Project Horizon: Semantic Shift Drives 2026 ROAS

It’s 2026, and digital marketing is way past just matching exact search queries. If you want your campaigns to perform, your strategy’s performance now hinges on understanding semantic keywords. We ran an initiative in Q1 2026 called “Project Horizon” that was set up to move beyond old-school keyword targeting to get at what users actually mean which forced us to rethink our entire approach to content relevance. So, did all that work actually move the needle on our metrics?

Campaign Teardown: Project Horizon’s Semantic Shift

The goal for Project Horizon was straightforward: drive more qualified leads for our new B2B SaaS platform, which does AI-driven data analytics for mid-market enterprises in North America. We had a strong suspicion that our old exact-match keyword strategies were bringing in traffic but completely missing a huge chunk of the audience that uses more conversational, intent-based searches. So we ran this campaign from January 1 to March 31, 2026, specifically to test that hypothesis.

Strategy: From Keywords to Concepts

The big change with Project Horizon was ditching the simple keyword list and adopting a semantic keyword cluster model. We stopped obsessing over just terms like “AI analytics software” and instead dug into the broader topics and questions real users have. To do this, we got our hands dirty with NLP tools, mostly using Semrush’s Topic Research and Ahrefs’ Content Gap features, which let us map out related concepts and problems people were trying to solve. We tore through competitor content, forum threads, and our own anonymized support transcripts to build a full picture of user intent. So, for “predictive analytics,” we started exploring phrases like “how to forecast sales accurately with AI” and “challenges in demand forecasting.” This work let us create content that spoke directly to those specific queries. We made the call to put 40% of the ad budget into creating content for these semantic clusters, with the other 60% going to our usual product-focused ads.

Creative Approach: Answering the “Why”

Our creative team then built a set of long-form articles, whitepapers, and videos that articulated solutions to the problems we found in our research, rather than just listing product features. A top-performing piece, for example, was titled “Eliminating Data Silos: A Guide to Unified Business Intelligence in 2026,” hitting a major pain point we’d uncovered. The CTAs in these assets were softer too. We offered a “resource kit” or an “industry benchmark report” instead of pushing for an immediate demo. The whole idea was to build trust and attract people much earlier in their buying process. Our ad copy changed, too, moving to questions and problem statements. An ad would ask, “Struggling with fragmented data? Discover unified AI analytics,” which felt a lot more helpful than just yelling “Buy AI analytics software.”

Targeting: Precision Through Intent

We used these semantic insights to really tighten up our targeting. In Google Ads, we leaned into phrase match and the new-and-improved broad match, loading them with our long-tail semantic phrases and keeping a close eye on the search query reports for new ideas. We also went hard on Google’s custom intent audiences, feeding them lists of URLs and keywords from our research to find people who were already reading about these topics. Over on Meta Ads Manager, we built interest-based audiences around our semantic clusters, like “business intelligence trends” or “data science for SMBs.” This let us find users who weren’t looking for our exact product but were clearly interested in solving the problems our platform is built for.

Campaign Performance Metrics: A Data-Driven Verdict

The campaign ran for 90 days on a $150,000 budget. Here’s the raw data on how it performed:

Impressions went down, which we expected when we stopped chasing high-volume exact match terms, but our CTR shot up by 47.4%. That told us we were reaching the right people. Even better, total conversions grew by 31.6%, and our CPL fell by 22.6%. The ROAS going from 1.5x to 2.1x shows how much more efficient this approach was. For context, HubSpot’s 2026 Marketing Report puts the average B2B SaaS ROAS around 1.8x, so we were pretty happy with where we landed.

What Worked: Precision and Engagement

The biggest win was how much our content relevance improved. By creating content that actually answered questions and solved problems, we saw way higher engagement. People spent 2.5 times longer on our semantic-driven pages than on our old product feature pages. That deeper engagement led directly to more conversions, proving that answering intent first really works. The long-form content, especially our whitepapers, was gold for capturing leads at a lower CPL than the direct demo requests. We also stumbled into a whole new market segment. By combing through search query reports, we found a huge amount of interest in “AI ethics in data governance”, something we hadn’t targeted at all. We spun up a mini-campaign around it, and that topic ended up being responsible for 15% of our Q1 conversions and even sparked internal talks about new product features. That’s the real payoff of applying latent semantic indexing (LSI) principles. It shows you what people care about right next to what you’re already selling.

What Didn’t Work: Overly Broad “Pillar” Content

Not everything was a win. Some of our first big “pillar” content pieces, like “The Complete Guide to AI in Business,” were a flop. They were so broad that they tried to be everything to everyone and ended up converting almost no one. We learned that while big topics are great for organic traffic, you need much more focused, granular content to actually get a conversion. The conversion rate on those huge guides was a dismal 0.8%, compared to 3.5% for our more targeted pieces. Also, our initial budget split for Meta Ads was off. The platform’s interest-based targeting just couldn’t find the super-specific, high-intent users we were after as well as search did, which meant we burned some impressions. It looks like when it comes to pure intent, search is still king.

Optimization Steps: Refining the Approach

We immediately made some changes for Q2 2026 based on what we learned. First, we’re now using a “hub and spoke” model for content, where a broad pillar piece just serves to link out to much more specific, intent-driven articles that are built to convert. Second, we rebalanced the budget, pulling 15% from Meta and putting it into Google Ads, specifically into Performance Max campaigns and more refined custom intent audiences. We’re also getting much more aggressive with our negative keyword lists in Google to clean up the traffic. Third, we built a new internal content scoring system. A new piece of content has to hit a certain “semantic relevance score” to get published, which basically forces the writer to address a specific user problem. This has already cut our content production time for Q2 by 10% because the briefs are so much clearer. Finally, we’re now scoring leads differently based on whether they came from a direct product search or our semantic content. The leads from the semantic content might be earlier in the funnel, but we found they had a 20% higher close rate over 90 days, which is a huge insight for prioritizing the sales team’s time. This move toward semantic keywords and content relevance isn’t a fad. It’s a fundamental change in how we have to connect with our audience. When you understand user intent this well, you not only make your campaigns more efficient, you find new ways to grow the business.