Semantic search

Semantic search is retrieval by meaning rather than by matching the words of the query. Engines model intent, context and entities to judge what a page is about and whether it answers the need.

Google moved this way over time: Hummingbird in 2013, RankBrain in 2015, BERT in 2019 for reading queries in context, MUM in 2021, and the generative systems behind AI Overviews and AI Mode as of 2026.

For you, this means a clear, topic-led page can rank across many phrasings. Repeating one keyword does nothing if the page misses the point.

You no longer write one page per slight wording. You write one strong page per user need and cover the subtopics users expect.

• Queries fan out. A good /pricing guide might earn impressions for “pricing”, “cost”, “how much is yourproduct”, and “pricing tiers”.
• Entity clarity matters. Name products, features, audiences and use cases plainly, with consistent terms across pages.
• Structure matters. Headings, short paragraphs and descriptive anchors help models map sections to intents.

Open the Performance report and filter to a key page. Scan the Queries tab for variety of phrasings and intents, not just the head term.

1. 1

   Check query spread

   Count distinct queries. If /guides/getting-started shows clicks from “setup”, “install”, “configuration”, you are covering the topic well.

2. 2

   Group by intent

   Bucket queries by jobs to be done. For /pricing, group “cost”, “discount”, “nonprofit pricing”, “enterprise plan”.

3. 3

   Spot gaps

   If you see impressions for “annual discount” with low position, add a clear section and an anchor like /pricing#annual-discount.

4. 4

   Look at impressions without clicks

   Rising impressions across many phrasings with flat clicks hints your snippet or on-page answers are weak. Fix title, meta and on-page headings.

• Start with search intent. State the job in the h1, for example “Set up YourProduct on Shopify”.
• Cover the expected subtopics. Use h2s like “Requirements”, “Step-by-step”, “Common errors”, “Pricing implications”.
• Name entities. Mention product names, features, industries and file types users search for.
• Use plain language. Prefer “how much does YourProduct cost” wording in a sentence over jargon.
• Link related pages. From /pricing, link to /guides/billing and /security if buyers ask about them.
• Write answers, not variations. One good answer section beats a list of keyword-stuffed rewrites.

A fixed page reads like a help doc that anticipates questions, not a thesaurus. It earns clicks from many close variants without repeating them.

• Keyword stuffing. Writing “semantic search” ten times will not move you.
• Thin pages per variant. “/pricing” and “/cost” as separate pages can cannibalise each other.
• Ignoring snippet quality. Weak titles and meta descriptions suppress clicks even when meaning matches.
• Overfitting to AI answers. Generative results change. Keep pages useful on their own and citeable.
• Entity drift. Inconsistent names across pages confuse models and users.