There’s A Better Way To Classify Search Intent

If you’ve kept a close eye on new content in the SEO community over the past year or so, one of the things I think you’ll notice is that search intent is on the minds of some really smart folks in our industry:

(There are definitely others but these are some of my favorites).

Why is search intent suddenly in the SEO zeitgeist?

I believe it’s because Google’s search results have shifted in the past few years. Google has gotten significantly better at producing search results that deliver what the user is looking for without relying solely on classical SEO factors.

In the process, I think we’ve seen search intent become a more dominant factor on multiple types of search results, often outweighing classic SEO ranking factors like links, title tags, and other SEO basics. Historical domain-wide link factors (domain authority, basically) no longer carry the importance that we saw in 2010-2015.

I’m burying the lede a little bit here, but I’m really excited to share that in the spring of 2019, Content Harmony will be releasing a new SaaS toolset that helps SEOs & content marketers produce better content.

One of the great features in that toolset is a new scoring system we’ve developed for classifying search intent.

In this post, I’m going to lay out why we felt this was a key part of the puzzle in producing better content. I’m not satisfied with current methods of classifying search intent, and I’ll walk you the new classification and intent scoring process that we have built and show you some examples.

How We Talk About Search Intent Hasn’t Changed Much Since 2002

There are two common ways of labeling search intent that I’ve seen over the years.

The first classification that most of us have encountered is Navigational, Informational, or Transactional.

This methodology dates back to a 2002 peer-reviewed paper from Andrei Broder at Altavista (thanks Tom Anthony for this post with the Alta Vista paper link and some other thoughts on this system):

Broder’s paper defines each category as follows:

  • Navigational: The immediate intent is to reach a particular site.
  • Informational: The intent is to acquire some information assumed to be present on one or more web pages.
  • Transactional: The intent is to perform some web-mediated activity.

The paper goes on to explain an interesting methodology they used to survey users to understand the intent behind their search.

Fast forward a few years. In the early 2010s Google starts referring to their own variation on this, talking about Know, Go, Do, & Buy “micro moments”. These are a bit more user friendly but basically cover the same categories that we see with Navigational, Informational, and Transactional.

I think these traditional classification systems are useful to help beginners understand how different searches are intended to yield different results, but they’re not very useful to SEOs and content creators on a day-to-day basis.

So where do these systems fall apart?

1) [Navigational / Informational / Transactional] Is Too Broad

As SEOs and content creators, we can produce much better content if we understand what a typical user is looking for. Unfortunately I think the systems that exist now are good for explaining search intent in hypothetical terms, but when you try to start applying them to keyword research and showing content creators and writers how to factor that into their content, I think their usefulness is diminished.

If you’re trying to understand the difference between a format-driven navigational search to reach Youtube, like [snowboard videos] or a branded navigational search, like [seattle metro routes], seeing them both labeled as Navigational isn’t super helpful. And grouping keywords together in this system is too generic to be very useful during the keyword research and mapping process.

2) Navigational / Informational / Transactional Don’t Account For Overlapping Intent

Additionally – I can think of many searches that fall into multiple categories. If somebody searches “amazon laptop deals”, would you label that Navigational (trying to reach Amazon) or Transactional (trying to buy a laptop)? So trying to categorize intent as a distinct category rather than a bunch of overlapping intents is problematic as well.

3) Search Intent Labels Are Often Guessed Manually Based Upon The Query Itself, Not The Actual Search Results

Most of the methods people use during the keyword research process involve adding modifiers that we expect to generate a specific intent. An SEO Strategist reviews a long list of keyword in the hundreds or thousands and fills in a column with Transactional, Informational, or Navigational.

For example, if users add words like “sale“, “cheap“, “deals“, or “for sale” to a query, we mark them as transactional and move on. If we encounter a query where it’s not clear what the intent is from glancing at the keyword, we mark it as Split Intent or with our best guess and then we keep moving down the spreadsheet. Who has the time to actually look at all of these SERPs and see what types of results are actually showing up, right?

(Interesting sidenote on manually assigning intent: that 2002 paper explicitly mentions that “…we need to clarify that there is no assumption here that this intent can be inferred with any certitude from the query.”)

Another approach is to look for a city name or modifier like “near me” to label a query as local – even though many of our non-modified search terms might be showing local packs at the top of the results.

No shame is intended if this sounds familiar – I’ve done it, too – but it’s a process that leads to mistakes and incorrect assumptions.

The experienced SEOs in the room should be able to think of many times when a search yielded a number of unexpected result types as Google seemed to shift the implicit intent they were trying to serve.

It’s also not hard to find shifts in intent that occur from query changes that don’t look significant when you’re looking at them in Excel sheet. Take a look at [texas electric] vs [texas electricity]:

[texas electric] results in a local & branded SERP, since Texas Electric Cooperative is the name of a specific entity.
[texas electricity] results in a more unbranded SERP with PowerToChoose.org from the Public Utility Commission of Texas alongside resellers and commercial sites like comparepower.com and saveonenergy.com.

The problem underlying these examples is that we’re trying to do this classification manually, rather than introducing tools into the process that can do it more reliably.

4) SEOs Don’t Need To Understand Intent The Same Way Search Engines Understand It

We (the SEO community) are not search engines and we are not trying to decide what results a user wants to see, so it’s OK for us to look at intent differently than a classical information retrieval model.

The goal we (Content Harmony) are trying to reach is not to understand the true intent(s) of all users. The goal of our tool is to understand the type of content that Google is looking to serve to users based upon what Google knows about the user’s intent, so that we can help SEOs & content creators make the best content for that keyword that they can. I’ve been really careful throughout this post to discuss “search intent”, not “user intent”, for basically that reason.

As SEOs, we mainly need to understand whether our content (and the format it’s in) fit the format that Google is looking to deliver.

How Content Harmony is Classifying Search Intent

OK, so, money where my mouth is – how exactly do I propose we measure search intent more usefully, and more reliably?

I believe that it’s more useful to classify search intent in a way that more closely aligns with SERP features. Here are the Intent Types we’ve begun using in our software (presented in no particular order):

1) Research Intent

One of the most common result types, this would generally consist of search phrases that generate results like Wikipedia pages, definition boxes, scholarly examples, lots of blog posts or articles, in-depth articles, and other SERP features that suggest users are looking for answers or insights into a topic.

Featured Snippets

Knowledge Carousels

People Also Ask Results

2) Answer Intent

Slightly different from research, there are quite a few searches where users don’t generally care about clicking into a result and researching it – they just want a quick answer. Good examples are definition boxes, answer boxes, calculator boxes, sports scores, and other SERPs that feature a non-featured snippet version of an answer box, as well as a very low click-through rate (CTR) on search results.

Weather results – user wants a clear answer.

Mortgage calculator – a mixed result here but by showing an answer box Google has likely greatly decreased CTR to organic results.

Example of a sports score box, where user likely wants a quick answer to score and whether it’s final or in progress.

3) Transactional Intent

Users looking to buy products or research them are easy to detect, since Google tends to be aggressive with Shopping boxes and other purchase-intent features. Other easy methods of detection would include multiple results from known ecommerce players like Amazon or Walmart, results consisting of /product/ types of URL structure, and multiple results that feature eCommerce category/page schema markup.

Many SEOs forget to consider paid results when looking at search intent, but a prominent shopping box is often a clear indicator of transactional intent.
Multiple prominent eCommerce product/category pages (alongside the shopping results) and product rating rich snippets are good indicators of transactional intent.

4) Local Intent

Dominated by a combination of local packs, geographic markers that have recently started showing up in organic results, Maps intent in the top navigation, and if there’s a way of detecting it, localized organic results for the IP being used to search.

Maps locations in knowledge panels are a decent indicator that we’re looking at a location-specific query.
The classic map pack, particularly at the top of search results, is a strong local intent flag for us.

5) Visual Intent

These are easier to track using Image packs and thumbnails, however prominence is a factor. Many SERPs have Image packs in the top 100 results by default – placement in the top 10 is a more significant sign, as well as 2 rows of results, or perhaps top 10 rankings from sites like Pinterest.

Image packs are a strong indicator of visual search intent.
Pinterest ranking in the #1 and #3 organic results, with a prominent image pack? Yeah… that’s gonna earn a strong Visual Intent score.

6) Video Intent

Originally I was going to classify Video Intent alongside images as a Visual Intent category, but as I started reviewing more and more search results, it became clear that Video was really it’s own type of intent. Between video carousels, video thumbnails, and even video featured snippets now becoming commonplace, video is becoming critical to ranking for certain types of queries.

These video carousels are the most common indicator of video intent that we see.
Video thumbnails are another very strong indicator of video intent.

7) Fresh/News Intent

When we see News Intent in the navigation, Top stories boxes, Recent Tweets, and heavy use of recent dates in the past day/week/month in the organic results, that tells us that there is a high volume of content being produced by this topic, and we can probably infer that Google sees higher user interaction with more recent results for this topic.

Top Stories in search results are a clear sign that users are looking for news.

Recent tweets are another result type typically triggered on Fresh / News Intent keywords.

8) Branded Intent

Brand intent queries tend to be dominated by brand homepage results with sitelinks, or domain clustering from a single domain when a keyword includes a known brand entity (eg “walmart TVs”).

Large sitelinks are strong indicators of branded intent.
10 blue links for the term [airbnb north oahu] is pretty clearly branded intent.

9) Split Intent

This last one is more of a “meta intent” type. When we see signs that there are a number of potential intents a user could have, we should be triggering a “Split intent” warning to the user, since these queries are often difficult to rank in if you’re not the first result for a certain intent. EG if you’re not the first Research Intent result, like Wikipedia, your blog post may not have a shot at ranking. This is covered more heavily in the sections below.

[train] is a classic split intent SERP example, with videos, knowledge panels, recent news, and other strong signals across multiple types of intent. As the father of a 2 year old I’ve spent a lot of time on this SERP…

[seinfeld] has multiple potential intents, including users who want to stream video, research the show Seinfeld, video clips, TV air times, info about Jerry Seinfeld, etc.

While our system is currently focused on the 9 intent classifications, we’ve built our scoring process in a way that allows us to add new intent types in the future. When Google rolls out virtual reality search results, we’ll be ready for it.

How Content Harmony Is Scoring Intent

One of the challenges facing us in search results these days is that it’s very common to find a SERP with multiple potential intent options. So we need a way to indicate overlapping intent.

We also need a way to differentiate between a SERP that contains an image pack at the bottom, with a SERP that contains an image pack in the #1 ranking slot – the Visual Intent of those two searches can’t be scored the same. Intent is not a boolean “Yes” or “No” status – intent can be stronger or weaker on different keywords.

So, we developed our own scoring method.

Publicly to our users, Content Harmony will report search intent uses a scale of 0-3 depending on the prevalence of certain search features.

  • 0 – A score of zero means no signs of this intent were found in the search results.
  • 1 – A score of one means we saw some signs of this intent type, but they didn’t appear to be very strong.
  • 2 – A score of two means we see enough signs to believe this is one of the top 2-3 types of intent for this query.
  • 3 – A score of three doesn’t happen on every result we analyze, but when it does you can be sure that this is a clear intent for that keyword.

We report the Primary Search Intent back to the user depending on which type of intent scored the highest on our scale:

A snapshot of the actual intent score being shown in the beta version of our tool (scroll down for more info about early access).

Internally, we track somewhere between 4 and 20 different possible signals for each type of search intent. For example, we might have 11 different signals that suggest local intent on a search result. Each of those signals has its own raw score relative to each other, and by tracking each signal that shows up for a keyword, we calculate a raw score. Again, this raw score is entirely internal since it will evolve over time as we add new intent signals.

Our software will support API queries out of the box. An example JSON result for search intent score might look like this:



{ "primary_intent" : "Transactional",
  "intent_scores" : {
         "Transactional" : 3,
         "Branded" : 2,
         "Visual" : 1,
         "Research" : 1,
         "Answer" : 0,
         "Fresh / News" : 0,
         "Local" : 0,
         "Video" : 0
        }
}


(Formatting is subject to change of course. I’m sure our developer will tell me I missed a semicolon or something).

FAQ & Other Search Intent Considerations

“Isn’t Some of This Actually Formats, Not Intent?”

Good question! Yes, we are mingling format with intent to a degree. I believe that format and intent are heavily intertwined. Formats are simply different type of results and intent is nothing but the user wanting to see a certain type of result, so, they all go hand in hand. Describing intent in relation to format is also a much clearer way to tell SEOs and content writers what type of content they’ll need to produce to meet the intent we’re seeing for that keyword.

A user may not know they’re looking for a video when they search for [how to tie a tie], but good luck trying to rank for that search query without a video if that’s the best format to serve the user’s intent, which is seeing how the hell to tie a half windsor.

A user may not know type the words “image gallery” when they search for [house paint inspiration], but good luck trying to rank for that query without some beautiful custom or user-submitted photography. Detecting visual intent on that search result is critical to understanding what that user wants.

This question of detecting formats vs intent is closely related with my notes earlier regarding SEOs needing to understand intent differently than search engines do. We’re not trying to present the user with a set of results that makes them happy, we’re trying to figure out what the best result is to show Google to reach those users.

“Wait, If Intent Is Always Changing Then What If Your Intent Score Is Wrong A Week From Now?”

Another good question! This is absolutely correct, there are excellent examples out there of keywords where the intent can change daily, or even hourly throughout the day (I’m looking at you, Black Friday SERPs).

This tweet from JR Oakes illustrates this well (follow him, it’s worth it):

Here’s the thing – when intent changes in a way that negatively affects your content’s performance, you should be able to spot that in other tools.

Your week-over-week traffic drop alert in Google Analytics will go off. Your rankings tracker will show you lost the top spot for a keyword that you’ve been #1 on for the past 2 years. Your conversions from that page will be down.

Those are normal daily occurrences in the life of an SEO. And sometimes, your content needs to change, too!

When that happens, sometimes you’ll be able to look at the search result and see what is happening (eg a competitor with similar content took over your spot). But sometimes you won’t be quite sure what has changed, and running a new report for keyword intent is the correct way to compare before and after if you previously ran a report for that keyword.

We have an interesting example from the August 1, 2018 algorithm update of how this affected a B2B SaaS Client we used to work with. I tweeted about it in this thread:

How We’re Determining Split Intent

If a SERP scores 2 or 3 in a number of areas, that would be an appropriate time to report “Split Intent” to the researcher.

For example, in the [hockey] example below, we see multiple results indicating intent that would score 3 on our scale:

  1. Fresh Intent and news results
  2. Multiple Research Intent signs from Wikipedia, knowledge graph boxes, and PAA results.
  3. Multiple Visual Intent signs from video packs, and Image Video intent suggestions in the navigation.
  4. Multiple Local Intent signs from a local pack in the top 20 results, and localized organic results such as SeattleThunderbirds.com in my area.
Seriously [hockey] might be one of the most split intent keywords I can find. Aside from the multiple carousels above the fold, if you keep scrolling there’s a map pack, People Also Ask box, and See Results About (other entities).

In this case it would be misleading to an analyst to call this a Fresh Intent query or a Research Intent query. I would report this as a Split Intent keyword, and perhaps tell the user a ‘secondary intent’ of Fresh Intent.

How We’re Handling Unknown Intent

There will always be “10 blue link” search results where the intent just isn’t clear to us. We’re currently deciding between returning a primary intent of “None” or “General” in these cases and we’d be open to feedback.

I feel that “General” indicates the intent is unclear, but it’s different than a nil result of some type where data was missing.


Want To Take It For A Spin?

We’re just about ready to share early / beta access to our product, but we’ll be rolling out access slowly in order to test and refine and make sure we’re delivering quality data with every report.

➡ Which reminds me – keyword difficulty is just one of multiple features in the new platform. If you run an SEO or editorial team (agency or in-house) that is invested in content marketing and consistently produces 10-50+ pieces of content per month, please let us know!

If you have a project that requires a fairly large number of keyword difficulty scores to be calculated (say, 1,000+ critical keywords) and are interested in a dataset export, we’re looking for demonstration projects like that as well.


Kane Jamison

Kane is the founder of Content Harmony, a content marketing agency based in Seattle. He's also the instructor for Content Marketing at the School of Visual Concepts in Seattle, and an industry speaker at events like Mozcon, EngagePDX, SMX Social, State of Search, LavaCon, Wordcamps, and more. You can find him on Twitter at @KaneJamison.

Comments (10)

  • Fantastic read, Kane. Can’t wait to see what you’re coming out with.

    I tried to sign up but noticed /product/ is a 404. Just a heads up.

    (Can I be selfish and ask you to email me when it’s back up…not sure I will get a ping if you reply to the comment here?).

  • Very well done. Worth mentioning in Local Intent is how mobile based search can pretty much append its location to many search terms. In other words, if a user is on their phone and searching for “plumbers” Google is going to consider that search as “plumbers” + near me OR “plumbers” + city / town / current location. A fun way to see this in action is to use a VPN and something like Simulator (on MacOS) and then run the exact same search, altering location via VPN and platform via the device category of Simulator then comparing to desktop. It’s also a handy visual for clients when helping them to understand this behavior.

  • Enthusiastically signing up for the beta! Any tool that shaves some time off how long it takes to manually assess each SERP for intent is worth my $$$. Looking forward to trying it out.

  • Really curious to test your product!
    The analysis results are linked only to Google.COM or to any version of it (and it works in any language supported by Google)?

    • Just English/US results to begin with. Once we make sure that data is coming back reliably we already have the infrastructure to build out multilingual results, but we don’t want to roll out too much at once.

  • An agency related question: how do you reconcile client insistence they be featured in a top of the funnel searcher intent keyword when the page works better lower down the funnel? Is it still possible to force Google to use your page with brute force backlinks and on-page optimization or is it just time to sit the client down and explain the new searcher intent paradigm?

  • Hi Pixeo – that’s generally a good time to show the other pages ranking for the top and bottom funnel terms and how they differ, and then compare that with the page you’re trying to show. Convincing the client to have separate pages (unless it is reasonable to target both terms with the single page) is the next step – or – convincing them to have content on the page that appeals to both audiences.

  • Really exciting software and well-crafted post about the new tool. You answered questions I didn’t even know I had. Can’t wait for beta testing!

Leave a Reply

Your email address will not be published. Required fields are marked *

Get new content marketing tips via email:

Software

Contact Us

info@contentharmony.com

Mailing Address:
900 1st Ave S
Suite 205A
Seattle, WA 98134


Content Harmony

Sitemap | Privacy | Terms
© 2011-2019 Content Harmony®.