- Topic research is the decision before you write: it answers whether the audience cares, whether anyone is searching, which problems sit inside the topic, and whether you have a unique point of view.
- Keyword volume is still the starting filter. Treat it as an order of magnitude, then use AI to collect real questions from communities, synthetic searches, and customer transcripts.
- Capture that work in a topic brief (search data, a library inventory, a search engine results page (SERP) read, and social research). Some briefs should not become posts, because the data shows weak demand or a poor fit.
- Generative engine optimization (GEO) and answer engine optimization (AEO) share the same topic list as search engine optimization (SEO). Research the questions models will quote from those topics, instead of building a separate GEO calendar.
- Someone on the team should still review recommendations: relevance, uniqueness, new versus refresh, and whether the question is yours to answer.
Topic research used to mean exporting keywords and checking them by hand. That check still matters, but the work around it is larger now, because the same brief has to serve Google results and AI answers.
Buyers now ask ChatGPT, Perplexity, and Gemini, then wander through Reddit and review sites, often before they hit your site. Rankings still matter, and so does being the source an answer engine can extract and cite.
Do not ask a model to invent the editorial calendar. Use AI to gather demand signals and real questions faster, so you can decide what to write, refresh, or skip for search engines and for answer engines.
Topic research is still important
Use topic research to pick promising topics before you spend production time. The four questions haven't changed:
- Does the audience care about this topic?
- Is anyone searching for it?
- Which specific problems sit inside it?
- Can we create unique content with a unique point of view?
If anyone can generate a page with AI, that last question matters more than it used to. Publish when you have a point of view the current search results and AI answers do not already cover.
Topics and keywords are different scopes. A topic is broad enough for many posts, and specific enough that your buyers live in it; keywords are the phrases inside that subject, and strategy needs both. A brief built around a topic (rather than a single keyword) shows related queries and a fuller journey: understand, a long-tail problem, then evaluate or pricing, and sometimes a solutions page on the same theme.
A brief is for alignment first. Ours usually includes long-tail keyword data (volume and competition), an inventory of related pages, a SERP read for intent, and a scan of how people talk on social platforms; from there, it can recommend new titles, refreshes, a campaign, or nothing.
A no-go is useful. Some topics are dear to the brand and still dead in search and community, and finding that out in a brief costs less time than finding it out after you publish.
What AI is good at in research
Keyword volume is still useful for introductory topic selection. Volumes from keyword tools are estimates, so treat them as orders of magnitude (10 searches a month versus 250) rather than exact counts. If a topic you love shows no volume, that is evidence that nobody is looking, and it may not be worth the time.
Volume is being amended. Once a topic looks viable, the extra data (community language, synthetic queries, transcripts) shows how to cover it.
Directional demand
Google Trends is a fast viability check. It won't give absolute volume, but it will show whether interest is moving, plus related and rising queries. In 2026, a Gemini-powered panel can suggest up to eight related terms to compare, which is a quick look at the related questions around a topic: budget, implementation, homegrown versus buy, and similar follow-ups.
A topic explorer workflow adds SEO data. You type a theme and get keywords, plus the page on your site that already ranks highest for each one. Ranking bands help: the top three often need no urgent work; positions 5–10 have room; not ranking is not an automatic brief for a new URL.
Write new when your strongest page sits below the top 10, when a refresh would be a rewrite, or when the relevant pages already win other queries you don't want to disturb. Updates are often a few hours; new posts take longer to earn traffic. Use those ranking and library facts to choose between a new URL and a refresh.
Real questions, at scale
Keyword volume will not show the language buyers use when they are stuck; community threads will. For years, marketers pulled “People also ask” boxes and hunted Reddit by hand. AI helps because it can search many threads, on many sites, against a short topic list.
The FAQ workflow we use is straightforward. You feed 5–10 topics that are central to messaging, product, and audience. A workflow (we build ours in n8n) searches Reddit, LinkedIn, Google, and similar places, then returns questions; a human deletes the odd and off-audience ones, and a second pass can draft answers. Sales-call friction, support tickets, and intake surveys fit the same pipe.
Synthetic queries extend that same research. When someone asks a large language model (LLM) for customer relationship management (CRM) options for a small business, the model may run extra searches on pricing, reviews, and implementation cost; those follow-ons show what a complete answer needs. Keep customer phrasing when you turn the questions into FAQs or posts, so that the next time someone types that into ChatGPT, you have a page that is easy to cite.
You still review the list. Automation will mix useful questions with junk: the workflow covers more threads than a person can, and a person still has to delete questions that are off-audience or poorly phrased.
How the same research serves SEO and GEO
GEO is sometimes called LLM search. AEO is the practice of structuring content so LLMs can extract, understand, and cite it. Traditional SEO was about ranking; AEO makes your pages easy for a model to extract and cite when it answers. It's a complementary subsystem. Search engines still ground a lot of what models say, so the research and authority you built for SEO still help.
You don't need a second topic map. AEO starts with topic discipline: 3–5 strategic topics tied to product, audience, and position. FAQ Finder style workflows often use about 5–10. Limit the list to topics your product, audience, and staffing can support.
Once a topic is researched, reuse that package in more than one place: a post, FAQs that match assistant queries, a Reddit reply, a LinkedIn note, and a line in a sales one-pager. Models favor information that shows up across reputable sources. Lead with the answer, keep paragraphs able to stand alone, and write FAQs in the language people type into assistants. Pages with FAQ structured data are cited more often by LLMs; that traffic is still smaller than classic search, and it often converts better.
GEO research questions are strategy questions:
- What are customers talking about on each platform?
- Is the brand in those discussions?
- Does existing content already answer the question?
- Are ChatGPT or Claude citing you when it comes up?
Include freshness in the brief. LLMs prefer current pages, and they will also cite an archive. If old “data mesh” posts still outnumber what you say about the current category, models may describe you as a data mesh company; run topic research so you notice that mismatch before buyers do.
Top-funnel encyclopedia pages lost a lot of clicks as AI Overviews rose, and “What is…” traffic often needs no click. LLM sessions are growing while some unbranded Google clicks fall, yet pipeline from organic has been more stable than those traffic charts. Put the time you would have spent on extra glossary pages into better questions instead.
What you should not hand to the model
AI can collect questions and demand signals at a scale a person cannot, but it should not make the publish, skip, or refresh call by itself.
After the brief, recommendations still go through the team: audience need, brand and product fit, a unique angle, and whether you have the people to write it. You leave with titles to create and pages to revise, each tied to specific queries, or with topics you will skip.
Production is a separate decision from research. Intro-level explainers can be LLM-assisted; opinion, original research, and proof still need humans. Keep the same split in research: let the workflow gather sources, and let a person decide what to keep. Keyword data points to the topic; community questions, synthetic queries, and transcripts show the coverage.
Conclusion
AI helps topic research for SEO and GEO when it widens the evidence behind the brief: directional demand, real questions, and a clear new-versus-refresh call. It doesn't replace the four questions, the uniqueness check, or the team review. Treat research as a system (same topics, many channels), and keep people on the decisions that still need judgment.
Frequently asked questions
Is topic research different for GEO than for SEO?
The topic list is shared. GEO and AEO change what you extract from the research: questions that match how people prompt assistants, modular answers, FAQs, and whether models already mention you. You still start with audience, demand, problems inside the topic, and a unique point of view.
Can I skip keyword volume now that models collect questions?
No. Volume is still useful for introductory topic selection. Treat it as an order of magnitude, then amend it with community questions, synthetic LLM queries, transcripts, and Trends. If a topic has no volume, that is often evidence that nobody is looking.
How many topics should I feed a research workflow?
Limit the list to topics your product, audience, and staffing can support. AEO guidance is 3–5 strategic topics; FAQ Finder style workflows often use 5–10 that are central to messaging, product, and audience. Feeding every possible theme into the scraper returns a lot of off-strategy questions.
Should I write a new page or update one I already have?
Create new content when your strongest page sits below the top 10, when nothing is close enough that a refresh wouldn't be a rewrite, or when the relevant pages already win other queries you don't want to disturb. If a page is already in the top three for a better query, leave it.
Do FAQs help with LLM citations?
Pages with FAQ structured data are cited more often by LLMs. The traffic is still smaller than classic search, and it often converts better. The FAQs need to be real buyer questions, not filler; human review of the research list keeps the weak questions out.
Where does Google Trends fit next to a keyword tool?
Trends is a fast, Google-sourced viability check. It won't give absolute volume. Use it to see trajectory, related terms, and rising queries, then spend slower keyword and community research on the topics that survive.
Related reading
- Get started with SEO topic research: The brief, the four viability questions, and how to turn research into assignments.
- What is replacing keyword volume?: Why volume stays in the stack, and which granular sources should sit next to it.
- Getting started with AEO: How answer engine optimization sits beside SEO, including topic discipline and extraction.
