From Keywords to Entities: How AI Search Is Rewriting Search Optimisation

For years, search optimisation started with a familiar question: what keywords are people searching for? That question still matters, but it no longer explains the entire discovery process.

Google has said its Knowledge Graph contains hundreds of billions of facts about billions of entities, reflecting how modern search systems understand people, places, organisations, and concepts rather than relying only on matching words. AI search takes this further by generating answers from information gathered across multiple sources.

For marketers with an SEO background, this requires a shift in thinking. Keywords remain useful for understanding demand, but repeating the right phrase is not enough to establish what a brand knows, what it offers, or how it relates to a topic.

The move is not really from keywords to no keywords. It is from optimising individual phrases toward building clearer meaning around topics, entities, relationships, and evidence.

Traditional SEO Started With the Query

Classic SEO workflows often begin with keyword research.

A team identifies phrases with relevant search volume, groups them by intent, assigns them to pages, and creates content designed to answer those searches.

That process is still useful. Search queries reveal how people describe their problems and what information they want.

The limitation appears when keywords become the entire strategy.

Consider a B2B software company targeting the phrase “marketing attribution software.” A traditional content plan might focus heavily on variations such as:

  • Marketing attribution software
  • Best marketing attribution software
  • B2B marketing attribution software
  • Marketing attribution tools
  • Attribution software for marketers

Those phrases may represent real search demand. However, creating a separate page for every variation does not necessarily create stronger authority.

Modern search systems can recognise that these phrases are closely related. More importantly, they can evaluate the broader context around the subject.

Does the company actually offer attribution software? What does its product do? Which platforms does it integrate with? Who uses it? Does the website explain attribution accurately? Do credible external sources associate the company with this category?

Those questions move beyond keyword matching.

What Changes When Search Becomes Entity-Aware?

  • An entity is a distinct thing that can be identified and understood independently.
  • A company can be an entity. So can its founder, product, location, software category, customer, partner, or technology.
  • Search systems can use relationships between these entities to understand context.
  • For example:
  • Company A ? offers ? attribution software
  • Company A ? integrates with ? HubSpot
  • Company A ? serves ? B2B SaaS companies
  • Company A ? founded by ? Person B
  • Company A ? reviewed by ? Publication C
  • This creates a much richer picture than counting how often “marketing attribution software” appears on a page.

For AI-generated answers, these relationships can become particularly important. A user may never type the exact phrase a company has optimised for.

They might instead ask:

“What tools can help a SaaS marketing team understand which campaigns create pipeline?”

The system needs to understand the concept behind that request, identify relevant solutions, and construct an answer from available information.

Exact keyword matching alone is not enough to handle that process.

Before: Optimise a Page for a Keyword

  • A traditional SEO workflow might look like this:
  • Keyword ? Page ? Rankings ? Clicks
  • The team identifies a target query, creates or optimises a page, builds authority around it, and monitors ranking performance.
  • That remains a legitimate part of SEO.
  • The problem comes when marketers assume the page only needs enough keyword relevance to compete.

This thinking led to practices such as targeting arbitrary keyword-density percentages, repeating close variants unnecessarily, or creating several nearly identical articles for phrases with the same underlying intent.

Google has long moved away from this type of mechanical matching. Its systems can understand synonyms, concepts, and relationships between words.

  • AI-generated search experiences make the weakness even more obvious.
  • An assistant does not need to reproduce a traditional list of ten blue links. It can synthesise an answer to the underlying question.
  • That changes what content is competing to achieve.

After: Build a Recognisable Topic and Entity Network

  • In an entity-led approach, a single page is part of a larger information structure.
  • Take the same attribution software company.
  • Instead of producing numerous near-duplicate keyword pages, its content ecosystem could establish several connected areas of expertise:
  • Marketing attribution fundamentals
  • First-touch and multi-touch attribution
  • Pipeline attribution
  • CRM attribution
  • Server-side measurement
  • Offline conversion tracking
  • Attribution model limitations
  • HubSpot and Salesforce integrations
  • Paid media measurement
  • Revenue reporting

These topics should not exist simply to create more URLs. Each page should answer a distinct question or support a genuine part of the buyer journey.

Together, they provide context around the company’s relationship with marketing attribution.

The product pages then establish what the company actually offers. Documentation can show how the technology works. Case studies provide customer evidence, while external mentions can strengthen the connection between the company and its category.

This is much closer to how topical authority should be approached: not as publishing everything possible about a subject, but as covering the important parts with enough depth and consistency to demonstrate genuine relevance.

Keyword Density Becomes a Poor Proxy for Relevance

  • Keyword density is appealing because it produces a simple number.
  • If a 1,000-word article contains a target keyword 20 times, the density is 2%. That makes optimisation appear measurable.
  • The problem is that there is no universal “correct” percentage that makes a page relevant.

A page can repeat a phrase frequently while offering very little useful information. Another can answer the query comprehensively while using the exact phrase only a handful of times.

  • For marketers, this means the optimisation question needs to change.
  • Instead of asking:
  • “Have we used the keyword enough?”
  • Ask:
  • Does the page clearly answer the intended question?
  • Are important related concepts explained?
  • Are factual claims supported?
  • Is the relationship between the brand and topic clear?
  • Does the content add information beyond what already exists?
  • Can readers easily identify the key answer?
  • Does the rest of the website reinforce this expertise?
  • Keywords still help communicate relevance. Forced repetition does not create authority.

AI Search Makes Relationships More Important

  • Traditional search results give users a selection of pages to evaluate.
  • AI search can perform part of that evaluation before presenting an answer.
  • Suppose someone asks:
  • “Which CRM platforms are suitable for a 100-person B2B SaaS company with a complex sales process?”
  • To provide a useful response, an AI system may need to understand several relationships:
  • CRM platform ? suitable for ? mid-market businesses
  • CRM platform ? supports ? complex sales workflows
  • CRM platform ? integrates with ? relevant marketing tools
  • CRM platform ? used by ? B2B SaaS teams
  • CRM platform ? has ? specific pricing or implementation requirements
  • The answer depends on more than pages containing “CRM for B2B SaaS.”

This is why factual clarity across product pages, documentation, structured data, third-party profiles, reviews, and industry coverage becomes valuable.

The clearer those relationships are across reliable sources, the easier it is to understand what a company represents.

Structured Data Helps Clarify Meaning, but It Is Not a Shortcut

Schema markup often enters the conversation when marketers begin thinking about entities.

That makes sense. Structured data gives search systems explicit information about what a page represents.

Depending on the website, relevant schema could describe an organisation, product, software application, article, event, person, or other supported entity.

However, schema should be treated as clarification rather than a ranking trick.

Adding Organisation markup does not automatically establish a company as an authority. SoftwareApplication markup does not make a product more credible.

Structured data works best when it accurately reflects information already visible on the page.

It can help systems interpret that information, but the underlying content still needs to be useful and trustworthy.

The same principle applies to AI optimisation more broadly. Technical signals can improve accessibility and clarity. They cannot compensate for weak information.

Third-Party Sources Become Part of the Entity Picture

Brands naturally spend most of their optimisation effort on their own websites.

Entity clarity extends beyond that.

Consider what happens when a SaaS company describes itself as an enterprise platform, while several major review profiles still classify it as a small-business tool. An industry article may use an outdated company name, while another profile lists features that were removed two years ago.

  • Humans can recognise that some information is outdated. Machines also have to reconcile those inconsistencies.
  • A stronger brand footprint includes accurate information across relevant external sources, such as:
  • Industry publications
  • Professional directories
  • Review platforms
  • Partner websites
  • Integration marketplaces
  • Customer case studies
  • Company profiles
  • Executive profiles
  • This does not mean creating hundreds of directory listings.

Relevance and credibility matter more than sheer volume. The goal is to reduce ambiguity around the company and strengthen legitimate relationships with its category, products, people, and partners.

Before and After: What Actually Changes?

  • The difference between traditional keyword-first optimisation and entity-aware optimisation can be summarised more practically.
  • Keyword-first thinking
  • Entity-aware thinking
  • Which keyword should this page target?
  • Which question or topic should this page own?
  • How many times should the keyword appear?
  • Is the topic explained clearly and completely?
  • Can we create another variation of this keyword?
  • Does that variation represent a genuinely different intent?
  • Does this page rank?
  • Is the brand visible across relevant discovery environments?
  • How many backlinks does this URL have?
  • Which credible sources reinforce the brand-topic relationship?
  • Did organic traffic increase?
  • Are search visibility, AI mentions, citations, traffic, and conversions improving?

This is not an argument for abandoning keyword research, rankings, or backlinks. It is an argument for putting them inside a broader model of search visibility.

What Optimising for AI Search Looks Like in Practice

For marketers, optimising for AI search means strengthening the information systems already surrounding the brand rather than chasing a new collection of prompt-specific keywords.

This is also where a broader GEO programme can help. GTMLab.ai approaches AI search optimisation across technical accessibility, entity clarity, content, and authority building. The aim is to make a brand easier for both search engines and AI systems to understand and reference.

Start with the entity itself.

Can someone quickly determine what the company does, who it serves, and how its offering differs from adjacent categories? Check whether that information remains consistent across important pages and external profiles.

Then look at topical coverage.

A brand does not need hundreds of articles to establish expertise. It needs useful coverage of the questions that matter within its category, from early education through product evaluation.

Next, examine the evidence supporting those claims.

Original research, customer results, documentation, expert contributions, and credible third-party coverage provide stronger signals than generic articles that repeat existing information.

Technical foundations still matter too. Important content should be crawlable, indexable, logically structured, and supported by appropriate structured data.

None of these activities is entirely new. What changes is how they work together.

Keywords Still Matter, Just Differently

It would be a mistake to interpret entity-led optimisation as the end of keyword research.

Keywords remain valuable because they reveal the language people use. Search volume can indicate demand, and keyword data can help teams understand how problems are framed.

They also remain important for traditional search results, paid campaigns, content planning, and performance measurement.

What is changing is the role keywords play.

A target phrase should guide the content toward relevant intent. It should not dictate every sentence or encourage teams to create pages that exist only because two keyword variations are slightly different.

Marketers can therefore keep the useful parts of established SEO: technical quality, search demand analysis, useful content, internal linking, and authority building.

What needs to disappear is the assumption that search visibility can be engineered mainly through phrase repetition.

As search becomes more capable of interpreting meaning, clarity about entities and their relationships becomes part of optimisation itself.

The practical shift is subtle but important. Start with what people are trying to understand, establish the entities involved, explain their relationships clearly, and support those explanations with useful evidence.

Keywords still help people find the conversation. They just no longer define the whole conversation.