Schema markup has long been debated in the SEO community as an essential component of any SEO strategy. On one side, there are the people who take what Google says at face value, in that it only impacts search appearance via rich snippets and contextual understanding. On the other side of the fence, a belief is held that if Google recommends structured markup through schema, then there must be some direct ranking influence.

Throughout the years, we’ve seen LinkedIn influencers, anecdotal “evidence”, and correlational studies that seem to give credit to each side. Anecdotal experience isn’t verifiable and typically lacks an appropriate sample size. Correlative studies show correlation, not causation, and don’t account for independent variables between subjects.

Most people and agencies aren’t set up to conduct massive experiments like this to isolate variables. Variables like industry, business location, authority, website scale, tech stack, and site structure are often additional variables that can’t be isolated.

That’s what makes our agency perfect to conduct these studies and determine what is fact and what is myth.

During a 10-week study at Evergrow Marketing, we put 29 clients in a controlled environment to see exactly how schema markup affects SERP rank across Google, Google Mobile, Bing, and Yahoo.

The study split the subjects into a control and a test group. 

• All within the landscaping industry
• All within the US and split by region
• All other SEO efforts were paused
• All sites of similar size and authority
• All sites on the same Content Management System (CMS) and built by our agency using the  same website template

The results weren’t quite what we expected.

The original methodology was an 8-page Google doc and has been reformatted into collapsible sections in the Methodology section of the original schema study.

For the sake of context, I’ll include the shorter version of it here.

A background of our agency and the study

A quick background of our agency gives more context to the study and why this industry was chosen for the experiment.

We’re a digital marketing agency for landscaping and lawn care businesses (as well as other green and outdoor living businesses). Our clients are strictly local and are extremely similar in scale, strategy, and services.

On top of this, we require nearly all new clients to have their site built or rebuilt by us using a templated site theme we built on WordPress. The only real difference between the sites is the branding, images, content, and localization.

The purpose of this study was to determine whether or not schema markup was an effective method for improving rank for our clients, given their local business classification.

The schema markup we used

One of the main considerations for this was determining which schema to use. While some client sites had blog posts, FAQs, and job listing pages, others did not. We needed schema markup that could be consistent with every test subject.

So with that, we went with LocalBusiness markup from schema.org. We had even consulted with Jarno Van Driel, Yoast, and Moz on the items and types within the LocalBusiness markup to be used in this study.

The biggest consideration in the schema markup used was the idea that we wanted to isolate the mere fact of the schema being added to the site, which affects rank, and not the idea that a higher click-through-rate (CTR) through rich snippets affects rank.

Aside from JobPosting schema, there isn’t much in the local service industry that produces rich results, especially with FAQ rich snippets being deprecated completely and only used on government and health organizations after August of 2023.

With that in mind, we avoided any schema markup that aided in producing rich snippets that were still applicable, such as priceRange or reviewRating schema.

The full markup used can be found in the original study under the “Methodology” section.

The pre-test & client sorting

With any controlled experiment, the hardest part is determining subjects that are eligible for the test. Out of our 50 clients, we found 29 that were eligible.

These 29 were:

• Only on our base-level SEO services
• Located in the US
• Had a site built by us (WordPress using Divi)
• Were not producing new or regular content
• Had been continuously working with us for six months or longer

These 29 clients would then be sorted into two groups:

• Control Group
• Test Group

The Control Group would have all SEO efforts paused and no schema added to their site. The Test Group would also have all of their SEO efforts paused; only this group would receive the LocalBusiness schema. 

A unique challenge with our industry is that search volume and competition are seasonal. Knowing that we were conducting this experiment between February and April, it’d be landing right at the beginning of the spring rush. Our Southern clients don’t feel that rush as much as our Northern clients coming out of a snowy winter.

To control the seasonal variable, we’d ensure both the control and test groups would have an equal number of clients in various US geographical locations.

The Control Group had 13 clients:

• Six South
• Three Northeast
• Three Midwest
• One West

While the Test Group had 16 clients:

• Six South
• Three Northeast
• Six Midwest
• One West

Note: The groups were more evenly divided by locations, as each client had varying numbers of locations. Despite the number of clients in each group, the number of Google Business Profile locations in each group was 18.

How we cleaned the environment and benchmarked

Part of any controlled study is sterilizing the environment to avoid any kind of bias or contamination of results. What that meant for our study was completely removing any and all schema from all clients in the experiment.

Since adding schema markup of any kind wasn’t an existing practice at our agency, the only schema we needed to remove was breadcrumb schema natively added by the Yoast plugin we use on all sites.

We then verified up-to-date indexing and crawl reports via Google Search Console to ensure that it wasn’t picking up any schema, and the sites were properly indexed in Google, given their new, sterilized environment.

Defining the control and test periods

Once the schema was removed and we had verified no schema was on any site, we waited a period of 5 weeks for any potential schema-related rank changes to level out.

This was the Control Period.

After the Control Period, we entered into the five-week testing period, where the schema would be added to the Test Group only.

A benchmark report would be created prior to day one of the five-week testing period, and a final report would be created on the last day of the testing period for all subjects.

The search engines and queries we tracked

When “SEO” is mentioned, everyone immediately goes to Google. Obviously, Google has the highest market share by a mile, but it’s still worth it to see how other search engines treat schema markup.

For that purpose, we tracked all of the results in the following search engines using Moz Pro:

• Google
• Google Mobile
• Bing
• Yahoo!

If Google doesn’t care, maybe Bing does.

For the purpose of this test, we tracked two different types of queries:

• General query
• Specific query

A general query would be “[service] company in [target location]” while a specific query would be, “can you recommend a [service] company that services the [location] area that’s open at noon on friday?”

Any SEO is likely going to only be concerned with the general query results, but since schema markup was created to provide specific, entity-based information to search engines, it only made sense to include a specific query that a search engine could pull information from.

A good test doesn’t simply take the average position of one group and compare it to the other at the end of the testing period. One outlier could throw off the entire data set.

Good studies will typically use a Welch’s T-Test, which calculates the confidence as a percentage in which the controlled variable caused the improvement in rank.

For this test, we used a One-Tailed Welch’s Two-Sample T-Test. This test does a few things that calculating benchmark and final averages don’t account for.

This formula not only considers the overall average in rank change, but also the standard deviation in each subject’s change in rank. It’s not enough to say one group improved on average over the other; this formula creates a statement of confidence that says:

“Not only did this group improve over the other group, but I am also 90 percent confident the variable introduced will improve rank because out of the 16 subjects in the test group, 14 of them saw improvement vs only seven out of the 13 in the control group, and the improvements were statistically significant.”

A quick note on confidence intervals: when you see confidence percentages in this report, understand that 50 percent is essentially a coin flip. A confidence of 50 percent won’t tell us anything.

Likewise, a confidence interval of 75 percent isn’t an indication that the variable was the cause of the rank improvement. Medical studies typically accept confidence levels that exceed 95 percent. Anything less than that is uncertain or unlikely.

We’re not in the medical field. Schema markup is a relatively low-effort practice, so for the purpose of this study, I’ll accept anything as a fact that is 90 percent or higher. I do, however, give grace to results that show 80 percent or higher.

I can’t be certain that 80 percent confidence means schema does have an impact, but it could, and it could not.

Having established that, let’s get into the results.

The results of the general query

In full transparency, I started this experiment with the hypothesis that schema markup did absolutely nothing for rank in Google. I’m always open to the idea of being surprised; unfortunately, Google didn’t do that for me.

The biggest surprise was that schema markup didn’t confidently affect Bing at all.

I figured out of all the search engines, Bing would have had a confident impact given Bing’s Principal Program Manager, Fabrice Canel’s comments on how Bing uses schema to help their LLM understand context.

This test showed that Google likely doesn’t need schema to understand context, and Gemini is able to utilize Google’s index and knowledge graph for contextual understanding. If CoPilot is using it for contextual understanding, why isn’t Bing?

Yahoo falls into the “potentially” category. There isn’t a 90 percent or higher confidence, but 81.96 percent is still significant. I would believe schema markup impacts Yahoo, as they likely don’t have the resources to understand context without it, like Google.

For Google, we can definitively say that schema markup has absolutely no direct impact on rank.

Note: When we see flat 0.00 percent, that indicates the test group did not perform better than the control group on average. In order to run the Welch’s T-test, the test group must perform better than the control group; if it does not, then “0.00%” must be entered.

The results of the specific query

We were curious to see how schema might affect ranking when specific information was asked within a query, forcing search engines to look for objectivity instead of solely subjectivity.

The results were, yet again, surprising.

We would have expected Yahoo to be right up there in the 80-90 percent range, as it was with the general query, but this time the control group outperformed the test group.

The other surprising results were with Google Mobile and Bing, both showing “potential” influences with LocalBusiness schema markup when a specific query is entered.

For this example, we’re asking the search engine to find very specific information about a business.

It makes me wonder if Google and Bing prefer showing objective information from sources with the most complete dataset.

For what it’s worth, as an SEO, I don’t particularly concern myself with the results from the specific query test for two reasons:

1) It’s too specific

This query is so specific that there is almost no volume for it in SERPs. I’d put more weight on this for Google Mobile and Bing if these were results for the general query test.

2) It’s likely a query for LLMs, not SERPs

It’s a query that would likely be asked in AI mode or with an LLM, in which case, any SEOs would be more focused on the query fan out and grounding queries of the retrieval than the input query itself.

The search visibility results

The last test we conducted using Moz was reviewing how adding the LocalBusiness schema impacted the search visibility report. In other words, the percentage of clicks we can expect to get out of ‘x’ number of searches for the keywords we’re tracking.

With each higher position each subject secures in SERPs, their search visibility improves. If a subject was at position 10 on page one, their search visibility might be 10 percent. If they bump up to and secure position one, the visibility may go to 50 percent.

Anything past position 20 (page two) is 0 percent.

The results of the Search Visibility test are the most important, in my opinion. In all search engines, the control group outperformed the test group on average, and the T-test wasn’t able to be conducted.

That means that the “potentially” label we were giving to Yahoo for the general query, as well as Google Mobile and Bing for the specific query, is likely “no”. 

Sure, there were rank increases more often than not in the test group, but the Search Visibility test shows that those rank increases weren’t even enough to make a significant impact or bring the average position above number 20 (page two of Google).

Given the results of this study, we can be confident in the idea that adding schema markup to a site or page does not improve rank, nor is it any kind of signal that influences rank.

That does not mean that schema markup is inherently useless. On the contrary, schema markup is an incredibly valuable tool in a lot of industries and applications.

If you want more visibility on SERPs, using rich snippet-producing markup is an intelligent way to stand out from the competition.

Unfortunately, in the service area business, rich-snippet-producing markup is limited to only a few options, like priceRange and the review property under the Product type.

Note: Be careful when marking up reviews on service area business websites, as it is against Google’s review markup policies to markup aggregate third-party reviews and reviews owned by the site owner, as it’s seen as a conflict of interest. Reviews that are marked-up must be user-generated for a specific product or service using the review property under the Product type and displayed only on the page about that product or service.

Where schema markup is imperative

Certain applications and industries almost require schema markup as an SEO strategy. One example local businesses can take advantage of is the JobPosting schema.

When you search for “landscaping jobs near me” in Google, the only way to show up in the results below is by marking up the job listing page with the appropriate JobPosting schema.

These are the first results Google shows before the organic listings. Without this schema type, your job post is virtually invisible on Google.

Product schema is another important type for displaying products in traditional search. Sure, you can use Merchant Center to show up in Google Shopping, but without product schema, your product details are ineligible to show up in traditional search results.

How LocalBusiness schema fits into an SEO strategy

LocalBusiness schema is what I consider “non-rich-snippet-producing schema”. It’s schema that makes absolutely no impact on search appearance or rank.

However, my stance on the matter is this:

LocalBusiness schema is incredibly easy to implement and you only have to do it once. We know that Google and other search engines crawl and contextualize content within schema, so it’s simply additional information that search engines can rely on.

It’s good housekeeping for local businesses.

Consider AI and LLMs

While foundational LLMs rely on their pre-trained knowledge, modern AI search engines use Retrieval-Augmented Generation (RAG) to fetch real-time data from the web. Because processing large amounts of retrieved text consumes tokens (which costs compute power and money). This idea gives credit to 'chunking' to feed only the most relevant snippets to the AI.

It’s also arguable that schema markup like JavaScript Object Notation for Linked Data (JSON-LD) is beneficial to token consumption. Its strict structure provides unambiguous context. By explicitly defining entities, prices, and relationships, JSON-LD allows automated AI web-scrapers and RAG systems to parse and understand page content with perfect accuracy, heavily influencing what data the AI chooses to include in its final response.

I see a future world in which these systems prefer JSON-LD to prevent hallucinating. On the flipside, I see a world in which LLMs become good enough to not need it.

Fortunately for you, we actually tested three different LLMs in the full test. You can check it out in the full study at Evergrow Marketing.

Despite the results of this study, this shouldn’t deter anyone from using schema markup or have them treat it as the antiquated strategy of placing meta keywords.

There are still very applicable use cases for it, and unknown variables.

It’s simple to add, and you can even check schema implementation using the free Mozbar.

Keep in mind that SEO is more than just “ranking”. It’s providing the best search experience possible for you or your clients. You can rank number one for a query but still lose clicks because position #3 has product markup schema and rich results.

Always be sure to follow Google’s guidelines on schema markup. For something that doesn’t inherently improve rank, they sure do give harsh manual action warnings.

Improperly using markup is high risk with little reward.

And as always with SEO, keep an open mind when you hear or read something new. What worked five years ago may no longer be relevant today. Don’t take anecdotal evidence as truth. Test for yourself and use the strategies that are best for you or your clients in your industry.

For more practical tips on how and when to implement schema markup, check out Moz’ beginners guide on structured markup.