Editorial Team The wake-up call for Aurora Digital came in early 2026. The Austin-based agency was watching as its biggest client, GreenThumb Gardens, saw its organic search visibility crater by nearly 30% during the last quarter of 2025. This collapse in market share for the e-commerce gardening retailer went straight to the bottom line. It was obvious their once-reliable keyword research strategies were now failing spectacularly in the face of AI-driven SERPs, meaning they had to tear up their whole playbook for organic search.
Key Takeaways
- Your old keyword research playbook is broken in AI search. You have to switch to analyzing conversational queries instead.
- Digging into the People Also Ask (PAA) boxes and related questions in AI results is the best way to see what users actually need and the kinds of questions they’re asking.
- Use prompt engineering to simulate how an AI thinks, which helps you find the specific language and context that large language models prefer in their search answers.
- Your content strategy needs to change to directly answer long, conversational questions and give a complete picture that satisfies a user’s entire line of questioning.
- You have to constantly audit AI-driven SERPs to track changes in snippets and layouts, because if you don’t, you’ll lose your organic visibility in a heartbeat.
“We were still using tools designed for a pre-AI world,” admitted Sarah Chen, Aurora Digital’s Head of SEO. “Our reports were full of high-volume, short-tail keywords that simply weren’t converting anymore. Google’s Search Generative Experience, or SGE, was spitting out conversational answers instead of just a list of blue links.” Because of this, old-school metrics like search volume and keyword difficulty, while still having their place, couldn’t give you the full story on user intent or potential traffic anymore. The real work became understanding the actual questions people were typing into the search box and figuring out how the AI decided which answers to show.
The first audits showed a clear, and frankly worrying, pattern. GreenThumb Gardens’ product pages that used to own the SERPs for terms like “organic fertilizer” or “heirloom seeds” were getting buried under AI-generated overviews. These AI summaries pulled info from all over the web and gave people direct answers, so they never had to click through to a site like GreenThumb’s. “It felt like we were competing with Google itself,” Sarah remarked. It was clear the agency had to stop chasing simple keyword matches and start figuring out the semantics and context the AI models actually cared about.
Their first move was to completely blow up their keyword research process. They stopped starting with broad terms and instead began by digging into the actual AI-generated results for their main queries. “We spent hours manually examining SGE snapshots,” Sarah explained, looking at the ‘People Also Ask’ (PAA) sections, the related questions, and exactly how the AI built its summary responses. This manual grunt work gave them a direct window into what the AI considered authoritative. It was time-consuming, sure, but it showed that users were asking really complex, multi-part questions like, “What is the best organic fertilizer for tomatoes in clay soil, and when should I apply it?” which their old tools just couldn’t pick up on with any real accuracy.
Doing that kind of qualitative analysis by hand wasn’t going to scale, so Aurora Digital started playing with prompt engineering to reverse-engineer how the AI understood topics. They would feed common search queries into large language models (LLMs) and watch the responses to spot patterns in how the information was put together. “We’d input a query like ‘sustainable pest control for roses’ into an LLM,” Sarah elaborated, “and then pick apart the generated text for themes, specific names it mentioned, and the follow-up questions it spit out. This helped us map out the topic cluster from the AI’s point of view.” This whole process let them find long-tail, conversational keywords that their standard tools were completely blind to, usually because the individual search volumes were low even though the collective user intent was huge.
What they found was that topical authority mattered way more than old-school keyword density. It was obvious that AI models in SGE were rewarding sources that showed a deep, complete grasp of a subject, not just sites that stuffed a keyword over and over. “We changed our whole content strategy from targeting single keywords to building out entire topic clusters,” said Mark Jensen, Aurora’s lead content strategist. For the GreenThumb Gardens account, this meant they stopped writing separate articles on ‘compost’ and ‘micronutrients’ and instead built huge guides on ‘soil health for organic gardening’ that covered everything from A to Z which seemed to be exactly what the AI was looking for.
The team also started pulling in new data for their keyword analysis. They started paying close attention to the internal site search data from GreenThumb Gardens, giving them a raw, unfiltered look at what people were searching for once they were already on the site. “Internal search logs are gold,” Sarah emphasized. “They show us the real questions users have, in their own words.” They would then take this data and compare it against their Google Search Console ‘Queries’ report, looking for those long-tail questions that were getting a ton of impressions but no clicks, a dead giveaway that there was an opportunity to create content that could give a direct answer in the AI results.
A perfect example of this new strategy working was for the query “how to start a no-dig garden.” GreenThumb Gardens already had a few blog posts that mentioned no-dig gardening, but none of them gave a single, complete answer that an AI could easily grab and summarize. Using their new process, the Aurora Digital team saw this was a high-intent, conversational query popping up all over the PAA sections. So they built a massive, step-by-step guide called “The Complete Guide to No-Dig Gardening: From Soil Prep to Harvest,” structuring it with clean headings, bullet points, and a big FAQ section at the end. The guide took off, ranking on its own and also getting pulled into featured snippets and AI summaries, which is exactly where you want to be since, according to a Statista report on SGE adoption, users are leaning more and more on these AI answers.
They also started analyzing the search result layouts for different types of queries. “For transactional queries, AI might still show you product listings,” Mark observed, “but for informational queries, it wants direct answers and structured data. We had to change our content format to match what the AI wanted to show.” This led to a much bigger focus on schema markup, especially for FAQs and how-to guides, which basically spoon-feeds the information to the AI models. They also began a routine of spying on competitor content that was getting featured in SGE snapshots, breaking down its structure, its depth, and how it used visuals.
The team even developed a process they started calling “AI persona mapping.” It’s basically a thought exercise: if an AI were a person answering a question, what kind of info would it lean on? Would it prefer expert opinions? Hard scientific data? Maybe user reviews? Figuring out these likely biases allowed them to tune their content to be exactly what the generative AI algorithms were looking for. “It’s like writing for a very intelligent, very literal editor,” Sarah mused. “You have to be clear, concise, and cover all the bases.”
And GreenThumb Gardens saw real results. Within six months of putting these new content and keyword research strategies into practice, their organic visibility started to bounce back, eventually shooting past its previous highs. They measured a 25% jump in organic traffic to their informational content and, more importantly, a 15% lift in conversions on product pages that were now supported by these deep, AI-friendly guides. The big lesson for the agency was that keyword research in the AI era is about understanding and predicting what a user really wants and then building your content so an AI can easily digest and serve it up.
If this whole experience taught them anything, it’s that search is always changing and using a static playbook is a recipe for failure. Now, Aurora Digital runs weekly audits of SGE results for all their core queries so they can adapt their content strategy on the fly based on how the AI is presenting information that week. This is a non-stop cycle of learning and tweaking, all based on the fact that AI is actively interpreting the web and changing how people get information, not just indexing pages.
If you want to win in AI-driven organic search, you have to get dynamic and focus on intent in your keyword research, which means getting way past your old tools and developing a real understanding of how AI actually processes and presents information.
Is search volume still a useful metric with AI search?
It’s less useful for short-tail keywords. Because AI-generated summaries answer questions directly in the SERP, users don’t need to click through, making high volume on its own a misleading metric. Your focus should shift to answering the conversational queries that AI is trying to synthesize and covering a topic completely.
What’s “AI persona mapping” for keyword research?
It’s a method where you try to figure out what types of information an AI prioritizes for a given query, does it prefer expert quotes, data, or user reviews? Knowing this helps you tailor your content to match the AI’s preferences, making it more likely to be used in a generated answer.
Why should I pay attention to “People Also Ask” boxes?
Because PAA boxes are a goldmine. They show you the exact follow-up questions real users are asking, which reveals their true intent and gives you a blueprint for the conversational style that AI models are looking for. Using these insights helps you build out content that answers every part of a user’s potential question.
How do I use prompt engineering for keyword research?
You feed your target search queries into an LLM and then you pick apart the response it generates. Look for themes, specific people or products it mentions, and what follow-up questions it suggests. This is a great way to find conversational, long-tail keywords and see how an AI connects ideas within your topic.
How should my content strategy change for AI search?
Stop optimizing for single keywords and start building out complete topic clusters to show you’re an authority. This means writing long, detailed guides, using a lot of structured data like schema markup for FAQs and How-Tos, and formatting your articles so an AI can easily pull direct answers for its summaries.
