Are Keywords Still Important for AI Search, or Should Marketers Focus on Entities and Semantics?

Where do keywords, entities and questions each fit?

AI answers are built in stages, and each stage depends on something different:

Keywords dominate the early stages; entities and answer quality dominate the later ones. A page that uses none of the words buyers search with may never be retrieved. A page stuffed with keywords but vague about what it offers may be retrieved and then ignored.

Does clear content matter more than keyword optimization for AI visibility?

Beyond a basic level, yes. Once a page uses the terms people search with, extra keyword repetition adds nothing, while clarity keeps adding value. The original GEO research found that keyword stuffing did not improve visibility in generative answers. Adding statistics, quotations and credible citations did. Google's spam policies treat keyword stuffing as a violation, and they now explicitly apply to generative AI responses too.

Clear content means:

• stating what the page covers in plain words near the top;
• naming the product, place or service rather than "it" or "our solution";
• answering the question directly before elaborating;
• using the terms buyers use, including the common alternatives, once each where they fit naturally.

Does writing content around questions work better than targeting keywords for AI search?

Usually, yes, because people ask AI engines questions, often long ones. A keyword such as "CRM for estate agents" hides many questions: which CRMs integrate with property portals, what they cost per user, whether they handle lettings as well as sales, how long migration takes. A page built around the keyword tends to cover the topic generally. A page built around those questions answers each one directly, giving the engine several passages that can match the narrower searches it runs.

Questions and keywords are not opposites. The keyword identifies the topic and its demand; the questions define the content. How to create pages around specific questions is covered in how to create content that answers specific questions people ask AI.

What is a buyer question in AI search, and how is it different from a keyword?

A buyer question is the full question a potential customer asks when choosing, in their own words and with their own constraints. A keyword is the short phrase that represents a topic in search data.

Buyer questions carry the constraints that decide recommendations: size, location, budget, integration, situation. AI engines use those constraints to filter brands. So they are the right unit for both content planning and AI visibility tracking. How to collect them is covered in how to find the questions people ask ChatGPT about your business.

Can AnswerThePublic questions represent buyers better than keyword data for AI search?

They help, but they do not represent buyers on their own. Tools like AnswerThePublic generate questions from search engine autocomplete, so they show what people commonly type into search boxes. That is useful for ideas and vocabulary. But autocomplete favours short, common phrasings. It rarely captures the detailed, constraint-heavy questions people ask assistants, or questions specific to your market. The best buyer question sets combine:

• sales call notes and objections;
• support tickets and onboarding questions;
• review text, which shows what customers care about;
• on-site search queries;
• "People also ask" and autocomplete tools such as AnswerThePublic;
• Search Console queries of eight words or more.

Does search intent research still matter when optimizing content for AI search?

Yes. Search intent, meaning whether someone wants to learn, compare, buy or find a specific site, still decides what a good answer looks like. AI engines respond differently to each. Informational questions get explanations, comparison questions get lists and trade-offs, and transactional questions get products and providers. Knowing the intent tells you which page type and format to create. What changes is the level of detail: AI questions express intent more explicitly and with more context than short keywords do.

What is conversational intent in AI search, and how is it different from search intent?

Conversational intent is what a person is trying to achieve across a whole conversation with an assistant, not just in one query. It differs from classic search intent in three ways:

• It builds over turns. "What should I look for in a CRM?" leads to "Which of those work for estate agents?" and then "Is Acme cheaper than Brightside?"
• It carries context. Later questions depend on earlier ones and on what the person has said about themselves.
• It moves towards decisions quickly. A conversation can go from learning to shortlisting in a few turns.

For content, that means covering the path, not just single queries: explanations, then criteria, then comparisons, then specifics like pricing and setup, with clear links between them. A site that answers only the first step leaves later turns to competitors. Tracking related questions as groups rather than single keywords is covered in whether GEO strategy should track intent clusters instead of individual keywords.

Should marketers focus on entities and semantics?

Yes, as a complement to keywords. Entity clarity means engines can tell exactly which business, product, place or person your content is about, and connect it to what other sources say. In practice:

• use one consistent name for each product and the business;
• state the category, location and audience plainly;
• connect related pages with descriptive internal links;
• use Organization, Product and LocalBusiness markup that matches visible content;
• keep profiles and listings consistent with the site.

How entities help engines understand a site is covered in how entities and semantic structure help AI understand a website.

What do people ask about keywords and AI search?

Is keyword research dead?

No. It still shows demand, vocabulary and topics. Use it to find what to cover, then plan content around the questions within each topic.

Should I optimise for long-tail keywords for AI?

Cover long, specific questions within strong pages rather than creating a thin page for each variation.

Do exact-match keywords matter in AI answers?

Less than they used to. Engines match meaning as well as words, but using the terms buyers use still helps pages get retrieved.

Can I estimate AI prompt demand from keyword volume?

Roughly, as a proxy for topic interest. See how to use keyword research to estimate AI prompt demand.