← PYC · guides
Guide / topical-authority

Topical authority, completely.

What it actually is, why it beats keyword chasing and holds through algorithm updates, and how to build it — the entity-first, coverage-plus-history method this studio runs, with real field data behind every claim.

By Phil Yarrow · 14 min read · updated 2026-07-15 · Download PDF ↓

Contents
  1. In one minute
  2. What it actually is
  3. Why it matters now
  4. The anatomy
  5. A worked example
  6. How to build it
  7. How to measure it
  8. Test it yourself
  9. Where it goes wrong
  10. Authority in the AI era
  11. Questions
  12. Further reading
  13. The proof

In one minute

The whole argument, up front
  1. 01 Topical authority is a search engine’s confidence that you cover a subject completely enough to be trusted on it. It is built, not bought — from coverage plus historical data, and it compounds.
  2. 02 The unit of work is the subject, not the keyword. You map an entire topic — its entities, attributes and the real questions — and cover it as one connected network of pages, linked by meaning.
  3. 03 It is slow first, then exponential. Coverage and history accrue for months with little to show, then rankings hold through updates and keep climbing. Our field study puts the foundation phase at roughly a year.
  4. 04 It is the durable answer to the AI era. The same entity-complete coverage that a search engine trusts is what a language model quotes — so authority is how you get ranked and cited, not just shown.
  5. 05 It cannot be faked with volume. A hundred thin pages is not coverage; it is noise. Depth in the centre beats scatter at the edge, every time.

What topical authority actually is

Topical authority is the most misread idea in search, because the name sounds like a metric. It is not one. It is a verdict — the conclusion a search engine reaches once it has enough evidence that you cover a subject completely, and have done so long enough to be trusted on it. You cannot set that number. You earn the verdict, and it compounds.

Earning it means changing the unit of work. The keyword era treated a website as a bag of unrelated pages, each chasing its own phrase. Authority treats the website as one thing that is about something. The work is no longer the keyword; it is the coverage of a subject — its entities, the questions people actually ask, and the way they connect. Keywords become an input to that, not the point of it.

This works because search engines stopped reading pages as bags of words a long time ago. They read them as statements about entities — real things with identities and relationships. “Misted double-glazing repair” is not a phrase to match; it is a service, tied to a set of symptoms, performed in a set of places, on a set of products. Topical authority is the engine’s judgement of how completely you cover that space — and how long you have credibly done it.

It helps to say what it is not. It is not “domain authority” or any third-party score — those estimate a whole site’s link strength and live in a toolbar; topical authority is subject-specific and lives in the search engine. It is not backlinks, though links still help. And it is not E-E-A-T, though it is how you demonstrate the expertise E-E-A-T asks for. It is narrower and more useful than all three: the engine’s confidence that, on this subject, you are a complete and credible source.

Why it matters now more than ever

Two shifts turned authority from nice-to-have into the whole game. Algorithm updates now reward genuine, comprehensive coverage and reset the thin tactics that used to work — so authority is the ranking that holds instead of vanishing with the next update. And AI answers changed what a ranking is even worth: the same entity-complete coverage a search engine trusts is precisely what a language model reaches for when it decides whom to quote.

We measured both. Across a 16-month study of five live businesses, coverage-built rankings held and climbed straight through the 2025–2026 core updates. One account showed the “great decoupling” — impressions up, clicks compressed — while the account whose position climbed the most (page five to page two) actually gained click-through as the era compressed everyone else’s. Authority is how you stay ranked and get cited; without it you are merely shown.

The anatomy of authority

Strip authority down and it is two ingredients: coverage — how completely you map and cover a subject — and historical data — how long you have credibly done it. Every idea below serves one of those two. Six terms, defined so nothing rides on jargon:

Topical authority
The trust a search engine assigns your site on a subject once your coverage of it is complete enough, and old enough, that ranking you is the safe bet. A conclusion the engine reaches, not a score you set.
Source context
What your whole site is fundamentally about — the lens every page is read through. A garden centre and a chemist can both publish “how to treat aphids”; the source context tells the engine which one to trust for what.
Central entity
The one thing your site is most about — the hub the whole topical map hangs from. Everything you publish either is the central entity, an attribute of it, or a bridge to it.
Topical map
The structured plan of every sub-topic worth covering, split into a core section (close to what you sell) and an outer section (supporting topics that build authority). Your table of contents for the subject, written before a word of content.
Contextual coverage
Covering each sub-topic’s attributes and questions completely — not one page that mentions a thing, but the full set of statements a genuine expert would make about it.
Historical data
The track record a domain accrues as its coverage is crawled, indexed and clicked over time. The part of authority you cannot shortcut: eighteen months of covering a subject is a different, more trusted signal than the same pages published last week.
Cost of retrieval
How expensive it is for a search engine to crawl, render and understand your site. Clean architecture, fast pages and unambiguous structure lower that cost — so more of your coverage gets crawled, indexed and credited. Authority you build but the engine cannot cheaply read does not count.
The topical map · core and outer
Central entity
Core section
  • What you sell
  • Buyer-intent terms
  • Comparisons
Attributes
  • How it works
  • Specs & options
  • Use cases
Outer section
  • The questions asked
  • Problems solved
  • Adjacent topics
The core section is close to what you sell and converts; the outer section is the supporting coverage that earns the authority the core needs to rank. Most sites build only the core — which is exactly why the core never ranks.
Worked example

The map, made concrete

Abstract maps are easy to nod at, so here is a real one — the subject behind the glazing business in our field study, sanitised. The central entity is the service; the core section is what converts; the outer section is the coverage that earns the authority the core needs to rank.

Double-glazing repair
Core section · converts
  • Misted / failed unit repair
  • Broken pane replacement
  • Emergency glass repair · [town]
  • Sealed-unit replacement
  • Conservatory glass
Outer section · earns authority
  • Why double glazing fails
  • Toughened vs laminated glass
  • U-values & energy ratings
  • How to spot a blown unit
  • Repair vs replace: the cost
The questions that anchor the outer section
“condensation between the window panes”“can misted double glazing be fixed”“how long does double glazing last”

Build only the core — those five repair services — and you have handed the engine no reason to trust you. So it does not rank you, even for the things you sell. Add the outer section, cover the questions completely, link them by meaning, and the whole map lifts at once. That is the account that went from page five to page two, then broke upward.

How to build it, in order

The sequence matters more than any single step. Get the first two right and the rest follow; get them wrong and every later step inherits the error.

  1. 01

    Name the source context and central entity

    Before any content, decide what the site is fundamentally about and the one entity everything hangs from. This single decision governs which topics belong to you and which dilute you. Get it wrong and every later step inherits the error.

  2. 02

    Draw the topical map — core and outer

    List every sub-topic, entity, attribute and real question in the subject. Split them: the core section is close to what you sell and converts; the outer section is the supporting coverage that earns authority. Skip the outer section and you have pages, not authority.

  3. 03

    Map queries to documents

    Each node becomes one page with one clear job — one query intent answered completely, with the entity’s attributes covered, not sprinkled. One page, one statement; the network makes the argument, not any single URL.

  4. 04

    Link by meaning

    Internal links should describe relationships — this service, for this problem, in this place — not just pass equity around. That linked structure is the semantic network a search engine reads to judge how completely you cover the subject, and it is what a language model traverses to quote you.

  5. 05

    Produce at a cadence

    Coverage only becomes authority if you keep shipping — and a complete topical map is more than a human content team can sustain by hand. Automating the production pipeline is how the network gets built and maintained faster than the competition can write, which is where the efficiency comes from.

  6. 06

    Let history accrue — and do not stop

    Publish consistently and hold the line. Authority is history plus coverage; history only accumulates while you keep covering the subject. The sites that win are the ones still shipping in month eighteen, not the ones that stopped at month three because the curve was still flat.

How to measure it

Authority is slow, so the wrong metric will get a good programme cancelled. Watch these, in order of how well they track the outcome:

  1. Average position — the leading indicator. A search engine ranks you better before anyone clicks. In our study, position improved in every account where it was measured, even the ones whose clicks fell. It is the earliest honest sign the work is landing.
  2. Non-brand share. The proportion of clicks that are category demand rather than people typing your name. That is the traffic authority captures — and the traffic that grows a business.
  3. Coverage. How much of the topical map is live and indexed. The input you actually control, and the one that predicts the rest.
  4. Clicks and CTR — last, and with care. In the AI era clicks decouple from value; judge them against position and conversions, not on their own.

The field study shows each of these with real Search Console and GA4 data — including the accounts where the click line fell while the value rose.

Evidence · methods

Test it yourself

You do not have to take any of this on faith. Coverage, relevance and novelty — the ingredients of topical authority — are measurable, with the same information-retrieval methods search runs on and open-source code you can run in minutes. Three checks:

1 · Does a page actually answer the query?

Embed the query and the page, compare them with cosine similarity. A high score means the page genuinely covers the intent, not just mentions the words — the Sentence-BERT method. 

# pip install sentence-transformers
from sentence_transformers import SentenceTransformer, util
model = SentenceTransformer("all-MiniLM-L6-v2")

query = "how to fix misted double glazing"
page  = open("your-page.txt").read()

score = util.cos_sim(model.encode(query), model.encode(page))
print(f"relevance: {float(score):.3f}")   # 0..1 — higher = the page answers it

2 · What does the topic cover that you don’t?

Extract the entities from the pages already ranking and subtract yours. What is left is your coverage debt — the outer section you are missing. Uses spaCy entity recognition. 

# pip install spacy && python -m spacy download en_core_web_sm
import spacy
nlp = spacy.load("en_core_web_sm")
ents = lambda t: {e.text.lower() for e in nlp(t).ents}

yours   = ents(open("your-page.txt").read())
ranking = ents(open("top-competitor.txt").read())

print("coverage gap:", sorted(ranking - yours))   # entities they cover, you don't

3 · Does your page add anything new?

Google’s Information Gain patent rewards pages that add information beyond what already ranks. Measure it as the entities and claims your page has that the top results do not — the inverse of the gap above. Novelty, not word count.

What this proves — and what it doesn’t

These measure the concepts topical authority is built on — semantic relevance, coverage and novelty — using the same primitives (embeddings, entity extraction, TF-IDF) that modern search runs on. They do not reverse-engineer Google’s ranking, and nothing here claims to. They give you an honest, reproducible read on whether your coverage is complete and your pages genuinely answer the intent — which is the part you actually control.

Where it goes wrong

The six failures that stop authority from ever forming:

  1. Keyword-first thinking Treating the site as a bag of unrelated pages, each chasing its own phrase. It ignores the thing that actually earns rankings now: how completely the site covers a subject.
  2. Breadth without depth A hundred thin pages scattered across a topic is not coverage — it is noise, and it can actively signal a low-quality source. Cover the centre completely before reaching for the edges.
  3. Skipping the outer section Publishing only the commercial pages that convert. Without the supporting coverage that proves expertise, the engine has no reason to conclude you are authoritative — so the commercial pages never rank.
  4. Orphan pages and lazy linking Pages with no meaningful internal links are pages outside the network. If the structure does not describe how topics relate, neither the search engine nor a language model can read the coverage as a whole.
  5. Stopping before the curve turns Cancelling in month four because the line is flat. The flat part is the work accruing, not failing — our data shows roughly a year of foundation before the breakout. Impatience is the most expensive mistake.
  6. Chasing clicks over authority in the AI era Optimising for a click count that AI Overviews are compressing anyway. The durable play is authority — rank high enough to keep the click, and become the source the AI cites.

Authority in the AI era

AI answers resolve more queries without a click, so raw traffic is a shakier target than it used to be. That does not weaken the case for authority. It sharpens it. The entity-complete, densely-linked coverage that makes a search engine trust you is the same structure a language model reads to decide whom to quote. Authority is how you get cited, not just ranked.

And the click has not vanished for the authoritative. In our study, the account whose position climbed the most saw its click-through rate rise even as the era compressed everyone else’s — earning authority beat the decoupling. Rank high enough and you keep the click; cover completely enough and you become the source the AI names. Both are the same job — the data is here, and how you actually get cited is its own guide.

Questions

How long until topical authority works?
Months, not weeks. Coverage and historical data have to accrue before a search engine will trust you on a subject. In our field study the clearest account sat roughly flat for a year before the curve broke upward — so budget for a foundation phase and judge the programme on the shape, not month one.
Is this just publishing more content?
No. Volume without structure is noise. Topical authority is complete, connected coverage of a mapped subject — the right entities and questions, each answered properly, linked by meaning. A hundred thin posts can hurt; thirty complete, connected pages can win.
Does topical authority survive Google core updates?
It tends to — because core updates are usually trying to reward exactly what it is: genuine, comprehensive coverage. Rankings built on thin tactics get reset; rankings built on authority hold and often gain. Our field reports show accounts holding and climbing across the 2025–2026 updates.
How is this different from keyword research?
Keyword research finds phrases to target. Topical mapping defines a subject to own — its entities, attributes, questions and how they relate — then covers it as one network. Keywords are an input; the map, the coverage and the links are the work.
Do I still need backlinks?
Links still help, but they are no longer the whole game. A search engine increasingly judges a site by how completely and credibly it covers a subject. Comprehensive coverage plus a coherent entity identity can rank without a heavy link profile — and earns links on its own once it is the best resource on the topic.
Can this be automated without becoming a content mill?
Yes — the automation is on the production pipeline, not the thinking. The topical map, the entity coverage and the editorial judgement stay human; the templating, internal-linking logic and publishing cadence are automated so the network ships faster than a manual team can sustain. That is the difference between a system and a mill.

Further reading

The academic and open-source grounding behind the methods above — the texts, the patent and the tools. Real sources, checked, not a citation wall.

  1. academic Introduction to Information Retrieval — Manning, Raghavan & Schütze (Stanford, free online) The canonical IR textbook: TF-IDF, BM25 and the vector-space model — how relevance is actually computed.
  2. academic Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks — Reimers & Gurevych (2019) Semantic embeddings compared by cosine similarity — the method behind scoring query-to-document relevance.
  3. patent Google’s Information Gain patent, explained — Search Engine Journal Google’s own patent: pages that add information beyond what already ranks score higher. Coverage and novelty, not word count.
  4. patent Ranking search results based on Information Gain scores — GoFish Digital A practitioner’s close read of the same patent.
  5. tool spaCy — industrial-strength NLP Open-source named-entity recognition: map which entities your content actually covers.
  6. tool sentence-transformers (SBERT) Open-source semantic-similarity models — score relevance yourself in a few lines.
  7. tool scikit-learn — TF-IDF & cosine similarity Open-source, lightweight coverage and relevance checks without a neural model.

The proof, and where to go next

None of this is theory here. The method runs on live retainers, and the receipts are public:

Book a teardown — £1,500 Read the notes