Keyword Research for Google and AI Search: A Dual-Channel Approach

Keyword research used to mean one thing: find what people type into Google, assess volume and difficulty, create content, rank. That playbook still works - but it now covers only half the discovery surface your audience uses. ChatGPT processes approximately 2.5 billion prompts daily, representing about 12% of Google's search volume and surpassing Bing's market share to become the second-largest search platform in the world.

Nearly a third (31.3%) of the US population will use generative AI search in 2026, according to an EMARKETER forecast.

If your keyword research process hasn't changed since 2023, you're optimizing for one channel while ignoring another that's growing exponentially. Organic click-through rates on queries with AI Overviews dropped 61% between June 2024 and September 2025. Meanwhile, the traffic that does come through from AI-referred visits tends to be higher quality - spending 68% more time on-site and converting at higher rates than traditional organic visitors.

The practitioners getting ahead aren't choosing between traditional SEO and generative engine optimization. They're running both processes in parallel, feeding the outputs of each into a single content strategy. Here's exactly how to do that.

Why a Single Keyword List No Longer Covers Both Channels

The fundamental problem is structural. Google and AI search engines discover, evaluate, and surface content through different mechanisms. Traditional search matches keywords to indexed pages and ranks them in a list. Unlike traditional search, where results appear as a list of links, AI engines synthesize information from multiple sources into a single conversational response.

This distinction matters for keyword research because the queries users type into each channel differ sharply. While traditional Google searches average 4.2 words, ChatGPT prompts average 23 words, allowing for more context and nuanced responses. A Google user might search "best CRM software." A ChatGPT user asks, "What CRM should a 15-person B2B agency choose if they need HubSpot-level automation but at half the price?" The overlap between what ranks in Google and what gets cited by AI is surprisingly thin. Only 12% of URLs cited by ChatGPT, Perplexity, and Copilot rank in Google's top 10 search results, and 80% of LLM citations don't even rank in Google's top 100 for the original query. Building one keyword list and hoping it serves both channels is a strategy with a measurable failure rate.

Traditional Keyword Research: What Still Works and What Needs Adjusting

The core mechanics of keyword research for Google haven't been invalidated. Volume, difficulty, intent classification, and SERP feature analysis remain essential. But you now need to add a new dimension: AIO vulnerability assessment.

AI Overviews now appear for 30% of U.S. desktop keywords - a new high as of September 2025.

In a study of over 300,000 keywords, there was a correlation between AIO presence and a 34.5% decrease in click-through rates for top-ranking pages. Not every keyword you're targeting is equally exposed, and understanding which ones carry AIO risk is now a required step.

Filtering Keywords by AI Overview Exposure

Start with your existing keyword list from Semrush, Ahrefs, or similar tools. Semrush now shows which of your ranked keywords trigger Google AI Overviews and whether your site is cited in them - this is critical for understanding your real organic visibility in 2026. Ahrefs offers similar capability through its clickstream data, which helps identify zero-click keywords where search volume is high but traffic potential is low. For each keyword cluster, categorize terms into three buckets:

• AIO-safe: Keywords that don't trigger AI Overviews (transactional queries, branded terms, local queries).

AI Overviews only have a 10% chance of showing for commercial or transactional keywords.

• AIO-competitive: Keywords where AI Overviews appear but your content has a realistic chance of being cited as a source.
• AIO-vulnerable: Informational keywords where AI Overviews dominate and your domain lacks the authority to earn a citation.

Informational queries - the kind that start with "how," "what," "why," or "best" - are by far the most likely to trigger an AI Overview, with 88% of queries that produce an AI answer being informational in nature. This doesn't mean you abandon informational content. It means you adjust your traffic expectations and pursue those keywords with an AI citation strategy in parallel.

Intent Classification Gets a New Layer

Traditional intent buckets - informational, navigational, transactional, commercial - still apply. But with the rise of AI tools like ChatGPT, SEOs are now seeing a sixth type: generative search intent. These are queries where users expect AI to synthesize, compare, plan, or draft - not just retrieve information.

With AI tools like ChatGPT, Gemini, Perplexity, and Google's AI Overviews, users now phrase their needs as longer, more natural prompts that often combine multiple intents at once. A single query like "Compare three affordable smartphones and recommend the best one for students" blends commercial and transactional intent in a way that traditional keyword tools wouldn't surface. Map every keyword in your list against both traditional intent and AI-trigger likelihood. Tools like Keyword Insights use machine learning to classify intent at scale, while manual SERP checks reveal whether Google is already serving AI Overviews for a given term.

Prompt Research: The AI-Side Equivalent of Keyword Research

Prompt research analyzes the questions people ask generative AI systems and how those prompts shape the answers those systems produce. In practice, it functions as the AI-era extension of keyword research: traditional keyword research analyzes search queries, ranking opportunities, and competition within the results page, while prompt research focuses on the prompts that lead AI systems to explain topics, compare options, or recommend specific tools, products, or brands.

This isn't speculative. It's already a formalized discipline. SEO teams have always lived and breathed their keyword research - they built methodologies for choosing terms through user research, intent mapping, volume thresholds, difficulty scores, and SERP features. AI search works the same way at a higher resolution: prompts are the new measurement unit and they need a methodology just as rigorous as keyword research.

How to Conduct Prompt Research

The methodology follows five steps: 1. Identify your personas and their decision stages. A better approach, and one that actually scales, is to optimize for personas - your users will adopt different AI assistants for different reasons. Map the questions each persona would ask at each stage: problem awareness, solution exploration, comparison, and purchase decision. 2. Test discovery prompts across platforms. Open ChatGPT, Claude, Perplexity, and Gemini in separate browser tabs. For each platform, test prompts representing how your target audience would ask for solutions: "What are the best tools for [your category]?" or "How can I solve [problem your product addresses]?" or "Compare solutions for [your use case]."

3. Document which brands appear and which sources get cited. Create a simple spreadsheet with columns for the AI platform, the prompt used, whether your brand was mentioned, which competitors appeared, and the context of any mentions.

4. Build prompt clusters. Prompt research expands the scope of content strategy beyond ranking individual pages to clusters of related questions. For SEO, that means ensuring content covers the full topic landscape. For GEO, it means ensuring content provides the context generative systems need to synthesize answers.

5. Cross-reference with keyword data. Traditional keyword research retains significant value in prompt research methodology but serves a different function - rather than identifying optimization targets, keyword research validates how audiences naturally describe problems, frame constraints, and express intent. This language provides the vocabulary for constructing natural-sounding decision-stage prompts.

The Data Gap Problem

One fundamental challenge separates prompt research from keyword research: AI platforms do not publish query volume statistics, user demographic information, or seasonal trend patterns for prompts. This data scarcity fundamentally changes research methodology. Rather than relying on volume-based prioritization, prompt research focuses on qualitative factors: persona characteristics, decision constraints, competitive positioning, and recommendation patterns.

This means you can't simply filter prompts by volume the way you filter keywords. Instead, prioritize prompts where:

• Your brand is absent but competitors appear
• The AI gives incomplete or inaccurate answers that your content could fix
• Multiple prompts cluster around a decision point your audience cares about

Making Content That Serves Both Channels Simultaneously

Here's where the dual-channel approach pays off. The winning strategy isn't choosing between traditional SEO and generative engine optimization - it's implementing both as complementary approaches. The content that succeeds will be comprehensive enough to earn AI citations, structured clearly enough to appear in featured snippets, authoritative enough to rank in traditional search, and compelling enough that users want to click through.

Structure for Extractability

AI engines don't read content the way humans do. They extract chunks. AI search engines don't rank pages - they extract answers. Structure headings, lists, tables, and definitions so AI systems can confidently cite your content.

Practical formatting guidelines:

• Lead each section with a direct answer.

According to Aja Frost, HubSpot's senior director of global growth, "the first sentence of a page should answer the primary question completely, because answer engines are looking for that quick validation."

• Make each section self-contained.

Every section should stand alone, since AI engines pull individual chunks.

• Include statistics and quotations. The original GEO research from Princeton, Georgia Tech, and the Allen Institute found that

methods such as Statistics Addition and Quotation Addition show strong performance improvements across all metrics, with the best methods improving upon baseline by 41% and 28% on Position-Adjusted Word Count and Subjective Impression respectively.

• Use clear entity references.

Search engines and generative systems rely on entities to understand context. Clearly referencing relevant companies, products, technologies, and concepts helps them interpret how information fits together.

Topical Authority Over Thin Coverage

Both Google and AI search engines reward depth over breadth - but for different reasons. Google uses topical authority as a ranking signal. AI engines use it as a citation signal. You can't predict every query variation an AI will generate, so you need comprehensive content that covers the full breadth of your category. Typically, LLMs pull from 5 to 16 sources per answer, so your presence should be multi-platform.

Build content hubs rather than isolated pages. Each hub should address every question a persona might ask about a topic, from foundational definition to comparison to implementation. This serves both Google's cluster-based topical authority model and the way AI systems evaluate whether a source fits into a broader body of knowledge.

Where to Show Up Beyond Your Website

One finding that catches most SEO practitioners off-guard is how much AI citation activity happens off your domain. Reddit accounts for 22.9% of the top-cited domains across AI models.

LLMs pull heavily from Reddit, YouTube, and Wikipedia.

Earned (third-party, editorial, or affiliate) remains the most frequent type of citation across ChatGPT, Gemini, and Perplexity. This has direct implications for keyword strategy: the keywords and prompts where you lack owned-domain authority might still be winnable through earned presence on third-party platforms. Practical steps:

• Identify which platforms AI engines cite most for your category (run the prompt audit described above)
• Develop a presence on those platforms - Reddit AMAs, YouTube explainers, guest posts on industry publications
• Monitor whether those third-party mentions lead to your brand being included in AI-generated answers

ChatGPT uses Bing's index when it browses the web. If you're not properly indexed in Bing, you're invisible to one of the most widely used AI assistants. Submit your sitemap to Bing Webmaster Tools. This takes thirty minutes and is foundational to AI search visibility.

Measuring What Matters: Beyond Rankings and Traffic

Traditional SEO metrics remain valid but incomplete. Most marketing teams track organic traffic, keyword rankings, and conversion rate well. What tends to be unmeasured is citation rate in AI-generated responses, mention frequency across AI platforms, the ratio of AI-referred visitors to total visitors, and whether branded search volume is growing as a downstream effect of AI visibility.

The Dual-Metric Dashboard

For Google, continue tracking:

• Keyword rankings (with AIO presence flagged)
• Organic click-through rates (segmented by AIO vs. non-AIO SERPs)
• Featured snippet capture rate

For AI search, add:

• Citation rate across ChatGPT, Perplexity, and Google AI Overviews
• Brand mention frequency when prompts relevant to your category are tested
• Referral traffic from AI platforms (check your analytics for chatgpt.com, perplexity.ai referrals)
• Sentiment and positioning - are you cited favorably or just mentioned?

Semrush's 2026 update introduced Semrush One, their AI-era platform that tracks brand visibility across AI assistants like ChatGPT, Perplexity, and Google Gemini.

Ahrefs focuses more on brand recognition, placement, and how AI perceives your brand against competitors through a tool called Brand Radar, available as an add-on at $199/month.

Neither tool is perfect yet. The measurement infrastructure for AI visibility is where keyword research tools were in 2010 - directional, not precise. Supplement automated tracking with manual prompt audits run monthly.

Redefining ROI for Informational Content

Here's a nuance most articles miss: the value of being cited in an AI answer isn't measured in clicks. The challenge is adapting to traffic patterns where even high-volume queries may generate minimal referrals. The opportunity lies in understanding that visibility itself has value - being the cited source in AI responses builds authority, trust, and brand recognition that may manifest in traffic through other channels.

Track branded search volume as a downstream indicator. If AI engines consistently cite your brand when users ask about your category, branded searches should increase over time - even if direct AI referral traffic remains modest.

A Step-by-Step Workflow for Dual-Channel Keyword Research

Putting it all together, here's the process a practitioner can run monthly: Week 1: Traditional keyword research sprint. Pull keywords from Semrush/Ahrefs. Classify by intent. Flag AIO presence. Identify keyword gaps against competitors. Prioritize by a composite score of volume, difficulty, AIO risk, and business value. Week 2: Prompt research sprint. Run 50-100 prompts across ChatGPT, Perplexity, and Gemini for your top content categories. Document brand mentions, competitor visibility, cited sources, and answer quality. Identify gaps where your brand is absent. Week 3: Cross-reference and prioritize. Overlay keyword data and prompt data. Identify topics where both channels show demand. Flag topics where AI visibility is strong but Google rankings are weak (and vice versa). Build content briefs that address both surfaces. Week 4: Content production with dual optimization. Write content that answers specific keyword queries with clear, extractable structure. Include statistics, expert citations, and entity references that increase AI citability. Ensure each section stands alone as a quotable chunk while the whole page serves the full topic journey. This cadence compounds. The brands doing best in AI search are generally those that had strong organic SEO foundations and then extended those foundations to include GEO practices. You're not rebuilding from scratch. You're adding a layer. --- The shift to dual-channel keyword research isn't optional or speculative - it's a direct response to where your audience already spends its time. Traditional search has not decreased; instead, the pie has gotten bigger, with total usage of search combining search engines and search on LLMs increasing by 26% worldwide and by 16% in the US.

GEO tactics overlap heavily with SEO fundamentals. As Lily Ray, VP of SEO strategy and research at Amsive, notes, "the overlap with what we've been doing in the SEO space and digital marketing space before AI search existed is very, very strong." The delta between what you're doing now and what dual-channel research requires is smaller than you think. But that delta is also where the competitive advantage lives. Start with the prompt audit. Run 20 queries on ChatGPT this afternoon. Note who shows up - and who doesn't. That gap between "who's there" and "who should be" is your roadmap for the next six months of keyword and content strategy.