Meet an Expert

Which Semantic Entity to Consolidate First for AI Citations

How to choose which semantic entities to build first when merging scattered keyword pages into references that AI models can recognise and cite.

Lior Eldan Lior Eldan
Nov 2, 2026 7 min read Strategy

Pick any mid-market SaaS site and count the pages that describe its main product. There is usually a feature page, three blog posts written for keyword variants, a glossary entry and a comparison page, each built in a different year for a different query. A buyer sees one product. A model sees six partial descriptions that disagree on what the product is. Consolidating them into a few semantic entities fixes that, as long as you build the right entity first.

Key Takeaways

  • Build first the entity that buyers already ask AI about and whose signals are currently split across several of your own URLs.
  • An entity is ready to consolidate when you can define it in one sentence that every page, schema block and external profile can repeat word for word.
  • Defer entities that no independent source describes yet, because merging pages on your own site cannot create agreement that does not exist elsewhere.

What a Model Sees When Your Content Is Organised by Keyword

Keyword-era content was built one page per query variant. That made sense when each page competed for its own ranking. A generative engine reads differently. Google’s documentation for AI Overviews and AI Mode says both may use a “query fan-out” technique, running multiple related searches across subtopics and data sources before composing a response. Fan-out hits your variant pages one by one, and each returns a fragment with slightly different wording, a different feature list and sometimes a different category label.

The training layer has the same problem at a slower speed. Trust in a model is a graph of agreement across independent sources. If your own pages disagree about what you sell, the graph breaks at its first node, and no amount of external coverage repairs a definition you have not settled yourself.

Which Entity Should You Build First?

Most teams start with the entity they care about most, usually the flagship product. That is often the wrong call. The better starting point sits where two conditions overlap: buyers already ask AI questions that this entity should answer, and your existing material on it is spread across many URLs that compete with each other.

The first condition comes from auditing AI answers directly. Run the questions your buyers actually ask through ChatGPT, Gemini, Claude and Perplexity, and record two things for each: whether you are named, and whether you are cited as a source. The gap between those two numbers is the gap between Share of Voice and Share of Citation. An entity with high Share of Voice and near-zero Share of Citation is the clearest signal you have. The web talks about the topic, your site holds plenty of material on it, and no single page is clean enough for a model to use as the reference.

The first entity to build is the one where you already have the most to say and the least agreement about how you say it.

The second condition is definitional stability. Before you merge anything, write the entity’s definition in one sentence: what it is, what category it belongs to, who it is for. If product, sales and content cannot agree on that sentence in a single meeting, the entity is not ready, and consolidating it will only produce one long page that still contradicts itself.

Vertical matters too. The GEO: Generative Engine Optimization paper on arXiv found that the strategies it tested for raising visibility in generative engine responses worked differently from one domain to the next. A finance entity and an e-commerce entity will not respond to the same treatment, so judge priority within your vertical.

Start Where Your Equity Is Already Split

Consolidation pays off fastest where you already have links, mentions and history spread across competing URLs. Google’s canonicalization guidance describes the mechanism plainly: choosing a canonical lets search engines consolidate the signals held by individual URLs, such as links to them, into one preferred URL. The same guidance ranks the methods by strength. Redirects are a strong signal, rel=”canonical” annotations are a strong signal, and sitemap inclusion is a weak one. Google also notes the methods stack, so using two or more together increases the chance that your preferred URL wins.

For entity work, this gives you a practical scoring method. List every URL that describes the candidate entity, then count the inbound links and AI citations each one already earns. The entity with the most equity scattered across the most URLs goes first, because merging it moves signal you already have. Redirect the weaker pages into the strongest one, and update internal links to point at the canonical page directly, which Google says helps it understand your preference.

+64%
growth in AI brand presence for NewDay USA, tracked weekly across five engines over six months.See the case study.

Weekly tracking matters here. Consolidation changes what retrieval systems find as soon as the merged pages are recrawled, while training-data presence moves over model release cycles. Measure both layers separately, or a slow training layer will make a working retrieval fix look like a failure.

Ambiguous Names Go First, Unsupported Claims Go Last

Two more factors reorder the list. The first is ambiguity. If your product name is a common word, shares a name with another company, or has been renamed, a model may merge you with something else or split you into two entities. Google’s documentation on Organization structured data says the markup helps it disambiguate an organization in search results, and that some properties, such as iso6523 and naics, are used behind the scenes to tell organizations apart. The sameAs property connects your entity to the external profiles that describe it. An ambiguous entity loses citations every day it stays unresolved, so it moves up the queue even if its search demand is modest.

Keep the markup honest. Google’s guidance for AI features asks site owners to make sure structured data matches the visible text on the page. Schema that claims a cleaner definition than the page itself delivers creates exactly the kind of disagreement you are trying to remove.

The second factor pushes entities down. If no independent publication, directory or review site describes an entity yet, consolidating your own pages will make it tidier without making it trusted. Citations compound, but they need a first independent source to compound from. Park these entities and hand them to PR, so external coverage exists by the time the reference page is ready.

What to Do Before You Merge a Single Page

Run this sequence on your candidate list before any redirects go live.

  1. Pull 30 to 50 buyer questions from sales calls and support tickets, run them across at least four engines, and log mention and citation separately for each.
  2. Map every URL that touches each candidate entity, with its inbound links and current AI citations.
  3. Write the one-sentence definition for each entity and get product and sales to sign off on it.
  4. Rank candidates by citation gap and scattered equity, then move any ambiguous name to the top and any entity without outside corroboration to the bottom.
  5. Build the first entity completely: one canonical page, redirects from the variants, updated internal links, and Organization or Product markup that matches the page text.
  6. Track it weekly for at least a quarter before starting a third entity, so you know which changes moved citations.

Build two entities properly before touching the rest. The second one usually goes faster, because the definition process and redirect map from the first are already in place.

FAQs

How Many Entities Should a Mid-Market Brand Consolidate Into?

There is no fixed number. Most brands need one entity for the organisation, one for each product a buyer would name separately, and a small number for the concepts they want to own. If two candidate entities share the same one-sentence definition, they are one entity.

Should I Delete Old Keyword Pages or Redirect Them?

Redirect them to the canonical entity page. Google’s canonicalization guidance treats redirects as a strong signal and recommends using them when a duplicate page is being retired, which carries the old page’s links into the new reference.

Does Structured Data Alone Make a Model Recognise an Entity?

No. Markup helps search systems disambiguate an entity, but it must match the visible page text, and models still weigh whether independent sources agree with your description. Structured data supports consolidation and cannot replace it.

How Long Before Consolidation Shows Up in AI Answers?

Retrieval-based answers can change once the merged page is crawled and indexed. Training-data presence moves more slowly, across model updates. Track the two layers separately so you can tell which one is responding.

Lior Eldan

About the author

Lior Eldan

Lior Eldan is the Co-Founder of Moburst and serves as its COO. He works at the intersection of marketing, AI and growth, helping brands' teams adapt to AI-driven discovery and decision-making through data-informed strategy and systems thinking.

Meet the whole team

Ready to make your brand the answer?

Our growth strategists take brands from overlooked to recommended. In one call you will see how the AI engines read your brand today, and leave with your own roadmap to dominate the new era of search.