Topical Authority SEO: How to Measure It and Build It

Topical authority is how consistently and completely a site covers one subject, so that search systems and readers treat it as a dependable source on that subject. For web search, Google publishes no topical authority score. Google's separate "topic authority" system ranks news: it helps decide which expert sources suit a newsy query in specialized areas such as health, politics or finance (Google Search Central).

What Google does publish is a set of questions a site should pass. Its helpful content guidance asks "Does your site have a primary purpose or focus?" and whether someone researching your site would see it as "well-trusted or widely-recognized as an authority on its topic." It also names the warning sign: entering "some niche topic area without any real expertise, but instead mainly because you thought you'd get search traffic" (Google Search Central).

Internal Google Search API documentation that leaked in 2024 (mirrored in a Hex package) lists two fields that match the idea. An embedding is a numeric summary of what a page or site is about. siteFocusScore is a "number denoting how much a site is focused on one topic." siteRadius is "the measure of how far page_embeddings deviate from the site_embedding," meaning how far individual pages sit from the site's overall summary (Content Warehouse API documentation). A field's existence shows Google can compute focus. It does not show how heavily ranking uses it.

Build authority on the subject your buyers research before they buy, narrowed until you could list every question they ask about it. "Marketing" cannot be covered. "AI search visibility for B2B software" can. A topic that never leads to a purchase decision will not bring customers, however much traffic it earns.

Our rule of thumb: you should be able to list the 20 questions that matter most to your buyers in an hour. If you cannot, the topic is too broad or the buyer research is missing. Our guide to B2B AI buyer prompt research shows how to collect those questions from the places buyers ask them.

Our recommendation for where to start: answer the 10 questions that matter most to buyers, each fully on one URL, and link them to a single overview page. Add a second layer only after each of those 10 pages has one owning URL and passes the linking check below.

Two problems that a sitemap audit can detect weaken a site's focus: drift, which is publishing pages that do not serve the core topic, and good pages that nothing links to. Our inference is that these hurt focus more than low volume does; no source proves it. Drift is easy to miss because each off-topic post looks harmless on its own.

We ran the audit below on our own blog on October 10, 2026. Our sitemap (the file listing a site's URLs) held 100 blog posts. The script, using the term list shown in the next section, matched 71 and flagged 29 to review. We then read all 100 slugs (the last part of each URL). One person judged them against one rule: on-topic means about AI search visibility, tracking, citations or content. That gave 73 on-topic and 27 off-topic, mostly funding and startup-news items. Of those 27, 24 were among the 29 flagged. The other 3 matched on a word that sits in the slug but not in the subject: "brand" inside "brandon" and "visibility" in two news stories.

That is one site's slugs on one date, sorted by hand. It shows what drift looks like in numbers (more than one post in four), and it shows why a term match is only a starting list. It does not show that those posts cost rankings. By the first check below, 27% fails the one-in-five limit, so the off-topic row of the fix table applies to our blog: review each of the 27 and keep, fold or redirect it. The same sitemap also lists general-purpose utilities, such as a password checker and a UUID generator, that sit outside AI search entirely.

Run the script below on your sitemap with a list of words that belong in on-topic URLs, then read every flagged URL and judge it yourself. The script matches on the URL path only, so your domain name cannot inflate the result. It counts and lists; you decide what is off-topic, because slugs mislead.

import re, sys, urllib.request
from urllib.parse import urlparse

def get(url):
    req = urllib.request.Request(url, headers={"User-Agent": "Mozilla/5.0"})
    return urllib.request.urlopen(req, timeout=30).read().decode("utf-8", "ignore")

def urls(sitemap):
    found = []
    for loc in re.findall(r"<loc>\s*([^<\s]+)\s*</loc>", get(sitemap)):
        found += urls(loc) if loc.endswith(".xml") else [loc]
    return found

sitemap = sys.argv[1]
terms = [t.strip().lower() for t in sys.argv[2].split(",") if t.strip()]
section = sys.argv[3].lower() if len(sys.argv) > 3 else ""
pages = [u for u in urls(sitemap) if section in urlparse(u).path.lower()]
if not pages:
    sys.exit("No URLs matched that section")
on = [u for u in pages if any(t in urlparse(u).path.lower() for t in terms)]
off = [u for u in pages if u not in on]
print(f"{len(on)} of {len(pages)} URLs match your terms ({100 * len(on) / len(pages):.0f}%)")
print("Read these and judge each one:")
print(*off, sep="\n")

Save it as audit.py and run python3 audit.py https://yoursite.com/sitemap.xml "term1,term2,term3" /blog/. The third argument limits the check to one section. Our run used: visibility, citation, aeo (answer engine optimization), geo (generative engine optimization), chatgpt, perplexity, gemini, copilot, sentiment, share-of-voice, seo, schema, search, ai-answer, brand, prompt, llm. Replace those with words from your own subject. For a fuller walk through what to inspect on a site beyond slugs, see our guide on how to run an AI visibility audit.

Is the section focused?

Sort each flagged URL as on-topic, off-topic or unsure. Pass: no more than one in five pages in the section is off-topic. Fail: more than one in five. That threshold is our rule of thumb, not a Google rule.

Do the pages link together?

Open your overview page and confirm every question page is linked from it. Open each question page and confirm it links back. Pass: both directions exist for every page. Fail: any page with no inbound link from the overview page.

What is your Topic Share?

Topic Share is the share of search traffic a site earns from keywords containing a topic term. In his May 2025 post, Kevin Indig measures it by taking a topic term that has a Google Knowledge Panel, listing every keyword containing it with search volume of at least 10, and dividing each domain's estimated traffic from those keywords by the total. His ecommerce example gave Shopify 11% and BigCommerce 10% (Growth Memo). Pass: your Topic Share is at or above the top competitor's in the same month. Fail: below it. That pass line is our rule; use the result to compare, not as an absolute score.

Fix the lowest-effort problems first: add missing links and merge duplicate pages before reviewing off-topic posts, and do all of that before commissioning new content. Use the table to choose, then keep the cluster map below so every page has one job.

Cluster map template. Make one row per buyer question:

Pass: every question has exactly one owning URL, linked from the overview page and linking back. Fail: a question with two owning URLs, or an owning URL with no links in.

Ranked coverage goes with more AI citations, but the evidence is correlation, not proof of cause. In the Floyi study of 354,955 AI citations, the correlation between a site's ranked coverage (top-20 Google positions across a topic map) and its citations was 0.51 for Google AI Overviews, 0.48 for Google AI Mode, 0.36 for Gemini and 0.33 for ChatGPT. The maps were curated, so the authors say the results should not be read as web-wide averages (Floyi Topical Authority Report).

Our inference: ranking across a topic's questions looks like a prerequisite for being cited, but the data cannot show that. Before writing more, check that the bots AI engines use to read pages can reach yours (ChatGPT SEO walks through it) and see which pages AI engines already cite for your buyer questions. Our page on getting AI search citations covers what to change once you know.

That second step is what we built our Content Engine around. It finds the pages AI engines cite instead of yours, writes a better one from six live sources (Reddit, G2, competitors, news, search data and AI answers), publishes it on your own domain, and tracks the citations it earns. As of October 2026, Launch ($299 per month) tracks ChatGPT, Google AI Mode and Perplexity and includes 25 content pieces a month. Growth ($699) adds Gemini and Grok, with 50 pieces. Enterprise ($1,499) adds Claude and Copilot, with 100 pieces. Agency pricing is custom (pricing). To see it on your own buyer questions, Book a demo.