Read Your Site Search Log

This is a one-afternoon job most stores have never done, and it’s the cheapest customer research you have.

Why this log is different from every other report

Every other way a shopper interacts with your store is a selection from a menu you wrote. Categories you named. Filters you defined. Products you listed. Those tell you which of your options people pick, which is a different thing from what they want.

The search box is the only input where the customer writes rather than chooses.

That makes the log a transcript of how your customers talk about your category: in their own vocabulary, unprompted, at the moment they want something. Surveys are shaped by the questions. Support tickets come only from people with problems. Reviews come from people who already bought.

Nothing else you have is unmediated language from a stranger with intent.

First, the reason to care

Before you read a single query, run one comparison on your own store: conversion for sessions that include a search against sessions that don’t. In most stores the searchers win, and the reason is structural. Using search requires knowing roughly what you want and bothering to type it, which selects for intent.

So a failed search isn’t just a failed session. It belongs to someone more likely than average to buy, who told you in plain words what she was after.

Then check your zero-result rate. It’s usually on the first screen of your search reporting, and usually unread.

Sort a month into three piles

Export a month of queries, sorted by frequency. It takes an hour with a spreadsheet.

Pile one: things you sell, under a word you don’t use.

She searched “jumper,” you call it a sweater. A model number where you list the retail name. A brand you stock but don’t tag. “Sofa bed” where your category is “convertibles.”

This is a defect, not a preference. You had the product, she asked for it correctly, and the store said no.

Action: add the synonyms and turn on typo tolerance. It’s a configuration change, not a project, and every query here is a sale you already had stock for.

Pile two: things you don’t sell.

This is demand data, arriving continuously and free, from people already interested in your category and already on your site.

Split it further: adjacent products you could reasonably stock, variants you dropped (a size, a color, a material), and categories you don’t belong in. The last group is useful too, because it tells you how you’re being perceived.

Action: send it monthly, as a list, to whoever buys stock.

Pile three: situations rather than objects.

Dress for a wedding in August. Something for a small kitchen. Waterproof boots for walking to work. Gift for a dad who has everything.

These aren’t product names. They describe a circumstance, and the shopper is asking your search box to translate the situation into a product. That’s what a good shop assistant does, and what a keyword index structurally cannot.

Action: see below. This is the pile that needs more than configuration.

Why pile three is growing

A few years of talking to AI assistants has changed how people phrase requests. If you spend your week describing situations to a chat interface and getting useful answers, you stop translating your need into keywords before you type.

Your search box now receives sentences from people who learned elsewhere that sentences work. It receives them and returns nothing.

There’s a trap in that, and it’s worth naming because it corrupts your own data:

If sentences don’t work, people stop typing sentences. She tries “waterproof boots for walking to work,” gets a blank page, and next time types “boots.” Your interface has trained her to speak in your terms, and destroyed the signal in the process.

So the log gets less informative over time, and the store concludes that customers search in simple keywords. Which is now true, because the store made it true.

Never return a blank page

The zero-result page is where the damage happens, and you can fix it this week.

Show close matches. Show the category she probably meant. And give her somewhere to put the question, rather than making her guess again in your vocabulary. A blank page teaches her one thing: that asking doesn’t work here.

This isn’t a search-quality problem

These two problems lead to different budgets, so keep them apart.

One is that your search engine is bad: slow, poor matching, weak ranking. That’s real, it’s fixable with software, and a whole vendor category exists to sell you the fix.

The other is that nobody reads the log. No vendor fixes that, because it isn’t software. Install the best search product on the market and the log stays unread, because reading it was never anybody’s job.

Better ranking improves how well you answer the questions you already understand. Reading the log tells you which questions you didn’t know were being asked. And pile three can’t be fixed by ranking at all, however good it gets, because the problem isn’t matching.

Where Immerss fits

Situational queries need something that can take a sentence, understand the circumstance behind it, and answer with a product. That’s a conversation rather than an index.

Immerss puts an AI sales agent on your store that answers in plain language from your catalog and what you’ve taught it about your products, and adds the right item to her cart. When she needs to see something before she decides, one tap brings a real advisor onto live video who can show it and add it to her cart on the call.

It also repairs the data problem. Every conversation is visible to you, so the sentences shoppers type stop disappearing into a blank page and become a second log to read next to your search report. And when sentences start producing useful answers, people write sentences again.

It installs on Shopify, WooCommerce, Salesforce or any store by script, and it’s free to start. See how other stores use it in our customer stories, or get started.

---

Every dashboard in your business reports what customers did. The search log is the only one that reports what they said.

It’s been arriving the whole time, free, in their own words, sorted by how many people said it.