Almost Timely News: 🗞️ AI and GEO Advice for Retailers for the 2026 Holiday Shopping Season (2026-05-24)

Almost Timely News: 🗞️ AI and GEO Advice for Retailers for the 2026 Holiday Shopping Season (2026-05-24) :: View in Browser

The Big Plugs

New things!

1️⃣ Prompt to Skill Plugin

2️⃣ My new GEO 101 course

Content Authenticity Statement

95% of this week’s newsletter content was originated by me, the human. The chart in section 2 was generated by Claude’s Python code. Learn why this kind of disclosure is a good idea and might be required for anyone doing business in any capacity with the EU in the near future.

Watch This Newsletter On YouTube 📺

Click here for the video 📺 version of this newsletter on YouTube »

Click here for an MP3 audio 🎧 only version »

What’s On My Mind: AI and GEO Advice for Retailers for the 2026 Holiday Shopping Season

This week, let’s talk about the holiday retail season. If you’re checking your phone to look at the calendar, you’re not mistaken. It’s not even June yet. But Michele asked a fascinating question to me about AI and retailers that I thought warranted a much deeper dive:

“What’s one thing retailers should be doing now to prepare for the future of search this holiday season? And how should they actually approach it?“

Let’s dig into this, because it’s a more complicated question than it seems, and it hinges on what the holiday season is, what search is, and what most companies are getting wrong.

Part 1: The Holiday Retail Season

As consumers, we think of the holiday season as mostly December. In the global north, that’s winter; in the global south, that’s summer. November and December are when B2B slows down and B2C is operating at a frenetic pace.

But that’s not when the holiday season starts, not for retailers. The holiday season starts... well, started a month or so ago. Retailers have to think about inventory, and with global chaos caused by the US federal government’s attacks on Iran and the subsequent closure of major shipping lanes and disruption of the energy economy, there are many more wildcards in play.

Retailers have to try figuring out what demand will be, and when consumers will be interested in what they have to offer. That’s where tools like forecasting come into play, and with generative AI, that forecasting is easier and more accessible than it’s ever been before.

Let’s start with trend data. If we want to understand what’s likely to happen, and we have a fairly durable set of past history to draw from, we can use AI to do that inference. There are two catches here. First, this is classical AI - AI that’s rooted in regression analysis, math and statistics, as opposed to generative AI, which is what almost everyone else means today when they use the word AI.

Second, we need data to forecast from. Some of the best free data sources come from government sources (especially boring ones that are politically neutral, like the number of air travelers through USA TSA checkpoints daily, one of my favorite travel industry datasets) or Google Trends data.

Let’s say we want to know when people will start thinking about shopping for the holidays. Where might we go? We’d start with search data, because search data is durable and highly predictable. Take a look at Google Trends for trends phrases like “holiday gift guide”. These remained durable even through the pandemic, because people never stopped looking for gift ideas. You’ll see it begins its inflection roughly every September.

The second thing we need to determine is whether or not the consumer is in a buying mood. That’s where things like consumer sentiment play a role, such as the University of Michigan consumer sentiment survey. When sentiment is low, people tend to buy less.

The Google Trends data tells you what’s likely to happen. The consumer sentiment survey tells you the probable magnitude.

Now, if you’ve never done forecasting before, the good news is that generative AI makes this easier than ever to get started. You don’t need to own any software other than an AI coding agent like Claude Code, Claude Cowork, Google Antigravity, OpenAI Codex, OpenCode, or similar, and you don’t need the smartest, most expensive model for what’s essentially commodity statistics.

We’ll want to weight things like inflation as well, since life has gotten much more expensive in recent years, so we’ll get median inflation (inflation inclusive of everything) from the US Federal Reserve Bank, plus Adobe’s annual holiday shopping reports which are always excellent summaries of the season overall.

Put all your data in one location and open up your AI coding tool. Here’s a good starting prompt you’ll want to adapt and improve:

You’re a consumer shopping expert skilled at statistics, causal inference, and forecasting. Today you’ll be building a forecast for the 2026 holiday season. In the input folder, you’ll find the following datasets as raw data. [1] Google Trends data for the last 5 years for holiday gift guide related terms, week level data - googletrends.csv. [2] University of Michigan consumer sentiment data, month level data - UMCSENT.csv. [3] Adobe Consumer Shopping survey with multiple charts in it that need extraction into YAML, yearly data - adobe folder, multiple PNGs. [4] Median consumer price index inflation (percent change at annual rate) from the St. Louis Federal Reserve Bank, month level data - MEDCPIM158SFRBCLE.csv. All these data series are different frequencies and need to be accounted for in your forecasting; Google Trends tells us when people are most likely to start searching (intent), Adobe tells us what happened in the past as an outcome to measure past forecasts against, consumer sentiment tells us how likely people are to spend more or less, and median CPI tells us how much to adjust past forecasts and results to normalize spending over the years based on inflation. You’ll need to think through the data series we have and decide what transformations and feature engineering make the most sense. You’ll build your forecast by selecting time series forecasting techniques such as SARIMAX, Prophet, Greykite, Timer S1, and other techniques based on the data we have. Use your web search tools (today is 2026-05-23) to help you select FOSS Python-based time series forecasting libraries and packages such as statsmodels, prophet, scikit-learn, prophet. Packages must have been updated on Github on or after 2026-01-01. Do the feature engineering and transformations first, starting with the conversion of charts from the Adobe PDF into YAML datasets. Use the weekly Google Trends data to disaggregate the yearly Adobe data into weekly target variables before training the model. Then test different forecasting techniques against past outcomes with backtesting and robust overfitting prevention; use causal inference methods and techniques you know such as granger causality, vector autoregression, CausalImpact, and other relevant techniques to judge how much to weigh each input for the final forecast. Once you’ve determined the best techniques or ensemble of techniques, produce an explanation of your choices in the output folder along with the final forecast. The final forecast should be week level data along with a writeup of where the inflection for holiday shopping starts, and should also include a time series visualization in PNG format, 16:9 aspect ratio. All output files must be in the output folder with date-stamped file names such as forecast-chart-2026-05-23.png. Ask me questions until you have enough information to successfully complete the task.

I realize that’s a wall of text, but it’s easier to copy and paste, and AI will know what to do with it. Give this to your coding agent - a web-based chat LLM like ChatGPT will never, ever successfully complete this task.

What we see in the forecast for 2026?