How to Do Keyword Research for AI Search and Traditional SEO

TL;DR

Keyword research in 2026 requires two parallel tracks. The traditional track targets search intent and ranking difficulty to win Google clicks. The AI track targets question-shaped, evidence-rich queries to earn citations in ChatGPT, Perplexity, and Google AI Overviews. This guide walks through both, step by step, with a unified workflow you can run on any site.

By February 2026, ChatGPT had reached 900 million weekly active users (OpenAI, via TechCrunch). At the same time, Google AI Overviews now appear on roughly 25% of all tracked queries, and Ahrefs data shows a 58% drop in organic click-through rate for position-one results when an AI Overview triggers. That stat alone should change how you prioritize keywords. The era of "rank first, get traffic automatically" is over for a significant share of your target queries, particularly informational ones. Yet transactional and navigational keywords still produce strong organic ROI. The right keyword strategy for 2026 identifies which category a keyword falls into, then optimizes accordingly.

Why Traditional Keyword Research Falls Short in 2026

The standard keyword research workflow, pull volume from a tool, check keyword difficulty, map to a page type, has not disappeared. But it misses a parallel channel that now drives substantial discovery. AI answer engines like ChatGPT, Perplexity, and Google AI Mode now serve over a billion users (Google, 2026), and they cite sources in their answers. Being cited is the new version of ranking in the top three.

The problem is that traditional keyword tools measure Google search volume. They tell you nothing about which queries trigger AI citations, what phrasing AI models prefer, or whether a topic has enough evidence-density to make your page citation-worthy. You need a dual research workflow: one that finds rankable keywords for Google, and one that finds citation-worthy topics for AI engines.

Both tracks share a common foundation, understanding user intent, but diverge sharply in execution. The sections below walk through each stage in sequence.

Step 1: Audit What You Already Rank For

Before generating new keyword ideas, mine what you already have. Open Google Search Console and pull the Queries report filtered to the past three months. Sort by impressions descending, then filter for average position between 4 and 20. These are your quick-win keywords: pages Google has indexed and ranked, but not highly enough to get consistent clicks.

For each cluster of related queries, note the intent type: informational (how, what, why), navigational (brand + product), commercial (best, vs, review), or transactional (buy, get, sign up). Informational queries now face the highest zero-click rates, reaching 74% in some studies, because AI Overviews absorb them. Transactional queries remain largely click-through traffic and are worth defending hard.

GSC integration in Guru pulls this data automatically and surfaces which of your existing pages are stuck between positions 4 and 20, so you can prioritize without manual spreadsheet work.

Step 2: Build a Seed Keyword List Across Both Tracks

Start with 10 to 15 seed topics that describe what your business does and what your customers search for. Then expand each seed using two toolsets: one for traditional volume and one for question-intent.

For traditional volume and difficulty:

Semrush Keyword Magic Tool (25+ billion keywords, intent labels, difficulty trends)
Ahrefs Keywords Explorer (SERP feature indicators, traffic potential, parent topic grouping)
Google Keyword Planner (free, useful for confirming volume ranges and CPC as intent signal)

For question-intent and AI-engine alignment:

AlsoAsked (visualizes Google's People Also Ask trees, which reflect real user questions AI engines are trained to answer)
AnswerThePublic (preposition and question variants around a seed)
Semrush's Questions filter inside Keyword Magic Tool
Direct prompting in ChatGPT and Perplexity: type your seed topic and note the exact phrasing used in follow-up questions

One important insight from 2026 query data: the average ChatGPT prompt is 23 words when search mode is off, compared to roughly 3.4 words for a traditional Google search (SimilarWeb clickstream data, 2025). AI-targeted content needs to answer fully-formed questions, not match a three-word head term.

Step 3: Classify Keywords by Intent and Channel Priority

Once you have a seed list, classify each keyword across two axes: intent type and channel priority.

Figure 1: Classify every keyword by ranking difficulty and AI citation risk before assigning it a content treatment.

Keywords landing in the top-right quadrant (low difficulty, high AI risk) are your best dual-channel targets: create content that ranks and earns AI citations simultaneously. Keywords in the bottom-right quadrant are where traditional organic investment has the lowest return, and where GEO-first content formats deliver the most incremental value.

Step 4: Apply the GEO Layer to High-Risk Informational Keywords

For any keyword that is likely to trigger an AI Overview or appear in ChatGPT answers, your content needs to meet a higher evidence bar. The Princeton GEO study (KDD 2024, Princeton and Georgia Tech) showed that adding statistics increased AI citation rates by 41%, adding direct quotations increased them by 28%, and citing sources increased them by up to 115% for lower-ranked pages. These are not marginal improvements: they fundamentally change whether your page gets referenced in an AI answer.

Practically, this means your keyword research output for GEO keywords should include not just the target phrase but a content specification: how many statistics to include, which authoritative sources to cite, and whether a quotation from a named expert strengthens the topic. Running this research for a combined SEO and GEO content strategy takes more upfront work but compounds significantly over time.

The Guru GEO scoring module evaluates these signals at the page level, flagging pages that rank well in Google but score low on citation-worthiness, so you can prioritize GEO upgrades against your existing keyword map.

Step 5: Map Keywords to a Topic Cluster Architecture

Isolated keywords produce isolated pages. Isolated pages produce weak domain authority signals and poor internal-link equity. The alternative is a topic cluster model: one pillar page targeting a broad head term, supported by cluster pages targeting specific long-tail or question-intent variations.

For AI search, topic clusters serve an additional function: they establish your site as a topical authority on a subject. AI models evaluate source credibility in part by assessing how completely a domain covers a topic. A single well-optimized article on a subject does not signal authority the way a cluster of 8 to 12 interlinked, evidence-rich articles does.

When building your cluster map, use keyword grouping by semantic similarity rather than by exact phrase match. Tools like Semrush's Topic Research and Ahrefs' Parent Topic grouping help surface which long-tail variations belong to the same cluster. For question-intent keywords, AlsoAsked's visual PAA trees are particularly useful because they show how Google clusters related questions, which is a reasonable proxy for how AI models cluster related topics.

Read more on building this architecture in Guru's pillar page guide.

Step 6: Prioritize with a Scoring Model

Every keyword list needs a prioritization layer, otherwise you end up writing content in random order based on whoever is loudest in the strategy meeting. Use a simple scoring model with five inputs.

Keyword Prioritization Scorecard

Factor	Weight	What to Measure
Business relevance	30%	Does it map to a product, service, or conversion path?
Traffic potential	20%	Ahrefs traffic potential (not raw volume)
Keyword difficulty	20%	KD score adjusted for your domain authority
AI citation risk	15%	Does an AI Overview trigger? Is zero-click rate >60%?
Content gap	15%	Does a competitor rank here but you do not?

Score each keyword 1 to 5 on each factor, apply the weight, and rank the list. Keywords with high business relevance and moderate AI citation risk should get the most resource investment because they are worth winning on both channels. Keywords with low business relevance but high traffic potential are content for brand awareness and AI visibility, not conversion.

Step 7: Structure Keyword Assignments for Production

The final step is translating your prioritized keyword list into production-ready assignments. Each keyword or cluster needs a content brief that specifies the target URL, the primary keyword, two to four secondary keywords, the intent type, the content format, the target word count, and the GEO requirements (minimum statistics, source citations, structured data type).

Content Brief Minimum Checklist

[ ] Primary keyword confirmed with volume and difficulty data
[ ] Search intent type labeled (informational / commercial / transactional / navigational)
[ ] AI citation risk assessed (AI Overview present or absent for seed query)
[ ] Cluster assignment noted (pillar or supporting page, with pillar URL)
[ ] Target URL slug defined and checked for cannibalization against existing content
[ ] Minimum statistics count set (at least 3 sourced data points for informational content)
[ ] Structured data type specified (Article, FAQPage, HowTo, Product, etc.)
[ ] Internal link opportunities listed (at least 2 to 3 existing pages to link from)
[ ] Competitor pages benchmarked (word count, heading structure, schema usage)

Figure 2: The end-to-end keyword research workflow from GSC audit through GEO layer to a production content brief.

The Guru content module supports brief templates that combine these fields, keeping keyword research connected to what actually gets written and approved, rather than living in a separate spreadsheet that the writing team ignores.

Common Keyword Research Mistakes in 2026

Treating search volume as the only prioritization metric. Volume measures how often a query is typed into Google. It does not measure traffic potential, conversion likelihood, or AI citation frequency. A keyword with 500 monthly searches and high business relevance outperforms a 10,000-volume keyword with low intent alignment.

Ignoring cannibalization before expanding. Adding new pages targeting variations of a keyword you already rank for often splits authority between two URLs instead of strengthening one. Run a site:query or GSC check before creating net-new content.

Optimizing informational content purely for Google clicks. When an AI Overview or featured snippet absorbs a query, position-one organic clicks fall sharply. For informational keywords in that category, the value of the content shifts toward brand impressions and AI citations, not direct traffic. Adjust your success metrics accordingly, which is exactly what the E-E-A-T and trustworthiness signals guide addresses.

Skipping schema markup. FAQ rich results were removed from Google SERPs in May 2026, and HowTo rich results were removed in 2023. Neither schema type produces a rich result in the SERP anymore. However, both are still valid structured data that AI extraction engines read when deciding which pages to cite. Article + FAQPage schema, applied correctly, helps AI models understand content structure without any SERP rich-result expectation attached.

Not revisiting keyword assignments quarterly. AI coverage of query types is expanding rapidly. A keyword that drove strong organic traffic six months ago may now be absorbing into an AI Overview. GSC data shows these shifts clearly. Quarterly reviews let you catch the channel-mix shift before it becomes a traffic drop you have to diagnose retroactively.

Keyword Research for AI Search: A Comparison of Approaches

Dimension	Traditional SEO Keywords	AI-Search Keywords
Query format	2-4 word phrases	Full questions, 7-23 words
Primary metric	Search volume + KD	Topical relevance + evidence density
Intent signal	Google SERP type	AI answer preview + PAA trees
Content treatment	On-page optimization	Statistics, citations, quotable claims
Schema focus	Title + meta + headers	Article, FAQPage, structured evidence
Success metric	Organic clicks, position	AI citations, brand mentions in answers
Cannibalization risk	High (same URL, same query)	Medium (same topic, different platforms)
Refresh cadence	6-12 months	Quarterly (AI coverage shifts faster)

This table does not suggest abandoning traditional keyword research. It illustrates that the two tracks require different inputs, different content specifications, and different success metrics. A mature keyword program runs both tracks in parallel, with shared topic clusters at the foundation.

Frequently Asked Questions

What is the best keyword research tool for AI search in 2026?

No single tool covers both traditional search volume and AI citation likelihood. The most practical combination is Semrush or Ahrefs for volume, difficulty, and intent classification, paired with AlsoAsked for question-format keyword discovery, and direct SERP testing in ChatGPT, Perplexity, and Google AI Mode to see which queries produce AI-generated answers.

How do I know if a keyword is likely to trigger an AI Overview?

Search the keyword in Google and check whether an AI Overview appears above the organic results. Queries of eight or more words trigger AI Overviews approximately seven times more frequently than shorter queries. Informational queries show AI Overview rates above 40% in many tracking studies. Testing each seed keyword manually takes about 30 seconds per query.

Should I stop targeting informational keywords if AI Overviews absorb the clicks?

Not entirely. Informational keywords still build topical authority, drive brand awareness, and earn citations in AI answers, which is a new form of visibility even when there is no direct click. The strategy shifts from optimizing for Google clicks to optimizing for AI citation, which means adding statistics, sourced claims, and evidence-dense content rather than writing around a keyword density target.

How does topic clustering affect AI search visibility?

AI models assess source credibility partly by how completely a domain covers a topic. A cluster of interlinked, evidence-rich articles on a subject signals topical authority more strongly than a single page. Google's own documentation notes that topical depth influences how AI systems evaluate page quality. Building clusters around your core keywords improves both traditional ranking signals and AI citation likelihood.

What query formats work best for AI engine citations?

Full-question formats, direct comparison queries, and requests for specific procedures or statistics perform well in AI answers. Phrasing like "how to," "what is the difference between," "which is better for," and "what does X mean" maps closely to how users prompt AI engines. Using these formats in your headings and FAQ sections makes it easier for AI models to extract your content as an answer.

How often should I redo keyword research in 2026?

Full keyword research should be revisited every six to twelve months, or after a significant Google core update or a major shift in your industry. However, a narrower AI-coverage audit should run quarterly: pull GSC data, check which queries have developed new AI Overview coverage, and update content priorities accordingly. The rate of change in AI search coverage makes annual-only reviews a liability.

What role does structured data play in keyword research?

Structured data is not part of the keyword discovery process, but it is part of the keyword-to-page assignment process. When you assign a keyword to a page, you should specify which schema type applies: Article for editorial content, FAQPage for question clusters, Product for ecommerce, and so on. Pages with correct schema may receive meaningful citation uplift in platforms like Google AI Overviews, though research findings vary by platform. Schema remains a strong signal for content structure that AI extraction engines read, making it a worthwhile lever when assigning keywords to page types.

Can small sites compete for AI citations against high-authority domains?

Yes, more so than in traditional organic search. The Princeton GEO study found that pages with strong evidence density, statistics, sourced claims, and direct quotations received up to 115% citation improvements regardless of domain authority. AI models appear to weight evidence quality more heavily than link-based authority signals, which means a well-sourced, specific, expert-written page on a smaller domain can outperform thin content on a high-DA site.