How to Analyze Your Search Term Report

Turning the mountain of data into a molehill of effort

Tim Gilbert · 2 March 2019

Extracting the signal from the noise in your search term report

Your search term report contains information about:

  • What your customers are looking for

  • How they are expressing their needs and preferences

  • Which advertising and SEO approaches are working and which aren't

  • Where you are spending money on irrelevant keywords, and finding untapped market niches for existing

This information is valuable like gold, but only if the raw data can be mined and transformed into knowledge and insight. Otherwise it's just a huge pile of dirt.

Some knowledge is easy to get to like gold nuggets that have been washed down from the mine and are sitting clean, and sparkling in a stream. For example, it's easy to see the really effective searches by using number filters to find high click-through rates (CTR) and high conversions. Unfortunately, the majority of the knowledge and all of the insight requires actual digging into the data.

Since a typical search term report contains tens of thousands (and often millions) of searches, trying to get a thorough understanding and extract all the useful information is somewhere between highly impractical and completely impossible. The reason for this is that you need to examine all the long-tail searches since they can contain significant insight for improving your return on ad spend (RoAS), but by nature of them being long tail, that means there are thousands of them, often with only a few clicks/conversions each.

In order to turn the mountain of work back into a molehill, you'll need an effective to deal with groups of related searches instead of each individual searches. There are many different ways to break up the searches into groups, from simple to complex. I'll list a few of the methods below before diving into a deep explanation of them.

Analyses to breakdown a search term report

  • Filtering by search performance numbers, either straight less-than or greater-than, or by percentages, or by expected performance curves (a search that has a CTR of 5% for 10,000 impressions is relatively much more important than a search with a 50% CTR but only 4 impressions).

  • Breaking down each search into phrases (n-grams) and looking at the best/worst performing phrases.

  • Dividing your searches into product-related (about one or more specific products in your catalog) and non-product related (about the company, store locations, history, blog posts, topical research questions.

  • Dividing your product-related searches into ones that specifically mention your brand, ones that don’t mention a brand, and ones that mention a competitors brand.

  • Breaking down each search into product attributes to see the importance of attributes (e.g. material), particular attribute values (e.g. cotton, silk, leather), and combinations of attributes across products, and measure which get more impressions or convert better.

  • Mapping each search to the product(s) it most closely relates to so you can examine what your customer’s consider important on a per-sku basis.

  • Dividing your searches into topics/themes (especially for the non-product related). For example, if you sell pools, then grouping research questions about how much water a normal pool holds together into a single topic. This can be done with either exploratory clustering to find search topic groups you might not have already known about, or by classifying the searches to clusters you already have in mind (e.g. company history, sales/deals, contact information)

  • Manually creating/maintaining a separate adwords campaign for each product and webpage. This can be powerful, but requires a lot of manpower, only applies when you can create the campaigns, and can’t be applied/changed retroactively. More likely human time limitations means you’ll be maintaining campaigns for whole categories of products.

Although each method has trade-offs and each company’s situation is different, you shouldn’t necessarily be trying to select just one method to analyze your search term. Instead, you probably should use most/all of them.


The Types of Search Term Report Analyses

Filter by search performance numbers

This is by far the simplest thing you can do (needing only a spreadsheet tool and a grasp of basic math) to reduce the amount of work you’ll need to do, but the more you reduce the number of searches to just a few highlights, the more information and potential insight you are ignoring.

Really the minimum you should be doing is calculating expected performance for each search term and filtering by that instead of straight counts or percentages. One method is with an expected SQR performance curve (https://findwatt.com/blog/metric-to-optimize-sqr-performance)

Even with effective filtering, you are still examining searches one at a time if you take this approach by itself. It makes more sense to do this filtering in combination with another method (like topical clusters or search phrases).

 

Search phrases

This is done by breaking down each search into words/phrases (n-grams), and adding up the impressions/clicks/conversions/etc for all the searches that a word/phrase appears in. This is a simple way of starting to get at the different concepts contained across the searches.

Getting the phrases from 1,000 searches will almost certainly result in more than 1,000 words/phrases, so at first glance it seems like you’re going backwards in our effort to reduce the amount of work to review, but it can be an effective way to distinguish between the searches that are low frequency because they are a unique combination of common words instead of being just words that aren’t common in the whole search term report.

For example, if there are many long tail searches that contain “organic cotton”, that might be a good keyword to bid more for even if no single combination of “organic cotton” + product-type (shirt, pants, sweater, etc) has many impressions.

This approach still has several weaknesses. It is completely unaware of synonyms like “caution sign” and “warning sign” meaning almost the same thing. It is sensitive to order so it won’t combine the frequencies of “caution sign” and “sign, caution” even though they contain the same words. And unless you do an in-depth phrase analysis to determine which combination of words are meaningful and which are simply adjacent to each other by coincidence, you end up creating even more noise that will have to be filtered out. (https://findwatt.com/blog/crack-phrases-in-product-titles-to-unleash-their-power)

 

Classify product vs non-product searches

Divide your searches into product-related (about one or more specific products in your catalog) and non-product related (about the company, store locations, history, blog posts, topical research questions, etc).

There is a significant difference from the kind and quantity of traffic that you expect between people searching for products you sell, people searching for information about our company, and people looking to have generic questions answered. Having them all together in one report creates noise when looking at important customer phrases.

You should expect to see good CTR, Conversions and RoAS on your product searches, but you’re not going to expect the same kinds of numbers on the searches for general company information, not expect to see as many online sales from searches about the locations of physical stores. And you’ll want to make sure you’re not bidding too high on keywords that are just driving traffic to company blog posts and articles, even if they are getting good impressions and CTR.

If you’re getting a lot of traffic on research questions (looking for information about thing, not looking how to buy or which to buy), those are opportunities to either filter out irrelevant keywords to raise your quality score, or to write tailored articles on a topic to raise brand awareness/reputation.

 

Classify brand searches

Divide your product-related searches into ones that specifically mention your brand, ones that don’t mention any brand, and ones that mention a competitors brand.

It’s good to measure how effective your are in advertising for people already looking for your brand vs. people looking for generic products vs. people looking for your competitor’s products. An RoAS of 10 on your own products might be substandard while the same RoAS for generic products might be good, and outstanding for searches against competitors products.

You should be aware of how competitive you are on search terms for each of these types of searches.

 

Split product searches up into attributes

Break down each search into product attributes to see the importance of attributes (e.g. material), particular attribute values (e.g. cotton, silk, leather), and combinations of attributes across products, and measure which get more impressions or convert better. This is similar to the search phrase analysis, but vastly more useful since you can compare apples to apples, roll-up performance numbers by groups and patterns, and figuring the best way to standardize values to match buyer intent.

It enables you to create ideal patterns of attributes to put in the title for each category of product, instead of trying to add the less important attributes that would just go over the length limit without adding significant value. This is more effective than trying to use one formula for all your products, and less work than writing every single title by hand.

It also helps you check to make sure that you have the most important attributes in structured fields and in the product description. It can also be used to enhance the webpage titles and meta-tags on your own product pages.

 

Tie the searches to your products

Map each search to the product(s) it most closely relates to so you can examine what your customer’s consider important on a per-sku basis.

Knowing which searches are associated with which products allows a high degree of optimization of your product content because it reveals how important certain product attributes are for the user. Obviously the first step is making sure that your product feed actually contains the information if it applies.

For example, if you’re selling a motorcycle jacket and your customers are searching for “black leather motorcycle jacket”, you should make sure that “black” and “leather” are in the color and size fields, in the product description, and quite possibly in the product title as well. You can also harvest suggestions for other style words about that your customers might be thinking like “racing”, “riding”, “chopper”, and those can be great to include as benefit/use sentences in the description. Improving product content detail and adding customer-mindset focused content can lead to massive improvements in sales for a product.

Since you are reviewing the suggestions for each product individually, this is more useful for refining titles created using the formulas done with the attribute level analysis than for starting from scratch.

This is done with machine learning since it would be infeasible for a human to match every single search to the best product. This means it won’t match everything correctly, but since we are looking for aggregate numbers for the product data suggestions (for example, looking at the common phrases in the 250 closest searches to each product), small errors won’t cause a large problem in the suggestions.

 

Group non-product searches by topic/theme

Divide your searches into topics/themes (especially for the non-product related). For example, if you sell pools, then grouping research questions about how much water a normal pool holds together into a single topic. This can be done with either exploratory clustering to find search topic groups you might not have already known about, or by classifying the searches to clusters you already have in mind (e.g. company history, sales/deals, contact information, etc.)

Then you can add up the performance metrics by topic to see which are worth creating reverse keywords for, which might be used to create blog posts or articles, track your target market’s interest trends, etc.

Since this is done by machine learning, it won’t perfectly create all the clusters that a knowledgeable human would. In other words, sometimes it will break some themes into topics that are too specific which leaves too many topics to practically review, or it will group some searches together that really aren’t related to closely.

 

Highly specific adwords campaigns

You can create/maintain a separate adwords campaign for each one of your products and webpages. This can be powerful, but requires a lot of labor, only applies when you can create the campaigns from the beginning, and can’t be changed retroactively. More likely human time limitations means you’ll be working on campaigns for whole categories of products, which limits the specificity that you can roll-up on when summing search metrics


 

Share:

Social

Search


Highlighted Posts

  • Announcing the Product Title Performance Grader

    Today, we’re announcing the release of our new Product Title Performance Grader, a free tool...

    Read More
  • Case Study: Optimizing Product Titles for Google Shopping

    The Problem Our client suspected that their product titles were limiting the effectiveness of...

    Read More

FREE Product Title Grader

Are Your Product Titles Optimized?

  • Increase Impressions
  • Increase Clicks
  • Increase CTR

Tags