HomeFix Pro: 3.5x ROAS with AI in 2026

If you don’t have a content strategy for AI assistants, you’re already falling behind. Voice search isn’t a “nice to have” anymore, and your competitors are already eating your lunch on that channel if you’re not there. So how do you actually make content that AI assistants can find and that brings in real business?

Campaign Teardown: “Voice of the Customer” for HomeFix Pro

Here’s a breakdown of a three-month campaign we called “Voice of the Customer.” The client was HomeFix Pro, a home services company covering Atlanta’s northern suburbs (Marietta, Alpharetta, Roswell). The goal was simple: get more service requests from voice searches on Google Assistant and Alexa. Our bet was that if we optimized for how people actually talk and spelled out exactly what services were offered, we could catch users with a high intent to buy right when they needed help. We ran this from Q2 to Q3 2026.

Budget Allocation and Key Metrics

The total campaign budget was $85,000. Here’s how we sliced it up:

• Content Development & Optimization: $30,000 (focused on FAQ generation, schema markup implementation, and content rewrites)
• Paid Voice Search Ads (Google Assistant Action integration): $40,000
• Analytics & A/B Testing Tools: $10,000
• Contingency: $5,000

And here are the numbers at the end of the campaign:

• Total Impressions (Voice Search): 1.8 million
• Voice Search Click-Through Rate (CTR): 3.2%
• Total Conversions (Service Requests): 950
• Cost Per Lead (CPL): $89.47
• Return on Ad Spend (ROAS): 3.5x
• Conversion Rate (Voice Search): 12%
• Cost Per Conversion: $89.47

Yes, a CPL of $89.47 is higher than you might see on other channels. But for HomeFix Pro, it paid off because these leads came in hot. They converted faster and their average contract values were higher, which made the cost per lead completely worth it.

Strategic Pillars: Understanding the Conversational Interface

Our whole strategy came down to three things: conversational keyword research, getting our structured data right, and crafting natural language responses. We knew from the jump that old-school SEO keyword stuffing wouldn’t work here. Nobody types “plumber Atlanta cost” into a search bar anymore, let alone says it out loud. They ask their phone, “Hey Google, how much does it cost to fix a leaky faucet in Sandy Springs?” For keyword research, we dug into their existing search console data, pulling out long-tail queries that were already getting some traction, and combined that with work in dedicated voice search tools. We went all-in on “near me” searches, question-based queries, and phrases that described a specific problem and its solution. So instead of just “HVAC repair Atlanta,” we were targeting stuff like “AC not blowing cold air Roswell GA” or “furnace maintenance Marietta.” This level of detail in voice search optimization was the only way this was going to work. We also just sat down with the customer service team at HomeFix Pro. Listening to how real customers describe their problems over the phone gave us some of the best raw material. That qualitative work was how we ended up with a massive list of over 500 hyper-specific voice keywords and phrases.

Creative Approach: From Text to Talk

The creative part was all about turning HomeFix Pro’s existing web content into something an AI assistant could actually use. This meant a lot more than just copying and pasting text. We had to think about the back-and-forth of a real conversation. We started by building out detailed FAQ sections on their site, making sure to wrap everything in the right schema.org markup, especially `HowTo` and `FAQPage` schema. Doing that work up front lets search engines and AI assistants pull clean answers for questions like “How do I unclog a drain?” or “What’s the average cost of water heater replacement in Alpharetta?” We went through every service page to make sure it could spit out a clear answer to a direct question. There’s even an IAB report floating around that says structured data is a huge driver for voice visibility, with something like 60% of voice answers coming straight from content that’s been marked up properly. On the paid side, we used Google Assistant Actions. This let us build custom voice apps that a user could trigger on the spot. Someone asking, “Hey Google, find a plumber for a burst pipe near me,” might get an immediate prompt to connect with HomeFix Pro. The ad copy itself had to be incredibly tight: short, focused on the benefit, and have a clear call to action. Something like: “HomeFix Pro: 24/7 emergency plumbing. Say ‘Connect me’ to schedule immediate service.”

Targeting and Segmentation

For targeting, we kept it tight, focusing on the geographic areas inside HomeFix Pro’s service radius, Fulton, Cobb, and Gwinnett counties, specifically a 15-mile bubble around their hubs. We also layered on some demographic data, going after homeowners in income brackets that suggested they’d own smart speakers and actually use them to find services. The great thing about voice targeting, especially with Google Assistant, is the built-in intent. Someone asking for “HVAC repair” isn’t just kicking tires. They have an immediate need.

What Worked Well

Going deep on long-tail, conversational keywords was our single biggest win. The content team churned out hundreds of these little content snippets, each one designed to answer one very specific question. A query like “My garbage disposal is humming but not grinding” would get a direct, useful response from the HomeFix Pro integration, maybe offering a quick troubleshooting tip and then a link to book a technician. That level of specificity is what really moved the needle on conversion rates. The constant A/B testing of the AI responses was also a huge factor. At first, some of the answers were just too bland. We looked at where users were dropping off and tweaked the language. We learned, for example, that being direct and asking, “Would you like to speak to a technician now?” converted way better than a weak “Can I help you further?” Just making that one shift to more explicit, action-focused language boosted our voice search conversion rate by 12% during the campaign. Integrating `schema.org` markup was also absolutely key. We saw it plain as day in the data: pages with complete `FAQPage` and `HowTo` schema just performed better in voice results and got picked up by AI assistants as the source of truth. This is about becoming the authoritative answer for a query, which is a much stronger position than just being visible.

What Didn’t Work as Expected

We definitely stumbled on the creative at first. Some of our initial voice ads were way too formal because we were trying to write them like traditional search ads, short and packed with keywords. That doesn’t work in voice. People expect a conversation. An ad like “HomeFix Pro: HVAC solutions, expert service” just fell flat. We had to pivot, fast. Attribution for organic voice search was (and still is) a headache. You get clean tracking on the paid side, sure, but figuring out the direct ROI of your organic content for voice is tough. Is there even a reliable “voice search organic conversion” metric in most analytics platforms yet? We had to use proxies, looking at unexplained lifts in direct traffic, brand name searches, and service requests that didn’t come from other channels, to gauge success. It makes it hard to tell the client exactly which piece of content paid for itself.

Optimization Steps and Adjustments

Halfway through the campaign, we started A/B testing everything about our AI assistant responses: opening lines, CTAs, how much detail to give up front. We discovered that a short, direct answer followed by an offer to connect to a person (like, “Yes, a humming garbage disposal often indicates a jam. Would you like to speak to a technician at HomeFix Pro to schedule a repair?”) worked much better than long-winded troubleshooting guides. That change alone dropped our cost per conversion by 18% in the second half of the campaign. We also had to go back and expand our keyword list to account for how people *actually* talk, including mispronunciations and slang. We updated content to catch these variations so the AI could still make the match. Think about “HVAC”, some people say “H-V-A-C,” others say “H-vac.” You have to account for both if you want to maximize your reach. Finally, we set up a process for continuously monitoring voice search query logs (the parts that Google Search Console lets you see, anyway). This let us spot new questions as they popped up and tweak our existing answers. That kind of proactive work is what keeps a content strategy for AI assistants from getting stale. The “Voice of the Customer” campaign really showed that if you actually dedicate resources to AI assistant content, you can get great results. The key is rethinking how people get information in a conversational format, not just slapping a new coat of paint on old blog posts. To win here, you need a solid grasp of natural language and a willingness to constantly test and tweak your approach through iterative optimization.