Twenty thousand people searched "best running shoes for flat feet" last month. Three years ago, that traffic landed on ten blue links — review sites, brand pages, footwear publications. Last month, sixty-eight percent of those queries were answered inside Google AI Overview, ChatGPT search, or Perplexity, with three to five sources cited. The other thirty-two percent still went to blue links. The brands that show up in the cited answers are eating most of the actual buyer attention. The brands ranking #4 on the blue links are slowly disappearing.
Traditional keyword research is built for the blue-link world. You pick a keyword, pick a SERP position to chase, write a page that wins that position. The volume metric is "monthly searches", the success metric is "ranking #1-3", the optimisation target is "click-through rate from the SERP". None of those metrics map cleanly onto AI search.
AI search keyword research — the discipline that has emerged in 2025-2026 to fix this — answers a different set of questions. Which questions are AI engines actually answering for my topic? Which sources are they citing? What is the gap between what my competitors are saying and what my brand is saying? What content patterns earn citations vs the ones that get ignored? The tools, the workflow, and the success metrics are all different.
This is the honest 2026 version of how AI search keyword research actually works.
What AI search keyword research is, in one paragraph
AI search keyword research is the practice of identifying which questions AI engines are answering for your topic, what they are saying, who they are citing as sources, and where your brand has the opportunity to be cited. The deliverable is a list of question clusters with three columns next to each: search volume, AI Overview presence flag, and brand-citation gap. From that list, you build content that targets the specific passages, claims, and structured data patterns that earn citations.
Why traditional SEO keyword research breaks for AI search
Three structural reasons.
First — the unit of optimisation has changed. Traditional SEO optimises pages for SERPs. AI search optimises passages for retrieval. A 2000-word blog post that ranks #1 on Google for "best running shoes for flat feet" can be invisible inside ChatGPT search if its passages are not cleanly extractable. The H2 structure, the citation density, the named entities — these are now the optimisation surface, not the page-level title tag and meta description.
Second — the metrics that mattered no longer fully apply. Search volume still matters but is no longer the only volume signal — you also need "AI Overview impression count" (currently estimated, not measurable directly), citation frequency across LLMs, and brand mention volume in AI answers. Profound, Athena Intelligence, Otterly, and a handful of newer tools track some of these; none track all of them well yet.
Third — the SERP that traditional keyword tools see is not the SERP your customers see. DataForSEO and Semrush show you the blue-link SERP. Your customer is increasingly seeing an AI Overview at the top, with two or three blue links pushed below the fold. The keyword tools are reporting on a layer of the SERP that is shrinking.
The five things AI search keyword research actually surfaces
A real AI search keyword research workflow gives you five things traditional research does not.
1. The question, not the keyword
AI engines answer questions, not "match a keyword phrase". The unit of optimisation shifts from "running shoes flat feet" to "what are the best running shoes for flat feet, and why does arch type matter?". Question-shaped queries with explicit context get explicit answers; one-word keywords get a generic answer the AI can give without citing anyone.
2. The AI Overview presence flag per question
For every question on your list, you need to know whether Google is currently surfacing an AI Overview. DataForSEO has been tracking this since mid-2025. The presence flag tells you whether traffic on that question is being intercepted by the AI Overview before it reaches blue links. Roughly 60-70% of informational queries have AI Overviews in 2026; commercial-intent queries are at 30-45% and growing.
3. What ChatGPT, Perplexity, and Claude are actually saying
You need to query the AI engines themselves and read their answers. Perplexity is the cleanest because it shows source citations. ChatGPT search shows citations too. Claude with web_search returns citations. Bing Copilot has a citation panel. Each engine has slightly different retrieval patterns — Perplexity favours technical depth, ChatGPT favours brand-name authority, Claude favours recent and specific. Your content needs to fit at least two of those retrieval patterns to earn citations across engines.
4. The citation gap
For every question, list the domains that are currently being cited in AI answers. Compare that list to your domain. The gap is your opportunity. If your competitor is cited eight times across the four major engines for "what are the best running shoes for flat feet" and you are cited zero times, that is the gap to close. The closing move is content — specifically, content that ships citation-worthy data your competitor does not have.
5. The content brief that will actually earn citations
AI search keyword research outputs content briefs that are structured for retrieval, not just for ranking. The brief specifies: H2-level questions the page will answer, named entities to include, statistics or data points to cite, schema markup type (FAQPage, Article, HowTo), passage-level word count targets, and citation-worthy quotes from named experts. Generic "write 2000 words about X" briefs do not earn AI citations; specifically-structured briefs do.
The toolkit you actually need in 2026
No single tool covers all five surfaces above. The 2026 stack is six to eight tools combined.
DataForSEO covers the SERP layer — Google AI Overview presence, search volume, KD, autosuggest, "people also ask". Profound and Athena Intelligence cover brand mention frequency across LLMs. Otterly tracks AI Overview specifically. AlsoAsked covers deep PAA trees. Reddit and Stack Exchange APIs surface the actual questions in real-world communities. Tavily, or Claude with web_search, finds fresh sources that Google has not yet indexed.
At the synthesis layer, you need an LLM (Claude or GPT-4) to cluster questions, generate content briefs, and analyse citation gaps. The free tier of the AI search keyword research tool I have built — at /tools/ai-search-keyword-research/ — combines DataForSEO + Reddit + Perplexity + Claude into one workflow. It is not the only way to do this, but it is currently the only free way to get all five outputs in one place.
The workflow, in one paragraph
Start with a seed topic. Pull every "people also ask" plus autosuggest from DataForSEO. Pull every top-30 Reddit and Stack Exchange thread on the topic. Cluster the combined question list semantically with Claude (or use the tool above which does this for you). For each cluster, fetch the actual answers from ChatGPT, Perplexity, and Claude (cited sources included). For each cluster, read the cited domains and compare to your own. Score each cluster by search volume, KD, AI Overview presence, and brand citation gap. The top 10-15 clusters by combined score become your content roadmap.
What changes about the content you write
Three things, on every page.
Passage extractability. Each H2 should answer one specific question. Each H3 within that should answer a sub-question. AI engines extract passages, not pages — your structure has to make the passages findable.
Citation density. Real numbers, named experts, dates, and specific entities. AI engines preferentially cite sources that make verifiable claims. A page with three citable statistics beats a page with no statistics, regardless of word count.
Schema markup. FAQPage, Article, HowTo, Service, Person, Organization. The explicit signal layer that tells AI engines what your page is and who you are. Most agency-built content sites do not ship schema; the few that do compound their citations meaningfully.
What does not change
Three things stay the same. First, search intent matters as much as ever — the page that wins is the one that matches what the searcher actually wanted. Second, content quality compounds — thin or AI-slop content does not earn citations, never will. Third, brand authority matters more, not less — AI engines preferentially cite established sources over new domains, so the long-game investment in domain authority is not optional.
When AI search keyword research matters most
B2B SaaS, professional services, and content publishers in informational-query categories where AI engines are intercepting traffic. Marketing agencies whose clients are asking "why is my organic traffic down" — the answer is often AI Overview interception, and AI search keyword research is the diagnostic tool that proves it. Local-search-driven businesses (dentists, plumbers, restaurants) get less benefit because "near me" queries are still routed to Maps. Pure ecommerce gets less benefit because product queries are routed to Amazon and Google Shopping.
When traditional keyword research is still enough
Local-search-only businesses. Pure paid-search-driven traffic. Sites with under DR 15 in non-AI-Overview categories. The compounding payoff of AI search keyword research is roughly proportional to the percentage of your topic queries that have AI Overviews — if that number is under 30%, traditional keyword research is still your primary discipline and AI search optimisation is a layered addition.
Frequently asked questions
What is the difference between AI search keyword research and AEO/GEO?
AEO (Answer Engine Optimisation) and GEO (Generative Engine Optimisation) are the strategy-layer terms. AI search keyword research is the research-and-discovery layer that informs them. AEO/GEO is "what we are trying to achieve"; AI search keyword research is "the input data for that strategy".
How long does it take to start ranking in AI Overviews?
Faster than traditional SEO, often dramatically so. AI engines update their retrieval indices on a rolling basis — measurable changes in citation frequency typically appear in 30-90 days for content that ships with strong structured-data signals and clear passage segmentation. By contrast, traditional Google ranking improvements often take 4-8 months on competitive terms.
Do I need to do AI search keyword research separately, or is it part of regular SEO research now?
In 2026, it is a parallel discipline that overlaps about 60% with traditional SEO research at the foundations and diverges 40% on the synthesis and execution layer. You still need traditional research for the blue-link layer of the SERP. You also need AI-specific research for the AI Overview / AI engine answer layer. The right shape is one combined workflow that captures both — which is what the tool linked above is built for.
What tools do I need to do this myself?
At minimum: DataForSEO for SERP data ($0.10/search), an LLM API for clustering and brief generation (Claude or GPT, $0.02/search), Perplexity API for AI engine benchmarking ($0.005/query), and free Reddit + Stack Exchange APIs for community questions. Total cost: roughly $0.20 per uncached topic query. Or use the free tool I have built that combines all of these.
Is AI search keyword research the same as content brief generation?
Closely related but not the same. Keyword research surfaces what to write about; brief generation specifies how to write it. The two outputs feed each other — keyword research produces the prioritised topic list, brief generation produces the structured outline that will earn citations on each topic.
Where to go next
If you want to run this analysis on your own topic right now, the free tool at /tools/ai-search-keyword-research/ produces all five outputs (questions, AI Overview flags, AI engine answers, citation gaps, content briefs) in about 30 seconds per topic.
If you want the strategic version — what AI search optimisation looks like as an ongoing program rather than a one-off audit — the deeper write-up is at /solutions/ai-seo-geo/.
If you want the service brief — what working with me on AI search optimisation actually involves and costs — that is at /ai-search-optimization/.
And if you want to talk it through directly, book a 30-min call.