026. It forecasts much bigger jumps on the peak days.

Show the data

Period	Growth year over year
September 2026 (actual)	71%
Thanksgiving (forecast)	159%
Black Friday (forecast)	95%
Cyber Monday (forecast)	88%

Source: Adobe, October 29, 2026. AI traffic means visits from users clicking a link in an AI service. The three holiday figures are Adobe forecasts, not results.

Adobe also reported that AI traffic converted 90% better than non-AI traffic in September and generated 83% more revenue per visit. Those are retailer sites, not brand sites, and they are averages across categories, but the direction is clear. Shoppers who arrive from an assistant are more ready to buy than they were a year ago.

So there will be growth to report. The question is how much of what you see on your own dashboard is that growth.

The tags are not stable

Most analytics tools identify an assistant visit in one of two ways: a referrer from the assistant's domain, or a utm_source tag on the link. In the last two months both signals have broken, more than once.

Date	What changed	What it does to your reports
Around Sept 25	ChatGPT stopped passing a referrer on untagged clicks, per WISLR	Those clicks fall into Direct
Around Sept 28	Gemini stopped adding utm_source=gemini to cited links	Gemini is visible only by referrer, on the web app
Gemini iOS app (ongoing)	Opens links in its own WebView with no referrer	Visits land in Direct
Oct 8 to 15	ChatGPT links carried no referrer and no UTM tag, a bug OpenAI fixed on Oct 15	A week of ChatGPT visits filed as Direct
From Oct 15	utm_source=chatgpt.com restored, referrer still missing	The tag is the only signal

Sources: WISLR, October 14 and 16, 2026; Search Engine Roundtable, October 30, 2026.

The size of the effect is not small. On one store WISLR measures, identified ChatGPT arrivals from October 9 to 13 averaged 66% less than over October 1 to 7. A second store fell 57%. Lost clicks to support pages came back as Direct. Lost clicks to product pages, for US visitors, mostly did not come back at all, which WISLR connects, cautiously, to ChatGPT's new buttons sending shoppers to retailers instead.

Now picture the holiday comparison. The signals that identify an assistant visit this November are not the ones that identified it last November, and they have changed several times since September alone. A year-over-year line drawn through that is not measuring one thing.

The brand's site is the wrong place to look

There is a deeper problem for CPG brands specifically, and it would exist even if every tag worked. Assistants recommend brands and send the shopper to a retailer. WISLR found that 56% of ChatGPT's new Shop at buttons pointed to big-box stores and about 5% to the brand's own store, across four brands. We wrote that up in detail in today's data post.

For a brand that sells most of its volume through Walmart, Target, Kroger and Amazon, the brand website is a showroom. Its AI sessions are a tiny, unrepresentative sample of AI influence. Ahrefs' traffic benchmark of 118,671 sites put ChatGPT at 0.36% of visits in October 2026. A share like that, measured on a site that is not where the purchase happens, will be reported as "AI is small." We argued in AI traffic is small, and that is the wrong question that this framing misses where assistants actually shape a purchase. The holiday will test that argument at scale.

So the January slide can be wrong in both directions at once. Overstated, where a restored tag makes this year's identified AI visits look like a surge. Understated, where untagged visits sit in Direct and the real effect shows up as a retailer's sell-through number that nobody connects to an assistant.

What we think you should decide this month

We are not arguing for better attribution. We do not think it is available this season. We are arguing for writing down the reading rules before the data arrives, so nobody picks the convenient interpretation in January.

Annotate the breaks now. Put September 28, October 8 and October 15, 2026 on every chart that shows AI or referral traffic. Add any new break the day you see it. Our guide on Gemini's missing tags shows how to estimate a capture rate.

Pick one definition of an AI visit and keep it. For example: referrer from a known assistant domain, or a utm_source from a known assistant. Write it in the report. If a vendor's tool uses a different definition, say so next to its number.

Compare windows that sit on one side of every break. November 2026 against October 16 to 31, 2026 is a cleaner comparison than November 2026 against November 2025, even though it mixes in seasonality. Seasonality you can explain. A tagging change you usually cannot.

Watch Direct on product pages. If Direct traffic to product pages rises at the same time assistant traffic dips, you are probably looking at mislabeled assistant visits. If it does not, the visits may be going to a retailer instead.

Put a non-traffic measure next to the session count. The most useful one for a CPG brand is answer-level: run a fixed list of shopper questions in the main assistants every week of the season and record whether your product is named and which retailer the answer sends people to. Tip 12 describes the method. A brand that is named steadily while its own AI sessions fall is not losing to AI; it is losing the click to a retailer, which is a different problem with a different fix.

Ask your retail partners what they can share. If a retailer gives you traffic-source data through its media or data program, ask whether assistant referrals are broken out. Even a coarse number for assistant-referred sessions to your product pages on a retailer's site would tell you more than your own analytics will.

Why bother, if the numbers are messy anyway

Because budgets get set on these slides. A founder who sees "AI sessions up 120%" in January may move money into AI tactics for the wrong reason. One who sees "AI is 0.4% of traffic" may cut work that was shaping retail sales all season. Both mistakes are expensive, and both are avoidable with a page of rules written in advance.

There is also a simple honesty argument. The people who read your report deserve to know which numbers are measured, which are estimated, and which moved because a platform changed a link. Writing the caveats down before the season is easier than defending them after.

Our position, for what it is worth: treat holiday 2026 AI sessions on your own site as a floor, not a measure. Treat answer-level checks as the leading indicator. Treat retailer sell-through as the outcome. If those three move together, you have a story. If they do not, you have a question worth asking before anyone moves budget. Teams that want help with the answer-level part can do it in a spreadsheet; XENA Intelligence does it across assistants with its SearchPanel tool, but the rules matter more than the tool.